Smart Lighting Systems: State-of-the-Art in the Adoption of the EdgeML Computing Paradigm

Abstract

Lighting Systems (LSs) play a fundamental role in almost every aspect of human activities. Since the advent of lights, both academia and industry have been engaged in raising the quality of the service offered by these systems. The advent of Light Emitting Diode (LED) lighting represented a giant step forward for such systems in terms of light quality and energy saving. To further raise the quality of the services offered by LSs, increase the range of services they offer, while at the same time consolidating their reliability and security, we see the need to explore the contribution that can be derived from the use of the Artificial Intelligence of Things (AIoT) emerging technology. This paper systematically reviews and compares the state-of-the-art with regard to the impact of the AIoT in the smart LS domain. The study reveals that the field is relatively new, in fact the first works date back to 2019. In addition to that, the review delves into recent research works focusing on the usage of Machine Learning (ML) algorithms in an edge Cloud-based computing architecture. Our findings reveal that this topic is almost unexplored. Finally, the survey sheds light on future research opportunities that can overcome the current gaps, with the final aim of guiding scholars and practitioners in advancing the field of smart LSs. The study is reported in full detail, so it can be replicated.

Full text

Academic Editor: Filipe Portela
Received: 16 January 2025
Revised: 5 February 2025
Accepted: 10 February 2025
Published: 14 February 2025
Citation: Paolone, G.; Paesani, R.;
Pilotti, F.; Camplone, J.; Piazza, A.;
Di Felice, P. Smart Lighting Systems:
State-of-the-Art in the Adoption of the
EdgeML Computing Paradigm. Future
Internet 2025,17, 90. https://doi.org/
10.3390/fi17020090
Copyright: © 2025 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license
(https://creativecommons.org/
licenses/by/4.0/).
Review
Smart Lighting Systems: State-of-the-Art in the Adoption of the
EdgeML Computing Paradigm
Gaetanino Paolone 1,† , Romolo Paesani 2,† , Francesco Pilotti 2,† , Jacopo Camplone 1,† ,
Andrea Piazza 1,† and Paolino Di Felice 3,*,†
1B2B S.r.l., 64100 Teramo, Italy; g.paolone@b2binformatica.it (G.P.); j.camplone@b2binformatica.it (J.C.);
a.piazza@b2binformatica.it (A.P.)
2Gruppo SI S.c.a.r.l., 64100 Teramo, Italy; a.piazza@softwareindustriale.it (R.P.);
f.pilotti@softwareindustriale.it (F.P.)
3Department of Industrial and Information Engineering and Economics, University of L’Aquila,
67100 L’Aquila, Italy
*Correspondence: paolino.difelice@univaq.it
†These authors contributed equally to this work.
Abstract: Lighting Systems (LSs) play a fundamental role in almost every aspect of human
activities. Since the advent of lights, both academia and industry have been engaged in
raising the quality of the service offered by these systems. The advent of Light Emitting
Diode (LED) lighting represented a giant step forward for such systems in terms of light
quality and energy saving. To further raise the quality of the services offered by LSs, increase
the range of services they offer, while at the same time consolidating their reliability and
security, we see the need to explore the contribution that can be derived from the use of
the Artificial Intelligence of Things (AIoT) emerging technology. This paper systematically
reviews and compares the state-of-the-art with regard to the impact of the AIoT in the
smart LS domain. The study reveals that the field is relatively new, in fact the first works
date back to 2019. In addition to that, the review delves into recent research works focusing
on the usage of Machine Learning (ML) algorithms in an edge Cloud-based computing
architecture. Our findings reveal that this topic is almost unexplored. Finally, the survey
sheds light on future research opportunities that can overcome the current gaps, with the
final aim of guiding scholars and practitioners in advancing the field of smart LSs. The
study is reported in full detail, so it can be replicated.
Keywords: Internet of Things; artificial intelligence; Machine Learning (ML); smart lighting
systems; Edge Computing; edgeML; systematic literature review
1. Introduction
Lighting Systems (LSs) are one of the most revolutionary inventions ever; they have
sped up human progress, improving, at the same time, quality of life [
1
]. LSs are ev-
erywhere: at home, in public buildings, inside cities, along streets and highways. The
side effect from the massive, and often incorrect, use of LSs is quantified by the huge
energy consumption associated with them. In 2022, 25% of EU total energy consumption in
households was for electricity (source: https://ec.europa.eu/eurostat/statistics-explained/
index.php?title=Energy_consumption_in_households (accessed on 10 November 2024)).
Pasolini et al. [
2
] report that street lighting has a tremendous impact on energy consumption.
In the EU, for instance, street lighting alone accounts about 40% of municipal electricity
bills. Today, both academia and industry are engaged in a converging effort to turn LSs into
Future Internet 2025,17, 90 https://doi.org/10.3390/fi17020090

Future Internet 2025,17, 90 2 of 50
sustainable systems, reducing their energy consumption, increasing at the same time their
security, and improving the overall quality of service by preventing outages, preserving
citizens’ personal data, and also keeping their daily management as easy as possible.
The present Systematic Literature Review (SLR) aims at investigating the state-of-the-
art in the adoption of Internet of Things (IoT), Machine Learning (ML) methods, and the
Edge Computing paradigm in reaching a target that cannot be postponed any further. The
research described in this paper is not focused on a specific perspective, in contrast with
previous reviews. Soheilian et al. [
3
], for example, investigated the effects of smart LSs
on energy saving and people’s well-being in residential buildings by reviewing studies
published in the range January 2001 to February 2021. Ten studies, out of the 13 included in
the review, focus on energy outcomes, either alone or combined with people’s well-being.
In ref. [
4
], it has been pointed out that despite several studies having surveyed the smart
lighting literature, up to the end of 2021 none of them had investigated the adoption of ML
algorithms in such an application domain. Moving from such an observation, Putrada et al.
carried out an in depth SLR looking for ML methods suitable for increasing people’s
comfort, a socially relevant topic. The SLR covers the period 2014–2021.
The present study has the following merits:
•
It reports on an SLR realized with following the well-known method proposed by
Kitchenham and Charters in 2007 [
5
]. This method allows one to select, study, and sum-
marize the state-of-the-art with respect to a given set of research questions in a way
that is unbiased and (to a large extent) repeatable. It has been remarked that often
SLRs are conducted without following a well-defined methodology [6].
•
It promotes knowledge transfer from academia to firms. Knowledge transfer is
a motivating factor in the economic growth of Small and Medium-sized Enter-
prises (SMEs) all over the world. In the case of the EU, they represent 99%
of all businesses (source: https://single-market-economy.ec.europa.eu/smes/sme-
fundamentals/sme-definition_en (accessed on 10 November 2024)). Knowledge trans-
fer between academia and industry has received a lot of attention by scholars [
7
,
8
].
Ref. [
8
] reports on a review that summarizes the ways in which the transfer of knowl-
edge between academia and industry takes place. From such a study, it comes out
that knowledge is transferred mostly by means of published research findings; that is,
researchers write papers that, once published, become knowledge accessible to people.
Unfortunately, as the literature shows, this method used by academia is not entirely
working. The present SLR was carried out through strict cooperation between the
University of L’Aquila and an Italian SME called B2B S.r.l. which represents another
channel of knowledge transfer [7,8].
•
It follows multiple recommendations suggested by Santos et al. in their study (avail-
able online on 28 August 2024) that applies the notion of sustainability to the SLR
domain. Briefly, ref. [
9
] defines what sustainable SLRs are in terms of 15 major charac-
teristics, 15 critical factors that should be taken into consideration in the SLR process,
and 16 guidelines for carrying them out. For example, the research questions of
our SLR answer the needs of the stakeholders through evidence from the scientific
literature (this aspect intercepts the C11 characteristic in [
9
]). In the context of this
study, the stakeholders are B2B’s practitioners that took part in the SLR process and
directly will benefit from the outcomes. In addition, our SLR overcomes critical
factors CF2 and CF4 in [
9
], as will be clarified in the following section. In conducting
the SLR, all the guidelines (G1-G6) were followed. They pertain the communication
and collaboration among people taking part in the review process.
•
It provides an up-to-date state-of-the-art overview concerning the role of EdgeML in
the development of future smart LSs. In detail, the study (a) presents a map concerning

Future Internet 2025,17, 90 3 of 50
high-quality publications on the topic; (b) adopts a taxonomy of the topics that have
been investigated so far; (c) informs readers about the ML methods mostly adopted;
(d) reports on the degree of adoption of the EdgeML computing paradigm; and lastly,
(e) discusses several lines of future research opportunities. By exploring the potential
of the integration of IoT and Artificial Intelligence (briefly, AIoT) into LSs, this review
provides valuable insights into the next generation of smart LSs.
The remaining part of this work is structured as follows. Section 2introduces the
topics underlying the present review. Section 3presents the research method adopted in
the study. The section is quite long, so it comprises two sub-sections talking about planning
activities (Section 3.1) and conducting activities (Section 3.2). The former sub-section
presents the review needs and the research questions, while the latter sub-section details
the process followed to isolate the studies to be analyzed in detail. Inclusion/exclusion
criteria and quality criteria are adopted to reach this goal. Section 4replies to the given
research questions, while Section 5throws some light on current gaps which represent
future research opportunities both for academia and industry. Section 6focuses on two
related works, while Section 7analyzes potential threats to the validity of the findings of
the present study. Section 8ends the paper. The acronyms used within this paper are listed
at the end of this document.
Figure 1shows the organization of the remaining sections of this paper. Hereafter,
the arguments are presented in the same order.
Figure 1. Organization of this SLR. The numbers refer to the sections.
2. Background
2.1. IoT + AI = AIoT
IoT is a network of physical devices, actuators, interfaces, and connectivity. The basic
components of the network are briefly described in the following:
•
Sensors gather real-time data from the environment. They convert a physical phe-
nomenon into a digital signal that a machine can elaborate on. The term IoT device is

Future Internet 2025,17, 90 4 of 50
commonly used to denote a real-life device (e.g., a smart Light Emitting Diode (LED)
bulb) which embeds a certain number of sensors.
•
Microcontroller units (MCU) process the data collected by the sensors. An MCU is an
integrated circuit composed of memory, a processor, and input/output units.
•
Communication modules transmit (either in wireless or wired mode) the data over the
network. The following are some wireless communication protocols that are widely
used: Bluetooth, BLE (Bluetooth Low Energy), ZigBee, 6LoWPAN, WiFi, 802.11ax,
radio-frequency identification (RFID), NFC (Near-Field Communication), LoRa (Long
Range) (a Low-Power Wide-Area Network (LPWAN) technology), SigFox (a LPWAN
solution), LTE (Long-Term Evolution, also called “4G LTE”), and NarrowBand IoT
(NB-IoT). In ref. [
10
], they are described in some detail. A new entry in the pro-
tocol family is matter: an industry-unifying standard based on IP that facilitates
communication across smart home appliances (e.g., lights, TV, and washing machines)
and an application framework for home automation (www.buildwithmatter.com and
https://csa-iot.org/all-solutions/matter/ (accessed on 20 November 2024)). Matter is
the output of a Connectivity Standards Alliance project started in 2019 together with
relevant international firms (e.g., Amazon, Apple, Google, and Zigbee Alliances).
•Cloud includes storage and servers needed for the processing of the sensed data.
IoT devices sense huge amounts of data that need to be complemented with context.
AIoT is the context [
11
,
12
]. AI is beneficial for both real-time processing and post-event
processing. The advantages that come from the adoption of AIoT technology have been re-
ported for all IoT application domains, such as industry [
13
], retail [
14
], smart
cities [15,16],
and healthcare [17,18].
2.2. Cloud Computing
Cloud computing is the well-known term used to refer to the delivery of hosted
computing services (including servers, storage, databases, networking, software, analytics,
and intelligence) over the Internet with pay-as-you-use pricing. This paradigm frees users
from the need to purchase, operate, and maintain on-premises physical data centers and
servers. In connection with the IoT, the Cloud computing paradigm conceptually can be
implemented as a four-layer architecture composed of [19]:
• The Perception layer: this consists of sensors and actuators;
•
The
Network/Communication
layer: this ensures connectivity among the IoT devices
part of the network;
•
The
Cloud
layer (also called Service layer): this provides storage, computational power,
and software tools services vital in the implementation of IoT applications;
•
The
Application
layer: this consists of mobile apps and stand-alone applications
offered to users to control the IoT devices.
2.3. Edge Computing
The Cloud has been the deployment model of IoT systems for many years. However,
connecting IoT devices directly to the Cloud poses serious latency constraints to preserve
performance [
10
]. To mitigate the issue, a further computing layer has been added called
edge, moving from the Cloud-only computing paradigm to the Edge Cloud continuum
paradigm [
20
–
22
]. It is worth noticing that Edge Computing (EC) is a broad definition
that embraces several computing paradigms; in other words, this name is appropriate at
a high level of abstraction. References [
21
,
23
] discuss the available computing paradigm
variants from several perspectives. Singh and Gill [
21
] remark that in the years 2009–2014
four variants were proposed to implement the Edge Computing concept, namely: Edge
Computing cloudlets, fog computing, mobile Edge Computing and micro data centres.

Future Internet 2025,17, 90 5 of 50
In 2017, mobile Edge Computing was renamed multi access Edge Computing. Given fog
computing is a way to implement EC, in the following only two alternative paradigms are
distinguished: cloud-based computing and Edge Cloud-based computing.
2.4. Machine Learning
AI is an umbrella that brings together many methods. In this SLR, the focus is on
Machine Learning (ML) methods. The latter comprise Supervised Learning, Deep Learning
(DL), Ensemble Learning (EL), and Reinforcement Learning (RL). Ref. [
24
] is an interesting
source to start learning ML models.
Solutions based on fuzzy logic, agents, and natural language processing are labelled
as “out of scope” in the present SLR.
2.5. AIoT Implementation
In light of previous considerations, the present SLR looks at studies where the imple-
mentation of AIoT systems leverages ML methods. The latter comprise the well-known
training and inference two-stage process. The first stage uses data from IoT devices to train
the model, while the second stage uses the built ML model to make inferences from the
actual input data. Figure 2shows the two alternatives to implement the AIoT.
Figure 2a
depicts the Cloud-based path, where both model building and model inferencing are per-
formed by the server on the Cloud. On the other side of the IoT network, IoT devices
upload data and receive decisions from the Cloud. Figure 2b depicts the Edge Cloud
computing scenario where IoT devices are connected to the edge servers close to them.
Model building is still performed in the Cloud; then, the model is delivered to the edge
where inferencing takes place in near real time. Again, the decisions are returned to the
IoT devices. It is worth noticing that initially the edge still must send to the Cloud a
meaningful quantity of data sensed by the IoT devices for building the model, but while
this step usually takes place once, the model is used several times to make inferences. An
Edge Cloud collaborative computing platform (called Sophon Edge) for building AIoT
applications is described in [
20
]. Specifically, Rong et al. present, in sequence, the platform
computing model, its architecture, and how the evolution of the ML model is supported.
Figure 2. Two alternative implementations of AIoT.
2.6. EdgeML
As mentioned on introducing the Edge Computing paradigm (Section 2.3), running
ML models in the Cloud has severe counter indications, but, at the same time, running
them on edge-side electronic devices is challenging due to the lack of sufficient computing
resources. In the previous section, it was remarked that distributing the processing of ML
models between the edge and the Cloud is the right strategy to overcome these issues.

Future Internet 2025,17, 90 6 of 50
Today we observe a twofold convergent effort to improve the current state-of-the-art. On
the industry side, there is a massive expansion of the market of hardware devices and
complementary embedded software that can process ML algorithms, while academia is
investigating new methods for pushing the processing of IoT data as close as possible to
the IoT devices that the data come from.
Two proposals by the industry are the following. Microsoft is developing EdgeML
(https://www.microsoft.com/en-us/research/project/edgeml/ (accessed on 15 November
2024)), a suite of ML models that are trained on the Cloud and can run on resource-
constrained edge IoT devices. The code of ML algorithms for edge devices developed
at Microsoft Research India is available at the following repository https://github.com/
Microsoft/EdgeML. Google has released the LiteRT (short for Lite Runtime, formerly
called TensorFlow Lite) platform. LiteRT features tools for converting TensorFlow Neural
Network models into a simplified version that then can be run on edge-constrained devices.
An interesting project from academia is the development of the Edge Learning Ma-
chine (ELM) framework. The latter aims at supporting developers in designing and
deploying ML solutions on edge-constrained devices. In ELM, the ML model is created,
trained, and optimized on a desktop computer, while inferences are run on MCUs. The
framework implements three supervised ML algorithms (SVM, kNN, and DT); more-
over, it supports ANNs leveraging the X-Cube-AI package for STM 32 devices [
25
]. The
framework is open source and distributes a platform-independent C language implemen-
tation of those algorithms (https://github.com/Edge-Learning-Machine (accessed on 15
November 2024)). Another project from academia is the EdgeML framework described
in [
26
]. It controls the execution of DNN models by combining a workload-offloading
mechanism and dynamic neural architecture. To achieve good latency–accuracy–energy
performance on edge-constrained IoT devices, EdgeML adopts the RL model. Authors
implemented EdgeML for several well-known DNN models on the latest edge devices.
EdgeML is developed on top of TensorFlow. The source code of EdgeML is available at
https://github.com/Kyrie-Zhao/EdgeML.git (accessed on 15 November 2024).
In previous studies, the term EdgeAI (edge intelligence in [
27
]) has been largely used
to refer to the scenario we are talking about, that is to signify that ML models are run
either close to where the data are collected (for example, by leveraging MCUs connected to
the IoT sensors), or in dedicated hardware (also called edge server or micro-data center)
located near the IoT sensors. Ref. [
28
] is a recent work on EdgeAI. It reviews the hardware
suitable for implementing EdgeAI, the supporting APIs, and the applications that can
benefit from it. McEnroe et al. [
29
] review the convergence of Edge Computing and AI for
unmanned aerial vehicles, while Tu et al. [
30
] adopt the EdgeAI paradigm to implement an
EdgeAI-based vehicle-tracking system.
The architecture in Figure 3is suggested by recent studies on EdgeML, e.g., [
10
], to
implement the EdgeML paradigm. The robot head icon symbolizes that intelligence is
present on both sides, which enables a joint computation: part in the edge server and part
in the IoT device.
2.7. TinyML
TinyML is strictly connected to EdgeML. The TinyML community aims at promoting
the development of algorithms, software, and hardware suitable to run ML models on low-
cost and resource-constrained IoT devices. Prospectively, by leveraging TinyML the analysis
and interpretation of data will become viable on IoT devices and, if needed, replying in real
time will also become viable. Both academia and industry agree that TinyML will play a
fundamental role in the future in providing intelligent IoT solutions, with the least possible
access to the Cloud. TinyML, as a service project ongoing at Ericsson Research, is just an

Future Internet 2025,17, 90 7 of 50
example of this believe (https://www.ericsson.com/en/blog/2019/12/tinyml-as-a-service
(accessed on 16 November 2024)).
Figure 3. EdgeML supporting architecture.
Figure 4shows the TinyML workflow. It starts from data collection by sensors located
in the environment; then, the ML model is built in the Cloud or in a data center, compressed
to be deployed in MCUs where, eventually, the inference step will take place. The steps
in the figure are detailed in [
31
], where the available open source software frameworks
that support the advancement in ML research in MCUs are also listed. An introduction to
TinyML, followed by a discussion of hardware and software tools supporting it, and state-
of-art applications of TinyML may be found in ref. [
32
]. Oliveira et al. [
33
] provide a holistic
perspective on the challenging and rapidly evolving TinyML research field. Ref. [
34
] is
another useful study to obtain an overview on how TinyML-based systems are built. It also
focuses on the techniques suitable for reducing the complexity of the ML/DL algorithms
so that they can be run on MCUs.
Figure 4. TinyML workflow.
2.8. Smart Lighting Systems
The three principal benefits offered by the advent of high-power LEDs in the early
2000s regard the following: higher energy efficiency, longer lifespan (and then reduction
of maintenance costs), and basic dimming capabilities. The specific consumption per
dwelling for lighting is decreasing thanks to the phasing out of incandescent light bulbs.
In the EU, in 2021 they accounted for 14% of captive electricity, compared to 20% in 2000
(source: https://www.odyssee-mure.eu/publications/efficiency-by-sector/households/
household-eu.pdf (accessed on 10 November 2024)).

Future Internet 2025,17, 90 8 of 50
Building on the LED technology, smart LSs are emerging. They have the potential
to become the backbone of infrastructure for homes, buildings, cities, and rural areas,
featuring benefits that go beyond energy efficiency up to user comfort. For citizens, lighting
professionals, and urban/rural planners, understanding and leveraging this technology
is no longer optional: it is essential. The global smart lighting market is expected to
grow at an annual growth rate of 22.9% between 2022 and 2030 (source: https://energy-
cities.eu/the-evolution-of-public-lighting-from-torches-to-smart-services/ (accessed on
10 November 2024)).
In ref. [
3
], it has been pointed out that in the extant literature there are six alternative
synonyms for the phrase “smart LS”. Each terminology item aims at emphasizing a specific
characteristic of this category of system. When the notion of smart lighting was firstly
introduced in 2011 by Bhardwaj et al. [
35
], the authors meant just a system composed of
sensors and actuators connected through a network that cooperate to meet specific user
needs. In the present review, the sentence “smart LS” brings also embedded intelligence
from the adoption of ML methods. This definition corresponds to the meaning of the
phrase “intelligent smart LSs” adopted by Kim and Park [36].
Smart LSs comprise hardware and software. In turn, hardware includes LED lu-
minaires, sensors, and the communication network, while software consists of either a
manufacture’s app (this is frequent in homes’ smart LSs) or a central management system
(as in the case of smart street LSs). In addition, a hardware–software component is present
usually called a control system [
2
]. These elements are briefly described in the following.
Each LED luminary includes a certain number of LED bulbs (also called lamps) and a
built-in driver (called a light controller in [
2
]). The latter regulates the power to the bulbs,
ensuring proper current and voltage for optimal performance and longevity. Advanced
LED drivers include dimming capabilities and can communicate with the control system.
Smart luminaires embed different types of sensors. Table 1compares six commonly used
ones. Control systems manage the lighting based on sensor data and pre-programmed
parameters. Communication networks are wireless or wired networks that allow the
lights to communicate with each other and with either an app or the central management
system. The central management system allows operators to monitor and control the
lighting network.
Table 1. Comparison of sensors embedded in commercially available luminaires.
Sensor Type Detection Method Advantages Disadvantages Applications
Photosensitive Light Level
Automatic on/off
based on ambient light,
energy saving
Limited functionality,
affected by sudden
light changes
Outdoor lighting
Passive Infrared (PIR) Motion (body heat) Energy saving, motion
detection for security
Can be triggered by
pets or heat sources,
limited field of view
Indoor lighting,
hallways, entrances
Ultrasonic Motion
Wide detection range,
good for small object
movement
Affected by air
movement, requires
precise installation
Warehouses,
large spaces
Microwave Motion Long range, penetrates
walls (limited)
Expensive, potential
health concerns
Security lighting,
large spaces
Temperature Temperature
Over-temperature
protection for LED
lamps
No motion detection or
direct lighting control LED luminaires
Voice Sound Hands-free lighting
control
Requires specific voice
commands, privacy
concerns
Specific applications

Future Internet 2025,17, 90 9 of 50
The Philips Hue line of bulbs was the first smart bulb on the market. Now, several
alternative products are available. Each product consists of wireless RGB LED bulbs and
supports connectivity. The latter may change with the product. For instance, the Philips Hue
smart LED bulb released before 2019 communicates through the Zigbee Light link protocol,
a compatible subset of Zigbee 3.0, while Hue released later supports both Zigbee 3.0 and
Bluetooth. Adding the Hue Bridge component to the LS based on Hue smart LED bulbs and
using the official Philips Hue Bluetooth app on either a phone or tab opens the door to a
long list of smart functions (https://www.currys.co.uk/products/philips-hue-white-and-
colour-ambiance-smart-led-bulb-b22-800-lumens-triple-pack-10246932.html (accessed on
10 November 2024)).
LED and smart street LSs reduce energy usage on two levels: first from more efficient
LED luminaires and second from a dimming capability with smart controllers. A bench-
mark study carried out by Northeast Group in partnership with CityLab Insights on 16 US
cities, with population ranging from 8000 to over 4 million states that the cities using LED
and smart street LSs see a 60% to 80% reduction in energy usage [37].
The transition to smart LSs is still ongoing. Table 2lists recent research projects on the
topic, addressing three well-known application domains. All these projects do not adopt
ML methods, but the adoption of ML in the smart LS domain is a research opportunity
to further improve the user experience with this category of service. Processing the data
collected by indoor/outdoor luminaires is a precondition for the following:
•
Understandinging lighting-usage patterns by people in order to improve energy
efficiency, as noted in [38];
• Carrying out predictive maintenance of LED lights, as emphasized in [2];
• Obtaining fault notifications in real time in order to limit illuminance outages [2].
Table 2. Recent publications on smart LSs.
Application Domain Use Case Studies
Smart city Smart street LS [39–44]
Smart building [39,45–47]
Smart education Smart lab [48]
Smart agriculture Monitoring of lettuce growth [49,50]
Previous research has clarified the link between ML and Human Activity Recognition
(HAR), Computer Vision (CV), Voice Recognition (VR), and Speech Recognition (SpR), four
fields of AI that have a growing importance in modern smart LSs [
51
–
53
]. For example,
by using digital images from cameras and DL models, it is possible to train computers
to classify objects and then be able to react to what they “see”. In practical terms, image
recognition allows one to modify the light intensity in daily situations because of the
human presence in a room or a pedestrian walking along a city street [52]. Laad et al. [53]
discuss the transformative influence of CV methods on well-established ML methods with
respect to building versatile ML models that leverage the strengths of both disciplines.
Ref. [
54
] discusses an AIoT face-recognition-based LS which potentially can create light
conditions which adhere to intensity values initially set by home inhabitants. A review
of the state-of-the-art on both traditional and DL-based methods of speaker recognition
may be found in [
55
,
56
]. Sarbast and Mohsin [
57
] investigated the effectiveness of Random
Forest (RF), K-Nearest Neighbours (KNN), and Support Vector Machine (SVM) classifiers
in solving speech-recognition problems. In the smart LS context, SR is frequently used for
turning on/off lights in a room to enhance user experience. Putrada et al. [
4
] pointed out
that ML together with speech recognition, voice recognition, and face recognition look
very promising for enhancing people’s comfort.

Future Internet 2025,17, 90 10 of 50
Smart LSs are a domain of primary importance in the context of the creation of smart
cities. The EU, for example, is investing a lot of money to support projects consistent
with this objective (https://commission.europa.eu/eu-regional-and-urban-development/
topics/cities-and-urban-development/city-initiatives/smart-cities_en (accessed on 10
November 2024)). Manufacturers of smart LS products claim that this technology brings
substantial cost savings through energy efficiency and improved maintenance, enhances
public safety, and contributes to environmental sustainability goals. The present SLR aims at
assessing whether the state-of-the-art of the research confirms all these wonderful promises.
2.9. Smart LSs and EdgeML
Within this paper, the EdgeML term does not refer either to the EdgeML library under
development at Microsoft, nor to the EML framework mentioned in Section 2.6. Hereinafter,
EdgeML simply denotes that ML models are thought to be run as close as possible to the
edge-constrained IoT devices of an AIoT architecture that implements a smart LS. The
adoption of the EdgeML computing paradigm is relevant in the smart LS domain, as briefly
explained in the following. Both indoor and outdoor smart LSs can enhance considerably
the well-being and safety of citizens of advanced smart cities. But there are at least three
stringent requirements these systems must meet: cut down the latency, limit the network
overhead, and protect personal data:
•
The latency constraint is particularly stringent for smart street LSs where light con-
trollers are responsible for dynamically adapting the light intensity to traffic conditions
to ensure the safety of drivers.
•
The number of luminaires worldwide is becoming huge (about 360 million street lights
are foreseen on the globe by 2029 (source: https://smart-cities-marketplace.ec.europa.
eu/sites/default/files/2021-06/Smart%20Lighting%20Factsheet_0.pdf (accessed on
10 November 2024)), consequently the amount of the data collected by them is huge
as well. Processing the data collected by the luminaires at the edge of the IoT network
is a best practice to limit network overhead.
•
It has been remarked that because future indoor and outdoor LSs become smarter, they
need to collect more data about people and their daily activities [
58
]. This situation
poses a twofold ethical concern. Firstly, the collected data must be securely archived
to be protected against violation of privacy; secondly, the use of personal data must
comply with the personal-data protection laws. Limiting the transit of personal data
on the network is a precondition to protect it.
3. Research Method
The overall workflow of our study is composed of informal research followed by a well-
defined research methodology. By “informal research”, we mean unstructured research that
originated as our curiosity to conduct a preliminary search, with the understanding that
if it had produced an encouraging result, then we would have spent more time running
the second stage mentioned above. The informal research was performed by running the
following string against the Scopus database:
(internet AND things) AND
((lighting AND system*) OR (smart AND light*)) AND
(tertiary AND stud*)
As the output was the empty set, we learnt that so far no tertiary studies connecting
(smart) LSs to IoT technology have been published. So, we decided to carry out an SLR,
starting from primary and secondary studies (if any), to provide an overview of the selected
research area to find out if scientific studies already exist on the selected topic and quantify
the research evidence [5] (p. 44).

Future Internet 2025,17, 90 11 of 50
The SLR study was articulated in three phases, as suggested in [
5
]: planning the review,
conducting the review, and reporting the review results. The reporting phase is self-evident,
so below we focus on the other two phases. There are three basic planning activities:
• Identification of the review need;
• Specification of the research questions;
•
Elaboration of the review protocol (i.e., explanation of the method that will be used to
conduct the SLR).
3.1. Planning Activities
3.1.1. The Review Need
As previously mentioned, the present SLR is the result of a collaboration between B2B
S.r.l (an Italian SME) and the Department of Industrial and Information Engineering and
Economics of the University of L’Aquila (Italy). At the beginning of summer 2024, B2B S.r.l
expressed the need to write an unbiased report about the adoption of AIoT technology in
the development of smart LSs. The interest in the topic arose within IT projects currently
being carried out at B2B S.r.l. One of the authors of this article (Di Felice) is an academic,
while most of the remaining co-authors work at B2B S.r.l. It has been remarked that the
team’s experience in conducting SLR is a critical factor (CF4 in [
9
]) with a direct impact
on the quality of the results. That is why the review process was coordinated by the
academic person. B2B S.r.l. declared interest in acquiring up-to-date knowledge about the
state-of-the-art in the domain of delivering smart LSs leveraging the EdgeML.
3.1.2. Research Questions
The aim of the SLR was to answer the general research question mentioned above.
The latter, in turn, splits into the following Research Questions (RQs):
(RQ1) What is the map of published primary and secondary studies about AIoT-based LSs?
(RQ2) What are the main topics addressed for AIoT-based LSs?
(RQ3) What are the key ML methods enabling the implementation of AIoT-based LSs?
(RQ4) Do the selected studies through RQ3 implement the EdgeML computing paradigm?
While RQ1, RQ2, and RQ3 isolate the extant literature about smart LSs leveraging
the AIoT technology, RQ4 restricts the subset of the selected pool of studies to those that
actually leverage EdgeML. Studying this restricted set of papers (if any) is the final aim
of the present SLR. All the previous RQs translate specific stakeholders’ needs. The final
point was reached through constant communication with and the competent participation
of B2B S.r.l professionals in the planning stage of the SLR.
3.2. Conducting Activities
Standard database search and snowballing are the dominant approaches to search
for studies to be included in an SLR [
59
]. The review protocol adopted in the present SLR
comprises both approaches to cut down the risk of losing relevant studies. The left side of
Figure 5summarizes the four stages necessary to select the first set of studies that were
included in the review. Initially, the Scopus database was queried by entering the search
string, then the returned documents were filtered against the inclusion/exclusion criteria
followed by the quality assessment stage. The title of the (21) selected studies was the input
of the right side of the workflow in Figure 5that implements the Forward Snowballing
strategy [
60
]. The forward search was accomplished by making recourse to the Scopus
engine which for a given paper returns the list of later papers citing it (if any). (Google
Scholar is a relevant alternative tool to accomplish the forward snowballing.) The retrieved
studies were filtered against a distinct set of inclusion/exclusion criteria followed by the
quality-assessment stage. The reason why the inclusion and exclusion criteria adopted

Future Internet 2025,17, 90 12 of 50
for the snowballing activity are distinct from those used in the stages depicted in the left
side of Figure 5will be explained later. Seven further papers were selected at the end
of the filtering stages. So, 21 + 7 studies were included in the present SLR. Hereinafter,
the activities of the conducting phase are detailed.
Figure 5. The implemented SLR protocol.
3.2.1. Selection of the First Set of Studies
Search String
The search string was the following:
(“Internet of Things” OR IoT) AND
(“Artificial Intelligence” OR “Machine Learning” OR Learning) AND
(“Lighting systems” OR “Smart Lighting”)
Search Process
The search was realized (by one of the authors, actually a B2B practitioner) as a hand-
operated search of Scopus articles that mention the keywords in the search string either
in the title, in the abstract, or among the authors’ keywords. The entered search string
was enhanced by adding a filter about the language (“English”) and four more filters for
limiting the retrieval to the following document types: article, conference paper, review,
and book chapter. The complete syntax of the Scopus search string looks as follows:
TITLE-ABS-KEY ((“Internet of Things” OR iot) AND
(“Artificial Intelligence” OR “Machine Learning” OR learning) AND
(“Lighting systems” OR “Smart Lighting”)) AND
(LIMIT-TO (DOCTYPE, “cp”) OR LIMIT-TO (DOCTYPE, “ar”) OR
LIMIT-TO (DOCTYPE, “ch”) OR LIMIT-TO (DOCTYPE, “re”))
The search was carried out on 2 November 2024. A list of 89 items was retrieved.
Tables 3and 4show, respectively, the distribution over the years of the retrieved documents
and the number of publications aggregated by type. For each retrieved study, we asked

Future Internet 2025,17, 90 13 of 50
the Scopus engine to return the following metadata: authors’ names, title, abstract, author
keywords, publication venue, DOI, and number of citations.
Table 3. Distribution of papers over the years.
Year Documents Year Documents
2024 13 2018 4
2023 17 2017 2
2022 18 2016 0
2021 13 2015 1
2020 14 2014 1
2019 6
Table 4. Aggregation of papers by types.
Document Type Documents Document Type Documents
Conference Paper 54 (60.7%) Book Chapter 8 (9.0%)
Article 23 (25.8%) Review 4 (4.5%)
Once the likely pertinent studies have been collected, it is necessary to assess their
actual relevance. According to [
5
], the assessment is accomplished in two stages. Firstly,
inclusion/exclusion selection criteria are used to determine which studies are included
in, or excluded from, the SLR; then the actual quality of the remaining studies must
be assessed. By borrowing the hint from [
61
], it can be said that the application of the
inclusion/exclusion criteria represents the primary selection phase, while the application
of the quality assessment criteria implements the secondary selection phase.
Inclusion Criteria
All the documents returned by the Scopus engine (either primary or secondary English
study) were included since they contain the keywords in the search string. It is worth
noticing that all the papers indexed by Scopus are subject to peer review. Moreover,
since the search involved only items in the Scopus database, duplicate publications were
not possible.
Exclusion Criteria
The selection of articles returned by Scopus was narrowed by ignoring documents
belonging to at least one of the eight categories listed in Table 5.
Table 5. The reasons for exclusion of a study retrieved by the Scopus engine.
Exclusion Criteria Reason for Exclusion
EC1. In the retrieved paper, the keywords “Artificial Intelligence” and/or “Internet of Things”
are part of the name of the conference where the work was presented, but the manuscript
doesn’t touch the AIoT topic.
False positive
EC2. In the retrieved paper, the keywords “Artificial Intelligence” and/or “Internet of Things”
are listed as authors’ keywords, but the manuscript does not touch on the AIoT topic. False positive
EC3. In the retrieved study, the word “Learning” (part of the search string) concerns
education and not the AI. Out of scope
EC4. The study focuses on the adoption of AI in connection with LSs, but the methods
considered are not the ML ones. Out of scope

Future Internet 2025,17, 90 14 of 50
Table 5. Cont.
Exclusion Criteria Reason for Exclusion
EC5. The study mentions ML, but it does not elaborate on how ML methods can act as an
enabler for the advancement in the development of smart LSs. So, the paper is too generic
from the perspective of the stakeholders who commissioned the present SLR.
Not relevant
EC6. The study adopts an ML classifier just to automate lighting ON/OFF, for example
through face recognition. Not relevant
EC7. In the retrieved paper, the sub-string (“Lighting systems” OR “Smart Lighting”) is
mentioned as a potential application domain, but the study does not elaborate on the topic.
So, the paper is useless from the perspective of the stakeholders who commissioned the
present SLR.
Not relevant
EC8. In the full text of the study, the Edge Cloud continuum paradigm is either marginal or
not mentioned at all. So, the paper is useless from the perspective of the stakeholders who
commissioned the present SLR.
Not relevant
The selection of the studies suitable for the present review was conducted iteratively.
This is a best practice that contributes to the realization of sustainable SLRs, as they are
called in [
9
]. Firstly, two authors analyzed the abstracts and authors’ keywords of the
89 articles. They worked individually by taking into account the exclusion criteria to decide
which papers are to be entered into the full-text screening phase. The involvement of at
least two reviewers to perform SLR activities that implicate sensitive judgments is one of
the guidelines recommended in [
9
] for reducing bias. In this stage, 19 papers were removed,
obtaining a total of 70 publications (Figure 5). Table 6collects the reference to the 19 papers
excluded after reading the abstract.
Table 6. Papers excluded after reading the abstract.
Reference Aim Motivation for
Exclusion
Nieh H.-M. and Chen H.-Y., An Arduino-Based Experimental Setup for Teaching Light Color Mixing,
Light Intensity Detection, and Ambient Temperature Sensing, Physics Teacher,61 (2), 2023, 133–137, DOI:
10.1119/5.0066060
On teaching light color mixing and
light intensity detection in physics
courses.
EC1
Klimek R., Proposal of a multi-agent system for a smart outdoor lighting environment, LNCS, 10246
LNAI, 2017, 255–266, DOI: 10.1007/978-3-319-59060-8_24
Proposal of a multi-agent system for a
rural environment. EC4
Mandaric K. et al, Agent-based approach for user-centric smart environments, Smart Innovation,
Systems and Technologies, 2020, 186, 37–46, DOI: 10.1007/978-981-15-5764-4_4
Proposal of an agent-based smart
environment system. EC4
Mandaric K. et al., A Multi-Agent System for Service Provisioning in an Internet-of-Things Smart Space
Based on User Preferences, Sensors, 24 (6), 2024, DOI: 10.3390/s24061764
Adoption of a context-aware
multi-agent negotiation algorithm for
a smart lighting use case.
EC4
Zandbergen D., The Unfinished Lampposts: The (anti-) Politics of the Amsterdam Smart Lighting
Project, (2020) City and Society, 32 (1), 135–156, DOI: 10.1111/ciso.12251
Overview of a smart lighting project
concerning lamp posts located in
Amsterdam.
EC2
Vale Z. et al., An overview on smart buildings, Encyclopedia of Electrical and Electronic Power
Engineering, 2022 DOI:10.1016/B978-0-12-821204-2.00066-0
Overview of building types,
technologies, enablers, risks, cultural
aspects, and smart building
applications.
EC5
Tsoukas V. et al., A Gas Leakage Detection Device Based on the Technology of TinyML Technologies, 11
(2), 2023, DOI: 10.3390/technologies11020045
Summary of a gas-leakage-detection
system based on TinyML. EC2
Wang J., Design of the Intelligent Elderly’s Lighting Emotional Interactive Experience System based on
Internet of Things, Proc. of Inter. Conference on Artificial Intelligence and Smart Systems, ICAIS 2021,
1347–1351 DOI: 10.1109/ICAIS50930.2021.9395946
Design of an intelligent elderly
person’s lighting emotional system
based on the IoT.
EC2
Biagetti G. et al., ToLHnet: A low-complexity protocol for mixed wired and wireless low-rate control
networks. Proc. of the 6th European Embedded Design in Education and Research Conference 2014,
177–181 DOI: 10.1109/EDERC.2014.6924383
Principles of a protocol and
presentation of a case study detailing
its implementation and performance.
EC3
Dalela P.K. et al., Surveillance enabled smart light with oneM2M based IoT networks, Communications
in Computer and Information Science, 2017), 775, 296–307, DOI: 10.1007/978-981-10-6427-2_24
Conversion of an existing lighting
infrastructure to centralized,
web-based, surveillance-enabled
smart lighting.
EC4

Future Internet 2025,17, 90 15 of 50
Table 6. Cont.
Reference Aim Motivation for
Exclusion
Sharma D. et al., Design of photo-voltaic source fed efficient corridor lighting system in green buildings,
Proc. 3rd Inter. Conference on Emerging Technologies in Computer Engineering: Machine Learning and
Internet of Things, ICETCE 2020, 58–62 DOI:10.1109/ICETCE48199.2020.9091767
Summary of savings following the
adoption of lighting controls and
usage of renewable energy for the
lighting system of Indian public
buildings.
EC1
Sun Y. et al., MagicHand: Interact with IoT devices in Augmented Reality environment, 26th IEEE
Conference on Virtual Reality and 3D User Interfaces, VR 2019, 1738–1743, DOI:
10.1109/VR.2019.8798053
Proposal of an AR-based
visualization and interaction tool that
offers a touchless mode to interact
with LSs.
EC5
Wakim P. and Mershad K., Using Internet of Things in a Learning Management System for Campus
Access Control, Inter. Conference on Computer and Applications, ICCA 2018, 46–51, DOI:
10.1109/COMAPP.2018.8460302
Integration into the university
learning management system of a
security feature through an Arduino
IoT device.
EC3
Yang Y.-T. et al., ImprovingStudents’ Learning Effectiveness by an AIoT Human Centric Lighting
System, 14th IIAI Inter. Congress on Advanced Applied Informatics, IIAI-AAI 2023, 180–181, DOI:
10.1109/IIAI-AAI59060.2023.00045
Investigation of the long-term impact
of a human-centric LS on students’
learning effectiveness.
EC5
Scholtz B. et al., An Internet of Things (IoT) Model for Optimising Downtime Management: A Smart
Lighting Case Study, IFIP Advances in Information and Communication Technology, 2019, 548, 89–104,
DOI: 10.1007/978-3-030-15651-0_9
Adoption of IoT in field service
management to manage data quality
and service-delivery challenges.
EC4
Asilian A. et al., The Role of Microelectronics for Smart Cities, Smart Grids and Industry 5.0: Challenges,
Solutions, and Opportunities, 13th Smart Grid Conference, SGC 2023, DOI:
10.1109/SGC61621.2023.10459310
Smart LSs are mentioned as an
application domain of
microelectronics.
EC4
Kouah S. et al., Internet of Things-Based Multi-Agent System for the Control of Smart Street Lighting,
(2024) Electronics (Switzerland), 13 (18), DOI: 10.3390/electronics13183673
Proposal of a smart street LS based on
IoT, fuzzy logic, and multi-agents. EC4
Tse, R. et al., Deepclass: Edge based class occupancy detection aided by Deep Learning and image
cropping, (2020) Proc. of SPIE—International Society for Optical Engineering, DOI: 10.1117/12.2572948
An image-processing-based
people-counting method to control
the presence of humans in classrooms
to set lights accordingly.
EC6
Buniel G. and Dela Cerna M., I-Detect: An Internet of Things Voice-Activated Home Automation with
Smoke and Fire Detection and Mitigation System, IEEE 13th Inter. Conference on Humanoid,
Nanotechnology, Information Technology, Communication and Control, Environment,
and Management, 2021, DOI: 10.1109/HNICEM54116.2021.9731884
A voice activation of luminaires
through Amazon’s Alexa. EC6
Then, all the authors read the full text of the 70 papers to make a decision concerning
their inclusion into the SLR based on the exclusion criteria. The admittance of a study to the
SLR was reached by consensus among the authors during meetings. In this step, 43 papers
were removed, obtaining 27 publications (Figure 5). Table 7presents the references to
the 43 papers excluded after full-text reading. One more study was deleted from the list
because retracted by the publisher. It is not mentioned at all.
Table 7. Papers excluded after full-text reading.
Reference Aim Motivation for
Exclusion
Vaidya M. et al., Energy Efficient Smart Lighting System for Rooms, Studies in Big Data, 2022, 92,
107–125 DOI: 10.1007/978-3-030-77214-7
Illustration of a self-adjusting LS and
a facial recognition-based lighting
management system.
EC5
Nigel F.T. and Longe O.M., Smart energy efficient lighting system for smart buildings, 2021 IEEE
PES/IAS PowerAfrica, PowerAfrica DOI: 10.1109/PowerAfrica52236.2021.9543273
Description of an LS that regularly
updates the count of occupants in a
room, detects any motion,
and measures light intensity in the
room.
EC4
Panicker, J.G. and Azman, M., Robust and Lightweight ControlSystem for IoT Networks: Enabling IoT
for the Developing World, Advances in Intelligent Systems and Computing, 2021, 1245, 73–92, DOI:
10.1007/978-981-15-7234-0_8
Proposal of a secure speech-based
automation and control system for
residential and commercial buildings.
EC4
Shanmugasundaram N. et al., Smart Lighting System Using the Internet of Things, 8th Inter. Conference
on Advanced Computing and Communication Systems, ICACCS 2022, 2037–2040, DOI:
10.1109/ICACCS54159.2022.9785327
Remote monitoring of home
appliances with an Android
application that uses Wi-Fi
technology.
EC4
Chen B. et al, Three-dimensional ultraflexible triboelectric nanogenerator made by 3D printing, Nano
Energy, 2018, 45, 380–389, DOI: 10.1016/j.nanoen.20 17.12.049
Proposal of a 3D-TENG able to charge
common electronics through
harvesting energy from human
motions.
EC1

Future Internet 2025,17, 90 16 of 50
Table 7. Cont.
Reference Aim Motivation for
Exclusion
Matveev I. et al., Comparative Analysis of Object Detection Methods in Computer Vision for
Low-Performance Computers Towards Smart Lighting Systems, LNNS, 2023, 548, 203–215, DOI:
10.1007/978-3-031-16368-5_10
Comparison of the performance of DL
algorithms for object detection
targeted at low-performance
embedded microprocessors.
EC5
Nusrat M.A. et al., Practicle Coordination and Aspect of IoT for Smart Cities and Healthcare System, 12th
Inter. Conference on System Modeling and Advancement in Research Trends, SMART 2023, 280–287,
DOI: 10.1109/SMART59791.2023.10428643
An overview of smart city projects. EC1
Al-Daweri M.S. et al., Dynamic Temperature, Humidity, and Lighting System for Smart Home Based on
Fuzzy Logic, Advances in Science, Technology and Innovation, 2024, 149–164, DOI:
10.1007/978-3-031-52303-8_11
Design and testing of a fuzzy logic
home system. EC4
Higuera, J. et al., Smart lighting system ISO/IEC/IEEE 21451 compatible, IEEE Sensors Journal, 15 (5),
2015, 2595–2602, DOI: 10.1109/JSEN.2015.2390262
Adoption of a fuzzy logic algorithm
to determine light levels on an office
desk during the working day.
EC4
Polepaka S. et al., Internet of things and its applications: An overview, LNEE, 2020, 643, 67–75, DOI:
10.1007/978-981-15-3125-5_8
An overview of IoT elements and
application domains. EC2
Sharma V. et al., A novel study on IoT and Machine Learning-based transportation, Machine Learning
Techniques and Industry Applications, 2024, 1–28, DOI: 10.4018/979-8-3693-5271-7.ch001
On the relevance of AIoT with respect
to advancing the smart transportation
sector.
EC5
Anagnostopoulos T. et al., Challenges and Solutions of Surveillance Systems in IoT-Enabled Smart
Campus: A Survey, IEEE Access, 9, 2021, 131926–131954, DOI: 10.1109/ACCESS.2021.3114447
A comparative assessment around
surveillance systems for smart
campuses.
EC5
Bierzynski K. et al., The Learning of the OpenLicht system, a self-learning lighting system at the edge of
the network, Smart Systems Integration 2018—Inter. Conference and Exhibition on Integration Issues of
Miniaturized Systems, 155–162
Discussion of problems and
challenges around the
implementation of self-learning LSs.
EC6
Subbarao V. et al., A survey on internet of things based smart, digital green and intelligent campus, 4th
Inter. Conference on Internet of Things: Smart Innovation and Usages, IoT-SIU 2019, DOI:
10.1109/IoT-SIU.2019.8777476
An overview of components of a
future smart and digital green
educational campus.
EC5
Kanthi M. and Dilli R., Smart streetlight system using mobile applications: secured fault detection and
diagnosis with optimal powers, Wireless Networks, 29 (5), 2023, 2015–2028, DOI:
10.1007/s11276-023-03278-9
A smart streetlight controller
facilitating the control and
management of lighting through an
app.
EC2
Pestana E. and Paice A., Learning Algorithms for Building Control Applied to the iHomeLab Lighting
System, 2021, CEUR Workshop Proceedings, 3116 Proceedings of FTAL 2021, October 28–29, 2021,
Lugano, Switzerland
Description of an approach based on
IoT, digital twins and ML for the
configuration of building
management systems.
EC4
Qin F., Modern Intelligent Rural Landscape Design Based on Particle Swarm Optimization, Wireless
Communications and Mobile Computing, 2022, DOI: 10.1155/2022/8246368
The PSO method is investigated to
modernize the workflow of rural
landscape design.
EC5
Choi Y. et al., Predicting wearable IoT Adoption: Identifying core consumers through Machine Learning
algorithms. (2024) Telematics & Informatics, 93, 10.1016/j.tele.2024.102176
Investigation of the performance of
ML algorithms in predicting
consumer adoption of wearable
devices.
EC5
Zhang J. and He S., Smart technologies and urban life: A behavioral and social perspective, (2020)
Sustainable Cities and Society, 63, DOI: 10.1016/j.scs.2020.102460
The foreword of the special issue:
“Smart technologies and urban life: a
behavioral and social perspective”.
EC5
Samuel, R. et al., Smart living: Role of the internet of everything and the challenges (2022) Internet of
Everything: Smart Sensing Technologies, 1–30
A survey on the role of the Internet of
Everything in future smart cities. EC5
Vinh P.V. and Dung P.X., Designing a Smart Lighting System for Illuminating Learning Experiences,
LNNS, 1062 LNNS, 2024, 296–305, DOI: 10.1007/978-3-031-65656-9_30
Proposal of an open and active
learning environment. EC2
Puig, S. and Foukia, N., CleverTrash: An ML-based IoT system for waste sorting with continuous
learning cycle, Inter. Conference on Electrical, Computer,and Energy Technologies, ICECET 2022 DOI:
10.1109/ICECET55527.2022.9872943
Proposal of a waste-recognition
system that aims at educating people
to properly recycle their waste.
EC5
Puig, S. and Foukia, N., CleverTrash: an IoT system for waste sorting with Deep Learning, IEEE Inter.
Conferences on Internet of Things, iThings 2022, IEEE Green Computing and Communications,
GreenCom 2022, IEEE Cyber, Physical and Social Computing, CPSCom 2022 and IEEE Smart Data,
SmartData 2022, 1–8, DOI:
10.1109/iThings-GreenCom-CPSCom-SmartData-Cybermatics55523.2022.00016
The performance of the CNN part of a
waste-recognition system is explored.
EC5
Altrad, A. IoTs Traffics Detection and Analysis Using Machine Learning for Cybersecurity Application,
IEEE 5th Eurasia Conference on IoT, Communication and Engineering, ECICE 2023, 78–83, DOI:
10.1109/ECICE59523.2023.10383018
Application of the feature-extraction
technique to detect IoT’s benign and
attack traffic features.
EC5
Park, J.S. et al., Building IoT-based Zero-Contact Experimental Environment for Studying Picture
Preference under Various Illumination Conditions, Digest of Technical Papers—IEEE Inter. Conference
on Consumer Electronics, 2021 January, DOI: 10.1109/ICCE50685.2021.9427755
Various experiments are reported
under various lighting conditions. EC5
John, J. and Mahalingam, P, Automated Fish Feed Detection in IoT Based Aquaponics System, 8th Inter.
Conference on Smart Computing and Communications: Artificial Intelligence, AI Driven Applications
for a Smart World, ICSCC 2021, 286–290, DOI: 10.1109/ICSCC51209.2021.9528186
Detection of excess fish feed on the
water surface by adopting an
object-detection algorithm.
EC5
Pridmore J. and Mols A., Personal choices and situated data: Privacy negotiations and the acceptance of
household Intelligent Personal Assistants, Big Data and Society, 2020, 7(1), DOI:
10.1177/2053951719891748
Investigation on how people
negotiate and make choices about
household intelligent personal
assistants.
EC5

Future Internet 2025,17, 90 17 of 50
Table 7. Cont.
Reference Aim Motivation for
Exclusion
Singh A.K. et al., Future Technology: Internet of Things (IoT) in Smart Society 5.0, (2023) Intelligent
Techniques for Cyber-Physical Systems, 245–265, DOI: 10.1201/9781003438588-15
A book chapter on the role of AIoT in
the development of Smart Society 5.0.
EC5
Thomas A.S. and Robinson A.Y., IoT, Big Data, Blockchain and Machine Learning Besides its
Transmutation with Modern Technological Applications, Intelligent Systems Reference Library, 2020,
180, 47–63, DOI: 10.1007/978-3-030-39119-5_4
Overview of ML techniques and IoT
applications in the transportation
domain.
EC5
Rahman, M.A. et al., IoT based Comprehensive Approach Towards Shaping Smart Classrooms, Proc. of
the 5th Inter. Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud), I-SMAC 2021, 103–109,
DOI: 10.1109/I-SMAC52330.2021.9640669
An application for reducing power
consumption in institutions by using
cameras to detect human presence
and operate appliances accordingly.
EC6
Yu, T. et al., Design strategy of green intelligent building using deep belief network, Inter. Journal of
System Assurance Engineering and Management, 14 (1), 2023, 196–205, DOI:
10.1007/s13198-021-01513-0
The LS of a medical building is
designed by adopting the biophysical
design theory and IoT technology.
EC4
Bernardo M. et al., End-Product of Solar-Sharing Smart Lighting Artificial Intelligence Driven Platform
for High-Valued Crops (Lactuca Sativa) on Indoor Hydroponics System, IEEE 10th Conference on
Systems, Process and Control, ICSPC 2022, 160–165, DOI: 10.1109/ICSPC55597.2022.10001821
An IoT-based LED LS to control the
light intensity in an indoor
environment. The system adopts a
fuzzy logic controller.
EC4
Bernardo M.S., DLI and PPFD throughput of Solar and AI-Based Smart Lighting Apply on Illumination
Stratums, IEEE 11th Conference on Systems, Process and Control, ICSPC 2023, 171–176, DOI:
10.1109/ICSPC59664.2023.10419940
A solar-sharing smart LS that controls
LED indoor illumination. The system
adopts a fuzzy logic controller.
EC4
Rahman M. et al., IoT and ML Based Approach for Highway Monitoring and Streetlamp Controlling,
Lecture Notes of the Institute for Computer Sciences, Social-Informatics and Telecommunications
Engineering, LNICST, 491 LNICST, 2023, 376–385, DOI: 10.1007/978-3-031-34622-4_30
An AIoT-based system to control
street lamps to provide illumination
according to the brightness of the
area.
EC6
Sung W.-T. et al., Smart Lamp Using Google Firebase as Realtime Database, Intelligent Automation and
Soft Computing, 2022, 33 (2), pp. 967–982, DOI: 10.32604/iasc.2022.024664
A discussion on the manufacturing of
smart lamps. EC1
Khoa T.A. et al., Designing efficient smart home management with IoT smart lighting: A Case Study,
Wireless Communications and Mobile Computing, 2020, DOI: 10.1155/2020/8896637
A proposal to strengthen home
security by using IoT. EC1
Thipards R. et al., Smart Street Lighting Control for Electrical Power on Saving by IoT, Inter. Computer
Science and Engineering Conference 2022, 55–60, DOI: 10.1109/ICSEC56337.2022.10049363
A system saving energy by dimming
the electricity when there are no
people or vehicles on the street.
EC7
Kumar P. et al., Smart lighting and switching using Internet of Things, 11th Inter. Conference on Cloud
Computing, Data Science and Engineering, 2021, 536–539, DOI:10.1109/Confluence51648.2021.9377078
Proposal of an IoT-based method for
sensing and monitoring of smart LSs
through an app.
EC4
Rauniyar K.R. and Khan J.A., Application of IoT and AI in the Development of Smart Cities Smart Cities
Concepts, Practices, and Applications, 2022, 181–196, DOI: 10.1201/9781003287186-7
A primer about the services that AIoT
can provide to smart cities. EC5
Garg S. et al., Real time adaptive street lighting system, Communications in Computer and Information
Science, 1192 CCIS, 2020, 223–239, DOI: 10.1007/978-981-15-3666-3_19
Proposal of a fuzzy logic-based
data-aggregation strategy for the
deployment of energy efficient street
LSs.
EC4
Shao Z. et al., Analysis of the opportunities and costs of energy saving in lighting system of library
buildings with the aid of building information modelling and Internet of things, Fuel, 352, 2023, DOI:
10.1016/j.fuel.2023.128918
About strategies for prototyping an
IoT- and BIM-based efficient LS of a
university library as an alternative to
the use of ML.
EC3
Goyal S.B. et al., Smart Luminaires for Commercial Building by Application of Daylight Harvesting
Systems, LNNS, 2022, 218, 293–305, DOI: 10.1007/978-981-16-2164-2_24
A daylight-harvesting strategy
integrated with AI to upgrade the
infrastructure of existing commercial
buildings.
EC4
Yan G., Intelligent Lighting System based on Digital Twins in Smart Home, Proc. of SPIE—The Inter.
Society for Optical Engineering, 12940, 2023, DOI: 10.1117/12.3010705
An introduction to the adoption of the
digital twin technology in the smart
home domain.
EC4
Study Quality Assessment
SLRs must adopt quality criteria suitable to exclude poor-quality studies that may
bias the result synthesis, while the best available evidence is taken into consideration.
References [
5
,
62
] recommend investigating the quality of selected studies in order to
enhance the filtering effectiveness of the inclusion/exclusion criteria. This statement has
been reiterated and enriched more recently in [
63
]. References [
61
,
64
–
66
] are SLRs where
such a perspective is adopted. In the present study, the output of the quality-assessment
stage of both primary and secondary studies is used for the same purpose. We evaluated
the quality of primary and secondary studies in sequence, as explained in the following.

Future Internet 2025,17, 90 18 of 50
Quality of Primary Studies
A primary source in science is a document or article that reports on a study, experiment,
case study, survey, research project, and so on. Primary studies are usually written by the
person(s) who carried out the research, conducted the study, or ran the experiment(s), and
include hypothesis, methodology, and results. So, the structure and aim of primary studies
are deeply different from secondary studies; that is why it is not possible to apply to them
the same quality criteria of the latter. It has been remarked that the quality of findings
and conclusions of an SLR directly depends on the quality of the primary studies selected
for the study [
63
]. Nevertheless, it is worth noticing that Budgen et al. [
67
] carried out a
tertiary study consisting in analyzing 37 secondary studies to assess how well these studies
are reported. They observed that just 8 papers out of 37 gave a score to primary studies.
Recently, Santos et al. [
9
] linked such behaviour of researchers with the time required for
scoring the quality of primary studies as their number increases constantly. Assessing the
quality of primary studies varies greatly as a consequence of the lack of a shared definition
about such a concept [
5
,
63
]. In the present SLR, we adopted a set of quality criteria based
on the recent SLR by Yang et al. [
63
], where a paper’s quality is expressed by what authors
call the “study characteristic”. By analyzing 241 SLRs in the software-engineering domain,
between 2004 and 2018, Yang et al. observed that the characteristics of primary studies were
scored by taking into account either (a) reporting, rigor, credibility, and relevance or (b)
rigor and industrial relevance. For example, ref. [
68
] follows the first approach. In contrast,
we followed the other one since such a method captures the perspective of the B2B firm
that commissioned the review reported in this paper. Hereafter, we briefly recall the second
alternative for readers to be able to understand the rationale behind the quality-assessment
criteria (QC) introduced shortly. Ivarsson and Gorschek [
69
] defined scientific rigor and
relevance as follows.
Rigor =context +study design +validity (1)
where:
•context
refers to the description of techniques, product, tools, and people necessary to
follow the study, compare it with others, and replicate it;
•
study design refers to the description of the approaches used for data collection
and analysis;
•validity
refers to the discussion of any limitations or threats to the validity of the study.
As pointed out by Ivarsson and Gorschek, the rigor characteristic refers to the extent
to which the method adopted in the study is presented, while the evaluation of its actual
rigor is out of the scope of the model.
Relevance =context +research method +subjects +scale (2)
where:
•context concerns whether the industrial context is representative;
•
research method concerns whether the method adopted in the study adds value to industry;
•subjects
concerns whether the subjects of the study are representative of practitioners;
•scale
concerns whether the size of the study is meaningful (“toy” examples are useless).
From Equation (1) the following QC follow:
QC1. Is there a description of the context?
QC2. Is there a description of the study design?
QC3. Are study limitations discussed?
From Equation (2) the following QC follow:

Future Internet 2025,17, 90 19 of 50
QC4. Is there a description of the context?
QC5. Is there a description of the research method?
QC6. Are the subjects of the study described and representative of practitioners?
QC7. Concerning the size of the study, is it on an industrial scale?
To the previous QC, the following one was added:
QC8. Is the publication venue a journal, a conference, or a book chapter?
QC8 is frequently considered in SLRs [61].
The QC score of each primary study was calculated using the following schema:
N(o) = 0, Y(es) = 1, P(artial) = 0.5. In detail, the QC were scored as follows:
QC1. The context is implicit: N; the context is clearly sketched, so the reader can
understand and compare it to other contexts: Y; the context can be inferred: P.
QC2. The study design is omitted: N; the study design is properly described, so the
reader can understand it: Y; the study design is just sketched: P.
QC3. The description of threats to validity is omitted: N; threats to validity are
discussed in detail: Y; threats to validity are partially defined: P.
QC4. The evaluation is either performed in a setting not representative of the reality
or omitted: N; the evaluation is carried out in an industrial setting representative of the
intended usage setting: Y; the setting is only partially representative of the intended final
setting: P.
QC5. The research method is not detailed: N; the research method used in the
evaluation is relevant for the actual practitioners: Y; the research method is just sketched: P.
QC6. The subjects used in the evaluation are either not mentioned or not representative
of the actual practitioners: N; the subjects used in the evaluation are representative of the
actual practitioners: Y; the subjects used in the evaluation are only partially representative
of the actual practitioners: P.
QC7. The evaluation is missing: N; the application used in the evaluation is of
industrial scale: Y; the evaluation was carried out by using a small application size: P.
QC8. The publication venue is either a conference or a book chapter: N; the publication
venue is a top-class journal (i.e., an ACM, IEEE, Elsevier, Springer, Wiley publication): Y;
the publication venue is not a top-class journal; however, it is indexed by Scopus: P.
Quality of Secondary Studies
The quality of the selected secondary studies was evaluated by applying the follow-
ing QCs.
QC1. Are the RQs explicit?
QC2. Is the review protocol well-defined?
QC3. Is the search string reported?
QC4. Are the inclusion criteria described and related to the study goals?
QC5. Are the exclusion criteria described and related to the study goals?
QC6. Are the years covered in the review declared?
QC7. Do the queried scientific databases ensure that the result of the search can cover all
the significant published studies?
QC8. Is the quality of the included studies assessed?
QC9. Is the data-extraction activity adequate to the purpose of the study?
QC10. Is the data-synthesis method described?
QC11. Are the selected studies adequately illustrated?
QC12. Does the study adequately answer the RQs?
QC13. Are the open issues adequately discussed?
QC14. Is there an analysis of the threats to validity?

Future Internet 2025,17, 90 20 of 50
QC15. Is the publication venue a journal, a conference, or a book chapter?
The QC score of each study was calculated using the following schema: N(o) = 0,
Y(es) = 1, P(artial) = 0.5. In detail, the QCs were scored as follows:
QC1. The RQs are implicit: N; the RQs are explicit: Y; the RQs can be inferred: P.
QC2. The review protocol is missing: N; the review protocol is well-defined: Y; the
review protocol is not fully defined: P.
QC3. The search string is not reported: N; the search string is reported: Y; the search
string can be inferred: P.
QC4. The inclusion criteria are not mentioned: N; the inclusion criteria are described
and they relate to the study goals: Y; the inclusion criteria are not explicitly described: P.
QC5. The exclusion criteria are not mentioned: N; the exclusion criteria are described
and they relate to the study goals: Y; the exclusion criteria are not explicitly described: P.
QC6. The years covered in the review are not declared: N; the years covered in the
review are declared: Y; the years covered in the review can be inferred: P.
QC7. The searched databases are not mentioned: N; the searched databases are
scientifically relevant and ensure a wide coverage of published articles: Y; the authors have
searched only a small set of scientific sources (either journals or proceedings). So, not all
the relevant published articles are investigated: P.
QC8. The quality-assessment stage of the selected studies is omitted: N; quality criteria
are defined and applied to each primary study: Y; the quality-assessment strategy can be
inferred: P.
QC9. The data-extraction activity is not described: N; the data-extraction activity
is adequate to the purpose of the study: Y; the data-extraction activity does not provide
sufficient details on all the aspects that are part of such a stage (e.g., the output of the
exclusion criteria, the procedure to remove duplicate publications, the type of paper (e.g.,
article, book chapter, or conference paper), etc.): P.
QC10. The data-synthesis method is not described: N; it is described: Y; it could be
derived: P.
QC11. The findings of each study are not specified: N; the information is carefully
summarized for each study: Y; only a recapped description is given for each individual
paper: P.
QC12. The study does not answer the RQs: N; the study adequately answers the RQs:
Y; the answer to the RQs is not fully satisfactory: P.
QC13. Open issues are not discussed: N; open issues are listed and adequately
discussed in the study: Y; open issues are listed, but emerging lines of research potentially
suitable for solving them are just touched: P.
QC14. Threats to validity analysis are omitted: N; there is an analysis of the threats
to validity and countermeasures are taken to limit potential threats: Y; the analysis of the
threats to validity is superficial: P.
QC15. The publication venue is either a conference or a book chapter: N; the publica-
tion venue is a top-class journal (i.e., an ACM, IEEE, Elsevier, Springer, Wiley publication):
Y; the publication venue is not a top-class journal; however, it is indexed by Scopus: P.
As suggested in [
5
], the score of each QC was extracted by one of the authors and
checked by the other ones to prevent errors. The score of the 26 analyzed papers is shown
in the following. Tables 8and 9show, respectively, the result of the quality-assessment
stage of the 22 primary studies and 4 secondary ones that entered the secondary selection
phase. Three primary studies and two secondary studies were excluded in this secondary
selection phase of the review methodology because their score is very low. Tables 10 and 11
list them, in sequence.

Future Internet 2025,17, 90 21 of 50
Table 8. Quality-assessment results for the 22 primary studies.
Reference QC1 QC2 QC3 QC4 QC5 QC6 QC7 QC8 Score
[70] 1 1 1 1 1 1 1 1 8
[71] 1 1 0 1 1 1 1 1 7
[72] 1 1 0.5 1 1 0.5 0.5 1 6.5
[73] 1 1 0.5 0.5 1 0.5 0.5 1 6
[74] 1 1 0 1 1 1 1 0 6
[75] 1 1 0 1 1 1 1 0 6
[76] 1 1 0.5 1 1 1 0.5 0 6
[77] 1 1 0 0.5 1 0.5 0.5 1 5.5
[78] 1 1 0.5 0.5 1 1 0.5 0 5.5
[79] 1 1 0 1 1 0.5 0.5 0 5
[80] 1 1 0 0.5 1 1 0.5 0 5
[81] 1 1 0 1 1 0.5 0.5 0 5
[82] 1 1 0 1 1 0 0 1 5
[83] 1 1 0 0.5 1 0.5 0.5 0 4.5
[84] 1 1 0 0.5 1 0.5 0.5 0 4.5
[85] 1 1 0 0.5 1 0.5 0.5 0 4.5
[86] 1 1 0 0.5 1 0.5 0.5 0 4.5
[87] 1 1 0 0.5 1 0.5 0.5 0 4.5
[88] 1 1 0.5 0.5 1 1 0.5 0 4.5
(Tang2021) 1 1 0 0.5 1 0 0.5 0 4
(Huang2022) 1 1 0 1 0 0.5 0 0 3.5
(Martin2021) 0 0 0 0 1 1 0 0 2
Table 9. Quality-assessment results for the secondary studies.
Ref.
QC1 QC2 QC3 QC4 QC5 QC6 QC7 QC8 QC9 QC10 QC11 QC12 QC13 QC14 QC15 Score
[4] 1 1 0.5 1 1 1 1 1 1 0.5 1 1 1 0 1 13
[89] 0.5 0 0 0 0 0 0 0 0 0 1 1 1 0 1 4.5
(Mukhopadhyay2024) 0.5 0 0 0 0 0 0 0 0 0 1 0.5 1 0 0.5 3.5
(Patil2018) 0 0 0 0 0 0 0 0 0 0 0 0 0.5 0 0 0.5
Table 10. Reference to the three primary studies excluded at the quality-assessment stage.
(Tang2021) Tang D. et al., An Intelligent Fault Diagnosis Method for Street Lamps, Inter.
Conference on Internet, Education, and Information Technology, IEIT 2021, 300–303, DOI:
10.1109/IEIT53597.2021.00073
(Huang2022) Huang Y., Design of Rural Road Lighting System Based on Internet of Things and
Deep Learning, Inter. Conference on Industrial IoT, Big Data and Supply Chain, IIoTBDSC 2022,
6–9, DOI: 10.1109/IIoTBDSC57192.2022.00012
(Martin2021) Martin G. et al., AI-TWILIGHT: AI-digital TWIn for LIGHTing— A new European
project, 27th Inter. Workshop on Thermal Investigations of ICs and Systems, THERMINIC 2021,
DOI: 10.1109/THERMINIC52472.2021.9626541
Table 11. Reference to the two secondary studies excluded at the quality-assessment stage.
(Mukhopadhyay2024) Mukhopadhyay S. et al., A Review and Analysis of IoT Enabled Smart
Transportation Using Machine Learning Techniques, Inter. Journal of Transport Development and
Integration, 8 (1), 2024, 61–77, DOI: 10.18280/ijtdi.080106
(Patil2018) Patil A.A. and Badgujar V.S., A Comprehensive Survey on Theoretic Perspective
Providing Future Directions on IoT, Inter. Conference on Smart City and Emerging Technology,
ICSCET 2018, DOI: 10.1109/ICSCET.2018.8537285
Despite the score of paper [
89
] being low too (4.5, Table 9) and close to that of
(Mukhopadhyay2024), we decided to keep it in the final set of studies to be investigated in
detail. This choice was determined by observing that [
89
] has 416 citations (up to 13 Octo-
ber 2024), of which 203 are from journal papers. Table 12 shows the citation distribution
over the years. Citation-based paper selection is frequent in reviews and this is because,
the higher the number of citations, the higher the paper quality [
7
]. The main reason for
such interest towards [
89
] by the researcher community is because it was the first review
investigating the ways in which the transportation domain can benefit from the AoIT. It

Future Internet 2025,17, 90 22 of 50
is also fair to remark that the study obtained a score of 4.5 because it does not follow any
well-structured research methodology. That is the reason why quality criteria QC2-QC10
scored zero.
Table 12. Distribution over the years of citations of reference [89].
Year 2019 2020 2021 2022 2023 2024 2025
Total citations 6 38 60 110 115 86 1
Citations from journal articles 4 15 26 56 60 41 1
3.2.2. Identification of Citing Studies
Below is reported in detail how the second set of studies to be considered for further
investigation was built through the forward snowballing (Figure 5). Detailing the snow-
balling strategy is a best practice that in published SLRs is often ignored, as observed by
Budgen et al. [
67
] in their tertiary study. The second column of Table 13 shows the number
of citations received by each paper belonging to the first set of selected studies, while the
third column shows the number of citations that are pertinent to the present SLR. The
papers without citations are not in the list.
Table 13. Total number of citations against the number of pertinent citations received by the studies
belonging to the first set of selected studies.
Studies in the First Set Number of Citations Pertinent Citations
[89] 425 6
[4] 46 16
[80] 6 2
[82] 87 2
[70] 33 0
[71] 22 6
[77] 9 8
[72] 8 0
[83] 6 4
[73] 3 0
[75] 3 0
[76] 4 1
Inclusion Criteria and Skipping of Repetitions
The papers found through the forward snowballing do not satisfy the search string
that was used to query the Scopus database; otherwise, they would have been extracted
at that stage. A citing paper is labelled as pertinent to the present SLR if the “
smart
lighting
” string is present either in its title, abstract, or among the authors’ keywords.
The total number of pertinent citations (#PCs) received by the 12 papers in Table 13 is
45 (third column). Each row in Table 14 denotes, in order, a paper in Table 13, the #PCs,
and the reference to the citing study. Table 15, in contrast, lists the citing papers and their
number of repetitions: 1 denotes no repetition. From such a table it emerges that there are
31 distinct citing papers. But, since the last eight papers are already present in the initial
set of 21 primary studies, it follows that the papers to be evaluated against the exclusion
criteria and then against the quality criteria are 23.

Future Internet 2025,17, 90 23 of 50
Table 14. List of papers citing the selected 21.
Papers in Table 13
#PCs Reference to the Citing Paper
[80] 2
Putrada, A.G. et al., EdgeSL: Edge-Computing Architecture on Smart Lighting Control With
Distilled KNN for Optimum Processing Time, (2023) IEEE Access, 11, 64697–64712. DOI:
10.1109/ACCESS.2023.3288425
Putrada, A.G. et al., Machine Learning Methods in Smart Lighting Toward Achieving User
Comfort: A Survey, (2022) IEEE Access, 10, 45137–45178. DOI: 10.1109/ACCESS.2022.3169765
[82] 2
Agramelal, F. et al., Smart Street Light Control: A Review on Methods, Innovations,
and Extended Applications, (2023) Energies, 16 (21), DOI: 10.3390/en16217415
Chiradeja, P. and Yoomak, S., Development of public lighting system with smart lighting
control systems and internet of thing (IoT) technologies for smart city, (2023) Energy Reports,
10, 3355–3372, DOI: 10.1016/j.egyr.2023.10.027
[71] 6
Wang, Y. and Durmus, D., Image Quality Metrics, Personality Traits,and Subjective Evaluation
of Indoor Environment Images, (2022) Buildings, 12 (12), DOI: 10.3390/buildings12122086
Vale,Z., et al., An overview on smart buildings, (2022) Encyclopedia of Electrical and Electronic
Power Engineering: Volumes 1–3, 2, V2-431-V2-440. DOI: 10.1016/B978-0-12-821204-2.00066-0
Putri, A.K. et al., The Smart Lighting System in the Coworking Space’s Meeting Room, 4th Inter.
Conference on Informatics, Multimedia, Cyber and Information System, ICIMCIS 2022,
534–538, DOI: 10.1109/ICIMCIS56303.2022.10017802
Daniel, W. et al., Integrated Smart Lighting Dashboard on the Office Desk to Accommodate
User Activity, 10th Inter. Conference on Cyber and IT Service Management, CITSM 2022, DOI:
10.1109/CITSM56380.2022.9935875
Widartha, V.P. et al., Advancing Smart Lighting: A Developmental Approach to Energy
Efficiency through Brightness Adjustment Strategies, (2024) Journal of Low Power Electronics
and Applications, 14 (1), DOI: 10.3390/jlpea14010006
Parise, G. et al., A Comprehensive Exploration of Smart Lighting Aspects: Area of Use,
Methodologies and Purposes, IEEE Industry Applications Society Annual Meeting, IAS 2023,
DOI: 10.1109/IAS54024.2023.10406744
[4] 16
Putrada, A.G. et al., EdgeSL: Edge-Computing Architecture on Smart Lighting Control with
Distilled KNN for Optimum Processing Time, (2023) IEEE Access, 11, 64697–64712, DOI:
10.1109/ACCESS.2023.3288425
Zhu, J. et al., Data-Driven End-to-End Lighting Automation Based on Human Residential
Trajectory Analysis, Inter. Conference on Smart Applications, Communications and
Networking, SmartNets 2024.
Barandas, M. et al., Iterative wireless node localization based on Bluetooth and visible light for
smart lighting systems, (2024) Wireless Telecommunications Symposium, DOI:
10.1109/WTS60164.2024.10536676
Aizono, Y. et al., Building Automation with Vision Transformer Using Synthetic Indoor Images
for Room Light Control, KST 2024—16th Inter. Conference on Knowledge and Smart
Technology, 40–44, DOI: 10.1109/KST61284.2024.10499683
Putrada, A.G. et al., Q8KNN: A Novel 8-Bit KNN Quantization Method for Edge Computing in
Smart Lighting Systems with NodeMCU, 824 LNNS, 2024598-615. DOI:
10.1007/978-3-031-47715-7_41
Mohammadrezaei, E. et al., Systematic Review of Extended Reality for Smart Built
Environments Lighting Design Simulations, (2024) IEEE Access, 12, 17058–17089, DOI:
10.1109/ACCESS.2024.3359167
Zhang, J. et al., Intelligent Personalized Lighting Control System for Residents, (2023)
Sustainability (Switzerland), 15 (21), DOI: 10.3390/su152115355
Agramelal, F. et al., Smart Street Light Control: A Review on Methods, Innovations,
and Extended Applications, (2023) Energies, 16 (21), DOI: 10.3390/en16217415
Cerpentier, J. et al., Adaptive museum lighting using CNN-based image segmentation, (2023)
Building and Environment, 242, DOI: 10.1016/j.buildenv.2023.110552
Parise, G. et al., A Comprehensive Exploration of Smart Lighting Aspects: Area of Use,
Methodologies and Purposes, IEEE Industry Applications Society Annual Meeting, IAS 2023,
DOI: 10.1109/IAS54024.2023.10406744
Prabowo, S. et al., Camera-Based Smart Lighting System that complies with Indonesia’s
Personal Data Protection Act, ICADEIS 2023—Inter. Conference on Advancement in Data
Science, E-Learning and Information Systems: Data, Intelligent Systems, and the Applications
for Human Life, Proceeding, DOI: 10.1109/ICADEIS58666.2023.10271086
Petkovic, M. et al., Smart Dimmable LED Lighting Systems, (2022) Sensors, 22 (21), DOI:
10.3390/s22218523
Hadi, A. et al., Office Room Smart Lighting Control with Camera and SSD MobileNet Object
Localization, 2022 Inter. Conference on Advanced Creative Networks and Intelligent Systems:
Blockchain Technology, Intelligent Systems, and the Applications for Human Life, DOI:
10.1109/ICACNIS57039.2022.10055274
Putrada, A.G., et al., Recurrent Neural Network Architectures Comparison in Time-Series
Binary Classification on IoT-Based Smart Lighting Control, 10th Inter. Conference on
Information and Communication Technology, ICoICT 2022, 391–396, DOI:
10.1109/ICoICT55009.2022.9914831
Putrada, A.G. et al., CIMA: A Novel Classification-Integrated Moving Average Model for
Smart Lighting Intelligent Control Based on Human Presence, Complexity, 2022, DOI:
10.1155/2022/4989344
Putrada, A.G. et al., Synthetic Data with Nested Markov Chain for CIMA-Based Smart Lighting
Control Deployment Simulation, 11th Inter. Conference on Information and Communication
Technology, ICoICT 2023, August, 148–153. DOI: 10.1109/ICoICT58202.2023.10262430

Future Internet 2025,17, 90 24 of 50
Table 14. Cont.
Papers in Table 13
#PCs Reference to the Citing Paper
[83] 4
Putrada, A.G. et al., Homomorphic Encryption for Privacy Preservation in Occupancy
Sensor-Based Smart Lighting, Inter. Conference on Data Science and Its Applications, ICoDSA
2024, 168–173, DOI: 10.1109/ICoDSA62899.2024.10651987
Prabowo, S. et al., Camera-Based Smart Lighting System that complies with Indonesia’s
Personal Data Protection Act, ICADEIS 2023—Inter. Conference on Advancement in Data
Science, E-Learning and Information Systems: Data, Intelligent Systems, and the Applications
for Human Life, Proceeding, DOI: 10.1109/ICADEIS58666.2023.10271086
Putrada, A.G. et al., Machine Learning Methods in Smart Lighting Toward Achieving User
Comfort: A Survey, (2022) IEEE Access, 10, 45137–45178, DOI: 10.1109/ACCESS.2022.3169765
Putrada, A.G. et al., An Evaluation of Activity Recognition with Hierarchical Hidden Markov
Model and other Methods for Smart Lighting In Office Buildings, (2022) ICIC Express Letters,
16 (1), 91–100, DOI: 10.24507/icicel.16.01.91
[76] 1 Putrada, A.G. et al., Machine Learning Methods in Smart Lighting Toward Achieving User
Comfort: A Survey, (2022) IEEE Access, 10, 45137–45178, DOI: 10.1109/ACCESS.2022.3169765
[89] 6
Putrada, A.G. et al., Machine Learning Methods in Smart Lighting Toward Achieving User
Comfort: A Survey, (2022) IEEE Access, 10, 45137–45178, DOI: 10.1109/ACCESS.2022.3169765
Cerpentier, J. et al., Adaptive museum lighting using CNN-based image segmentation, (2023)
Building and Environment, 242, DOI: 10.1016/j.buildenv.2023.110552
Cerpentier, J. et al., Smooth output from adaptive illumination systems with pixelated LED
arrays, (2023) Proceedings of SPIE—The Inter. Society for Optical Engineering, 12765, DOI:
10.1117/12.2688496
Cerpentier, J. et al., Controlling the target pattern of projected LED arrays for smart lighting,
(2023) Optics Express, 31 (22), 37316–37324, DOI: 10.1364/OE.504077
Sharma, V. et al., A novel study on IoT and Machine Learning-based transportation, (2024)
Machine Learning Techniques and Industry Applications, 1–28, DOI:
10.4018/979-8-3693-5271-7.ch001
Mukhopadhyay, S. et al., A Review and Analysis of IoT Enabled Smart Transportation Using
Machine Learning Techniques, (2024) Inter. Journal of Transport Development and Integration,
8 (1), 61–77. DOI: 10.18280/ijtdi.080106
[77] 8
Putrada, A.G. et al., Homomorphic Encryption for Privacy Preservation in Occupancy
Sensor-Based Smart Lighting, Inter. Conference on Data Science and Its Applications, ICoDSA
2024, 168–173, DOI: 10.1109/ICoDSA62899.2024.10651987
Putrada, A.G. et al., NearCount for Model Compression on Edge Computing-Based Smart
Lighting with Product-of-Sum Function, Inter. Conference on Smart Computing, IoT and
Machine Learning, SIML 2024, 13–18, DOI: 10.1109/SIML61815.2024.10578110
Putrada, A.G. et al., Q8KNN: A Novel 8-Bit KNN Quantization Method for Edge Computing in
Smart Lighting Systems with NodeMCU, (2024) LNNS, 824 LNNS, 598–615. DOI:
10.1007/978-3-031-47715-7_41
Putrada, A.G. et al., SLTAM: Remodelling Technology Acceptance Model to Measure User
Comfort in Smart Lighting with Exploratory Factor Analysis, 3rd Inter. Conference on
Intelligent Cybernetics Technology and Applications, ICICyTA 2023, 414–419. DOI:
10.1109/ICICyTA60173.2023.10428882
Putrada, A.G. et al., Synthetic Data with Nested Markov Chain for CIMA-Based Smart Lighting
Control Deployment Simulation, 11th Inter. Conference on Information and Communication
Technology, ICoICT 2023, August, 148–153, DOI: 10.1109/ICoICT58202.2023.10262430
Hadi, A. et al., Office Room Smart Lighting Control with Camera and SSD MobileNet Object
Localization, ICACNIS 2022—Inter. Conference on Advanced Creative Networks and
Intelligent Systems: Blockchain Technology, Intelligent Systems, and the Applications for
Human Life, Proceeding, DOI: 10.1109/ICACNIS57039.2022.10055274
Putrada, A.G. et al., Recurrent Neural Network Architectures Comparison in Time-Series
Binary Classification on IoT-Based Smart Lighting Control
Putrada, A.G. et al., Machine Learning Methods in Smart Lighting Toward Achieving User
Comfort: A Survey, (2022) IEEE Access, 10, 45137–45178, DOI: 10.1109/ACCESS.2022.3169765
Table 15. List of citing papers and the number of repetitions.
Citing Papers # of Repetitions
Agramelal, F. et al., Smart Street Light Control: A Review on Methods, Innovations, and Extended
Applications, (2023) Energies, 16 (21), DOI: 10.3390/en16217415 2
Barandas, M. et al., Iterative wireless node localization based on Bluetooth and visible light for smart
lighting systems, (2024) Wireless Telecommunications Symposium, DOI:
10.1109/WTS60164.2024.10536676
1
Cerpentier, J. et al., Smooth output from adaptive illumination systems with pixelated LED arrays, (2023)
Proceedings of Inter. Society for Optical Engineering, 12765. DOI: 10.1117/12.2688496 1
Cerpentier, J. et al., Controlling thetarget pattern of projected LED arrays for smart lighting, (2023) Optics
Express, 31 (22), 37316–37324. DOI: 10.1364/OE.504077 1
Cerpentier, J. et al., Adaptive museum lighting using CNN-based image segmentation, (2023) Building and
Environment, 242, DOI: 10.1016/j.buildenv.2023.110552 2
Chiradeja, P. and Yoomak, S., Development of public lighting system with smart lighting control systems
and internet of thing (IoT) technologies for smart city, (2023) Energy Reports, 10, 3355–3372, DOI:
10.1016/j.egyr.2023.10.027
1
Daniel, W. et al., Integrated Smart Lighting Dashboard on the Office Desk to Accommodate User Activity,
10th Inter. Conference on Cyber and IT Service Management, CITSM 2022, DOI:
10.1109/CITSM56380.2022.9935875
1

Future Internet 2025,17, 90 25 of 50
Table 15. Cont.
Citing Papers # of Repetitions
Putrada, A.G. et al., Recurrent Neural Network Architectures Comparison in Time-Series Binary
Classification on IoT-Based Smart Lighting Control, 10th Inter. Conference on Information and
Communication Technology, ICoICT 2022, 391–396. DOI: 10.1109/ICoICT55009.2022.9914831
2
Hadi, A. et al., Office Room Smart Lighting Control with Camera and SSD MobileNet Object Localization,
ICACNIS 2022—Inter. Conference on Advanced Creative Networks and Intelligent Systems: Blockchain
Technology, Intelligent Systems, and the Applications for Human Life, Proceeding. DOI:
10.1109/ICACNIS57039.2022.10055274
2
Mohammadrezaei, E. et al., Systematic Review of Extended Reality for Smart Built Environments Lighting
Design Simulations, (2024) IEEE Access, 12, 17058–17089. DOI: 10.1109/ACCESS.2024.3359167 1
Parise, G. et al., A Comprehensive Exploration of Smart Lighting Aspects: Area of Use, Methodologies and
Purposes, IEEE Industry Applications Society Annual Meeting, IAS 2023. DOI:
10.1109/IAS54024.2023.10406744
2
Petkovic, M. et al., Smart Dimmable LED Lighting Systems, (2022) Sensors, 22 (21), DOI: 10.3390/s22218523
1
Prabowo, S. et al., Camera-Based Smart Lighting System that complies with Indonesia’s Personal Data
Protection Act, ICADEIS 2023—Inter. Conference on Advancement in Data Science, E-Learning and
Information Systems: Data, Intelligent Systems, and the Applications for Human Life, Proceeding. DOI:
10.1109/ICADEIS58666.2023.10271086
2
Putrada, A.G. et al., EdgeSL: Edge-Computing Architecture on Smart Lighting Control with Distilled KNN
for Optimum Processing Time, (2023) IEEE Access, 11, 64697–64712. DOI: 10.1109/ACCESS.2023.3288425
2
Putrada, A.G. et al., Q8KNN: A Novel 8-Bit KNN Quantization Method for Edge Computing in Smart
Lighting Systems with NodeMCU, (2024) 824 LNNS, 598–615. DOI: 10.1007/978-3-031-47715-7_41 2
Putrada, A.G. et al., CIMA: A Novel Classification-Integrated Moving Average Model for Smart Lighting
Intelligent Control Based on Human Presence, Complexity, 2022. DOI: 10.1155/2022/4989344 1
Putrada, A.G. et al., Homomorphic Encryption for Privacy Preservation in Occupancy Sensor-Based Smart
Lighting, Inter. Conference on Data Science and Its Applications, ICoDSA 2024, 168–173. DOI:
10.1109/ICoDSA62899.2024.10651987
2
Putrada, A.G. et al., SLTAM: Remodelling Technology Acceptance Model to Measure User Comfort in
Smart Lighting with Exploratory Factor Analysis, 3rd Inter. Conference on Intelligent Cybernetics
Technology and Applications, ICICyTA 2023, 414–419. DOI: 10.1109/ICICyTA60173.2023.10428882
1
Putrada, A.G. et al., Synthetic Data with Nested Markov Chain for CIMA-Based Smart Lighting Control
Deployment Simulation, 11th Inter. Conference on Information and Communication Technology, ICoICT
2023, August, 148–153. DOI: 10.1109/ICoICT58202.2023.10262430
2
Putri, A.K. et al., The Smart Lighting System in the Coworking Space’s Meeting Room, 4th Inter.
Conference on Informatics, Multimedia, Cyber and Information System, ICIMCIS 2022, 534–538. DOI:
10.1109/ICIMCIS56303.2022.10017802
1
Wang, Y. et al., Quality Metrics, Personality Traits,and Subjective Evaluation of Indoor Environment
Images, (2022) Buildings, 12 (12), DOI: 10.3390/buildings12122086 1
Widartha, V.P. et al., Advancing Smart Lighting: A Developmental Approach to Energy Efficiency through
Brightness Adjustment Strategies, (2024) Journal of Low Power Electronics and Applications, 14 (1), DOI:
10.3390/jlpea14010006
1
Zhang, J. et al., Intelligent Personalized Lighting Control System for Residents, (2023) Sustainability
(Switzerland), 15 (21). DOI: 10.3390/su152115355 1
Mukhopadhyay, S. et al., A Review and Analysis of IoT Enabled Smart Transportation Using Machine
Learning Techniques,(2024) International Journal of Transport Development and Integration, 8 (1), 61–77.
DOI: 10.18280/ijtdi.080106
1
Aizono, Y. et al., Building Automation with Vision Transformer Using Synthetic Indoor Images for Room
Light Control, KST 2024—16th Inter. Conference on Knowledge and Smart Technology, 40–44. DOI:
10.1109/KST61284.2024.10499683
1
Zhu, J. et al., Data-Driven End-to-End Lighting Automation Based on Human Residential Trajectory
Analysis, Inter. Conference on Smart Applications, Communications and Networking, SmartNets 2024 1
Sharma, V. et al., A novel study on IoT and Machine Learning-based transportation, (2024) Machine
Learning Techniques and Industry Applications, 1–28, DOI: 10.4018/979-8-3693-5271-7.ch001 1
Vale, Z. et al., An overview on smart buildings, (2022) Encyclopedia of Electrical and Electronic Power
Engineering: Volumes 1–3, 2, V2-431-V2-440. DOI: 10.1016/B978-0-12-821204-2.00066-0 1
Putrada, A.G. et al., An Evaluation of Activity Recognition with Hierarchical Hidden Markov Model and
Other Methods For Smart Lighting In Office Buildings, (2022) ICIC Express Letters, 16 (1), 91–100, DOI:
10.24507/icicel.16.01.91
1
Putrada, A.G. et al., NearCount for Model Compression on Edge Computing-Based Smart Lighting with
Product-of-Sum Function, Inter. Conference on Smart Computing, IoT and Machine Learning, SIML 2024,
13–18. DOI: 10.1109/SIML61815.2024.10578110
1
Putrada, A.G. et al., Machine Learning Methods in Smart Lighting Toward Achieving User Comfort: A
Survey, (2022) IEEE Access, 10, 45137–45178. DOI: 10.1109/ACCESS.2022.3169765 5
Exclusion Criteria
The 23 distinct papers satisfying the inclusion criteria have been narrowed down by
ignoring those belonging to the category in Table 16.

Future Internet 2025,17, 90 26 of 50
Table 16. The reason for exclusion of a study retrieved by the Scopus engine.
Exclusion Criteria The Way Excluded Studies Are Tagged in the Tables of This SLR
EC9. The study does not leverage ML. Out of scope.
All the authors read the full text of the 23 papers to make a decision concerning
their inclusion into the SLR based on the exclusion criteria. The admittance of a study
into the SLR was reached by consensus among the authors during meetings. In this step,
15 papers were removed, obtaining 8 publications (Figure 5). Table 17 references the
15 excluded papers.
Table 17. Papers excluded after full text reading.
Reference Aim Motivation for Exclusion
Wang, Y. and Durmus, D., Image Quality Metrics, Personality Traits,
and Subjective Evaluation of Indoor Environment Images, (2022)
Buildings, 12 (12), DOI: 10.3390/buildings12122086
Experimental investigation with humans of the relationship
between the perceived quality of indoor environments, personality,
and image quality metrics.
EC9
Putrada, A.G. et al., Homomorphic Encryption for Privacy
Preservation in Occupancy Sensor-Based Smart Lighting, Inter.
Conference on Data Science and Its Applications, 2024, 168–173.
DOI: 10.1109/ICoDSA62899.2024.10651987
A computational method to convert data about human presence in a
room into an integer to perform data homomorphic encryption. EC9
Putrada, A.G. et al., Synthetic Data with Nested Markov Chain for
CIMA-Based Smart Lighting Control Deployment Simulation, 11th
Inter. Conference on Information and Communication Technology,
ICoICT 2023, 2023, 148–153, DOI:
10.1109/ICoICT58202.2023.10262430
A deployment simulator for smart lighting control that uses
synthetic datasets. EC9
Cerpentier, J. et al., Adaptive museum lighting using CNN-based
image segmentation, Building and Environment, 242, 2023, DOI:
10.1016/j.buildenv.2023.110552
A lighting fixture comprising a LED, a sequence of lenses, and a
diffuser to obtain an adaptive LS. EC9
Cerpentier, J. et al., Controlling the target pattern of projected LED
arrays for smart lighting, Optics Express, 31 (22), 2023, 37316–37324.
DOI: 10.1364/OE.504077
A new method for calculating the optimal LED pixel addressing
scheme to match a target distribution. EC9
Cerpentier, J. et al., Smooth output from adaptive illumination
systems with pixelated LED arrays, (2023) Proc. of SPIE—The Inter.
Society for Optical Engineering, 12765. DOI: 10.1117/12.2688496
A computational, image-processing based method to achieve
adaptive LED array illumination with smooth output. EC9
Mohammadrezaei, E. et al., Systematic Review of Extended Reality
for Smart Built Environments Lighting Design Simulations, (2024)
IEEE Access, 12, 17058–17089, DOI: 10.1109/ACCESS.2024.3359167
A SLR aiming at exploring the use of extended reality for smart built
environments. EC9
Barandas, M. et al., Iterative wireless node localization based on
Bluetooth and visible light for smart lighting systems, (2024)
Wireless Telecommunications Symposium, DOI:
10.1109/WTS60164.2024.10536676
An approach that simplifies the configuration process of smart LSs. EC9
Parise, G. et al., A Comprehensive Exploration of Smart Lighting
Aspects: Area of Use, Methodologies and Purposes, 2023 IEEE
Industry Applications Society Annual Meeting, IAS 2023, DOI:
10.1109/IAS54024.2023.10406744
An overview of relevant topics about smart LSs as an urban service.
EC9
Daniel, W. et al., Integrated Smart Lighting Dashboard on the Office
Desk to Accommodate User Activity, 10th Inter. Conference on
Cyber and IT Service Management, CITSM 2022. DOI:
10.1109/CITSM56380.2022.9935875
A case study devoted to assessing whether the lighting intensity
level in offices meets the standard. EC9
Petkovic, M. et al., Smart Dimmable LED Lighting Systems, (2022)
Sensors, 22 (21), DOI: 10.3390/s22218523
An energy-efficient method for the design of the positioning of LED
lamps to illuminate an indoor floor plan. EC9
Widartha, V.P. et al., Advancing Smart Lighting: A Developmental
Approach to Energy Efficiency through Brightness Adjustment
Strategies, (2024) Journal of Low Power Electronics and
Applications, 14 (1), DOI: 10.3390/jlpea14010006
A prototype application that combines IoT sensors and daylight
data to raise energy efficiency in smart LSs. EC9
Putri, A.K. et al., The Smart Lighting System in the Coworking
Space’s Meeting Room, 4th Inter. Conference on Informatics,
Multimedia, Cyber and Information System, ICIMCIS 2022, 534–538.
DOI: 10.1109/ICIMCIS56303.2022.10017802
A case study that investigates the impact on workers of the light
intensity within a conference space. EC9
Putrada, A.G. et al., SLTAM: Remodelling Technology Acceptance
Model to Measure User Comfort in Smart Lighting with Exploratory
Factor Analysis, 3rd Inter. Conference on Intelligent Cybernetics
Technology and Applications, 2023, 414–419. DOI:
10.1109/ICICyTA60173.2023.10428882
A new model for measuring the user comfort and degree of
acceptance of the smart lighting technology. EC9
Chiradeja, P. and Yoomak, S., evelopment of public lighting system
with smart lighting control systems and internet of thing (IoT)
technologies for smart city, (2023) Energy Reports, 10, 3355–3372,
DOI: 10.1016/j.egyr.2023.10.027
Design and development of public LSs integrated with IoT
applications within smart cities. EC9

Future Internet 2025,17, 90 27 of 50
Study Quality Assessment
The quality of primary and secondary studies belonging to the pool of the 8 papers
captured through the forward snowballing was assessed by applying the quality criteria
previously introduced. Tables 18 and 19 show the result of this activity. None of the primary
studies were excluded, in contrast to the single secondary study because of its low score.
Table 20 references this latter work.
Table 18. Quality-assessment results for the 7 primary studies captured through forward snowballing.
Reference QC1 QC2 QC3 QC4 QC5 QC6 QC7 QC8 Score
[90] 1 1 0.5 1 1 0.5 1 1 7
[91] 1 1 0 1 1 0.5 1 1 6.5
[92] 1 1 0 1 1 0.5 0.5 1 6
[93] 1 1 0 1 1 0.5 1 0 5.5
[94] 1 1 0.5 0.5 1 0.5 1 0 5.5
[95] 1 1 0 1 1 0.5 0.5 0 5
[96] 1 1 0 0.5 1 0.5 1 0 5
Table 19. Quality-assessment results for the secondary study.
Ref.
QC1 QC2 QC3 QC4 QC5 QC6 QC7 QC8 QC9 QC10 QC11 QC12 QC13 QC14 QC15 Score
(Agramelal2023) 0.5 0 0 0 0 0 0 0 0 1 1 0.5 0.5 0 1 4.5
Table 20. Reference for the secondary study excluded at the quality-assessment stage.
(Agramelal2023) Agramelal, F. et al., Smart Street Light Control: A Review on Methods,
Innovations, and Extended Applications, (2023) Energies, 16 (21), DOI: 10.3390/en16217415
Data Extraction
The advice in ref. [
5
], p. 65, about this stage was implemented as follows: first, we
designed a data extraction form (Table 21) listing the metadata describing the information to
be extracted from the selected studies and the link between these metadata and the research
questions to be answered. Then, we added other tables comprising as many columns as
the number of items in Table 21. These latter tables were filled with the actual information
during the data-synthesis stage (Figure 5). To prevent the risk of bias, the data-extraction
form was defined when the study protocol was planned.
Table 21. Data items to be extracted from the selected studies.
Metadata Explanation RQs
Type of study Primary or secondary study RQ1
Source Article, conference paper,review, book chapter RQ1
Year Year of Publication RQ1
Length Number of pages of the paper RQ1
Scientific DB(s) Scientific databases queried by authors (applies to reviews) RQ1
Time interval The range of years covered by the study (applies to reviews) RQ1
References Number of references listed in the study RQ1
Citations Numberof citations got by the study RQ1
Number of RQs Number of RQs investigated (applies to reviews) RQ2
Surveyed papers Number of papers investigated in deep (applies to reviews) RQ2
Topic Problem solved by the study (applies to primary studies).
Topic addressed by the study (applies to secondary studies) RQ2
Paper’s keywords List of keywords proposed by authors of the study RQ2
The RQs The research questions addressed in the study (applies to reviews) RQ2
ML method(s) The ML method(s) adopted in the study (applies to primary studies) RQ3
Computing paradigm The computing paradigm proposed in the study (applies to primary studies) RQ4
Employed IoT devices Employed IoT devices in the implementation of EdgeML-based lighting systems (applies to primary studies) RQ4
Contribution Contribution of the study RQ3, RQ4

Future Internet 2025,17, 90 28 of 50
Data Synthesis
The aim of this step of the study protocol (Figure 5) is to synthesize the selected
28 papers
corresponding with each of the rows of Table 21, which allows one, in turn, to
derive an answer to the research questions (Section 3.1). As already said, each author
of this SLR read all the selected studies. A few meetings were sufficient to debate the
differences that occurred during the process. The synthesis is spread into nine tables (i.e.,
from Tables 22–30).
Table 22. Data synthesis from secondary studies.
Ref. Source Year Length Scientific DB(s)
[4] Journal 2022 42 Google Scholar
[89] Journal 2019 32 Not explicited
Table 23. Data synthesis from secondary studies (1).
Ref. Time Interval References Citations Number of RQs Surveyed Papers
[4] 1993 until 2021 434 47 3 Not declared
[89] It is not explicit 74 425 0 Not declared
Table 24. Data synthesis from secondary studies (2).
Ref. Topic Paper’s Keywords
[4]On the adoption of ML methods in
smart lighting to increase user comfort.
SL, SLR, ML, user comfort, activity
recognition
[89]A review on extant AIoT techniques
for smart transportation applications.
Big data, IoT, ML, Intelligent
transportation systems, Smart city,
Smart transportation
Table 25. Data synthesis from secondary studies (3).
Ref. The RQs
[4] RQ1. What are the topics discussed in studies about smart LSs?
RQ2. What smart lighting studies adopted ML?
RQ3. What smart lighting studies implemented ML to improve user comfort?
[89] They are not explicit
Table 26. Data synthesis from secondary studies (4).
Ref. Contribution
[4]
The review (a) maps ML methods for smart lighting research since 2014;
(b) groups the ML applications in smart lighting; (c) discusses the ML topics in
smart lighting able to boost people’s comfort; (d) mentions research gaps in the
application of ML in smart lighting related to boosting people’s comfort.
[89]
Authors structure the smart transportation sector into six categories (lights is one
of them). Then, they review the studies that addressed the six categories by using
IoT and/or ML techniques. Algorithm name, algorithm learning type,
and number of times each algorithm is used in the literature are also given.

Future Internet 2025,17, 90 29 of 50
Table 27. Data synthesis from primary studies.
Ref Source Year Length References Citations
[70] Journal 2020 16 159 33
[82] Journal 2021 15 108 87
[71] Journal 2020 26 54 22
[72] Journal 2023 11 42 8
[73] Journal 2024 10 33 3
[77] Journal 2022 10 31 10
[80] Conference 2020 16 38 6
[83] Conference 2019 5 12 6
[75] Conference 2023 6 35 3
[84] Conference 2024 6 8 0
[86] Conference 2024 5 11 0
[74] Conference 2024 6 31 0
[78] Conference 2024 6 14 0
[79] Conference 2023 8 12 0
[76] Conference 2019 5 20 4
[87] Conference 2024 4 10 0
[85] Conference 2022 4 11 0
[81] Conference 2022 8 25 0
[88] Conference 2020 13 16 0
[90] Journal 2022 22 68 13
[91] Journal 2023 11 74 6
[92] Journal 2023 12 30 3
[93] Conference 2024 18 39 0
[94] Conference 2022 6 25 5
[95] Conference 2023 6 23 1
[96] Conference 2022 5 20 3
Table 28. Data synthesis from primary studies (1).
Ref. Topic Paper’s Keywords
[70]Personalized energy-use behaviors in
commercial buildings.
Commercial buildings, IoT, Smartphone,
Wi-Fi network, energy-use behavior, DL
[82] Digitalization of highways.
Highway, DL, vulnerable road safety, smart
city, IoT, vision node, renewable energy
[71] Smart LSs for the learning context.
IoT; Smart lighting, Smart classroom,
Environmental data-processing framework,
Learning context, LED lighting control
[72] Horticultural LS.
Daylight harvesting, energy-efficiency,
horticultural lighting, IoT, ML, NNs,
Photosynthetically active
radiation measurement
[73] Human activity recognition. Smart lighting, Channel state information
(CSI), HAR, EL
[77]Activity recognition for smart lighting in
office buildings.
Smart lighting, activity recognition, IoT,
KNN, Naive Bayes, Hierarchical hidden
Markov model, PIR sensor
[80]Personalized service provision in a smart
LS.
Smart lighting, Edge-cloud collaborated
learning, Edge intelligence, DRL,
Personalized service provision
[83]Smart LS that integrates dimming level
with light intensity.
Smart lighting, light intensity, dimming
level, particle swarm optimization
[75] Digital Twins of Smart Campus. CPSs, Digital Twin, Urban IoT
[84] Indoor automatic dimming system.
Smart lighting, daylight response dimming
system, PSO
[86] Building automation. IoT, smart lighting, building automation,
DL
[74]Model Compression on Edge
Computing-Based Smart Lighting.
Smart lighting, EC, NearCount, model
compression, product-of-sum

Future Internet 2025,17, 90 30 of 50
Table 28. Cont.
Ref. Topic Paper’s Keywords
[78] Lighting automation. Smart home, ML, automation, transformer
[79] Voice-controlled LS.
Smart village micro-utilities, Embedded
ML, Edge Impulse, Voice-activated
LED lighting
[76]Detection of occupancy sensor signal
anomalies.
Occupancy sensors, Connected lighting,
RF classifier.
[87] Smart home control. Smart home, IoT, ML, SR,
Pervasive Computing
[85]Human activity recognition for smart
lighting control. IoT, AI, NN, MCU, Kalman filter
[81]Data-centric anomaly-based detection
system.
IoT, interactive ML, intrusion detection,
anomaly detection, IoT security, Poisoning
attack, virtual sensors
[88]
Monitoring system for domestic appliances.
ANN, automation, Cloud computing, facial
recognition, image processing, IoT, ML,
microcontroller, NN, sensors
[90]Automatic LS control based on human
presence.
Smart lighting, classification model,
movement data, user comfort
[91]Edge-computing architecture for smart
lighting control.
Edge-computing, smart lighting, model
compression, knowledge distillation, kNN
[92]Lighting control system tailored on home
residents.
Personalized lighting, intelligent lighting,
prediction control strategy,
back-propagation neural network
[93]Quantization method for Edge Computing
in smart LSs.
Smart lighting, Edge Computing,
NodeMCU, kNN, model compression,
quantization
[94] Comparison of RNN architectures. IoT, smart lighting, binary classification,
DT, RNNs, LSTM, time-series
[95]Occupants’ privacy-preserving in smart
LSs.
Smart lighting, general data protection
regulation, privacy, camera,
image perturbation
[96] Smart lighting control of offices. Office room, camera, smart lighting, object
localization, SSD MobileNet
Table 29. Data synthesis from primary studies (2).
Ref. ML Methods Computing Paradigm Employed IoT Devices
[70] DL Cloud-based
CO2 sensors, infrared
sensors, motion sensors,
sound sensors, and
temperature sensors are
available for
occupancy detection.
[82] DL EdgeML
Motion sensor,
light-dependent resistor
sensor, light dimmer, and a
long-range RF module.
[71] RL Cloud-based Ambient light sensors and
PIR sensors.
[72]Multi-linear regression, RF,
NN, and DT Cloud-based
Sony IMX219 image
sensors, multi-channel
spectral light sensors
AS-7341.
[73]
RF, Gradient Boosting,
and Extreme Gradient
Boosting Classifier.
Cloud-based None.
[77]KNN, Naive Bayes,
and HHMM Cloud-based Light sensor and
PIR sensor.

Future Internet 2025,17, 90 31 of 50
Table 29. Cont.
Ref. ML Methods Computing Paradigm Employed IoT Devices
[80] DRL EdgeML Light sensor, ultrasonic
sensor, and infrared sensor.
[83] PSO Cloud-based Ultrasonic sensor, smart
lighting device, lux sensor.
[75]
Supervised Learning
methods (SVM, LR,
and KNN), Ensemble
Learning (RF), and a DL
method (LSTM)
Cloud-based Light sensor.
[84] PSO Cloud-based Light sensor,
illuminance sensor
[86] DL Cloud-based None
[74] KNN, NearCount-PoS EdgeML Motion sensor, PIR sensor
[78] DL, DNN, RL Cloud-based
Human presence sensors,
environmental sensors,
brightness sensors.
[79] NN Classifier. EdgeML Microphone.
[76] RF classifier. Cloud-based Light sensor, occupancy
sensor (PIR motion sensor).
[87] SR Cloud-based None.
[85] ANNs Cloud-based
Light sensor, MCU, Monitor
(Node-RED), Arduino MCU
(MKR WiFi 1010)
[81] RF classifier. Cloud-based
Smart camera, climate
sensmitter, smart lighting
sensor, smart phone.
[88]ANN, LSTM, CNN,
and Naïve Bayes Cloud-based PIR, LDR.
[90]KNN, SVM, DT, NB,
and Ensemble Voting. Cloud-based PIR
[91] DNN, KNN EdgeML PIR
[92]
CNN, RF, DT, polynomial
regression, SVR, Ridge,
Lasso, Elastic net, KNN,
and BPNN.
Cloud-based None.
[93] KNN EdgeML PIR
[94] DT, RNNs, and LSTM. Cloud-based PIR
[95] CNN EdgeML Raspicam
[96] CNN Cloud-based Raspicam
Table 30. Data synthesis from primary studies (3).
Ref. Contribution
[70]
An IoT-based smartphone energy assistant (iSEA) tool. iSEA aims at promoting smart
energy-aware behaviours among the occupants of commercial buildings. iSEA
methodology and its IoT architecture are detailed.
[82]
Classification of highway digitalization into five components: smart highway LS is one of
them. An architecture for smart highway lighting, smart traffic, and emergency
management is also proposed.
[71]
Design and implementation of a smart LS, which dynamically controls the classroom
lighting in accordance with the learning context. The primary aim is to improve students’
performance.
[72]
Description of a neural-network learning control system composed of light sensors,
dimmable LED light fixtures, cameras, and a firmware devoted to crop monitoring and
performance evaluation.
[73]
A solution to HAR through passive sensing is proposed and evaluated. Leveraging
ensemble ML algorithms, the data extracted from the ESP32 microcontroller is used for
classifying different human activities.

Future Internet 2025,17, 90 32 of 50
Table 30. Cont.
Ref. Contribution
[77]An AIoT proposal that applies the HHMM for AR in smart lighting in office buildings.
The performance of the solution is proven to be superior to well-known ML methods.
[80]Design and implementation of a DRL model devoted to offering to users a personalized
illumination in order to enhance the quality of their experience with the LS.
[83]The study proposes a smart LS composed of two NodeMCU sensor nodes connected via
MQTT. The system uses PSO to integrate dimming level with light intensity.
[75]
Adoption of a Digital-Twin-based CPS to collect data generated on a campus to be used to
determine space occupancy based on the ambient light sensors.
[84]An AIoT-based system capable of achieving optimal illumination through sunlight,
reducing, at the same time, the lighting energy utilization.
[86]An ML-based building-automation system that uses images captured by a camera inside
the room.
[74]
Proposal of a method which performs sampling for the compression of the KNN model in
Edge Computing-based smart LSs.
[78]Proposal of a smart LS able to predict the occupant’s next position in order to adapt the
light accordingly.
[79]
Assessment of the influence of Raspberry Pi Pico W-based voice-activated LED lighting on
smart village micro-utilities.
[76]Adoption of frequency and temporal features to feed an RF classifier to detect occupancy
sensor anomalies in an apartment.
[87]
An SR-based method to learn the behaviour in smart homes. It leverages the data gathered
from sensors in the rooms and the actuator settings.
[85]An NN technique combined with the Kalman filter is adopted for controlling IoT devices
to augment the intelligence of the controller.
[81]
Proposal of a data-centric anomaly-based detection system. Experiments are carried out in
a campus that involves, among other things, a smart lighting component.
[88]An AIoT-based monitoring system for interacting through vocal commands with usual
domestic appliances (e.g., the luminaires).
[90]A novel Classification-Integrated Moving Average (CIMA) model that leverages human
presence in a room to control the LS to enhance occupant comfort.
[91]
A novel Edge Computing architecture for smart lighting control (EdgeSL) based on the
CIMA model and a new distillation algorithm of the KNN model as a compression model
to deploy CIMA in NodeMCU on EdgeSL.
[92] A method to control lights in rooms of a house leveraging the dwellers’ habits.
[93]A quantization method to compress the KNN model to run in NodeMCU being part of a
smart LS.
[94]An IoT implementation of a smart LS is proposed and used to compare the classification
performance of distinct RNN models.
[95]Implementation of a camera-based smart LS architecture. The solution ensures
anonymization through pixelation; moreover, it supports Edge Computing.
[96]Implementation of an AIoT prototype of a smart lighting control system in office rooms.
The prototype combines the lighting control with the SSD MobileNet object localization.
4. Results
The RQs of Section 3.1 are answered below, considering the findings arising from the
accurate analysis of the 26 selected primary studies, while the two secondary studies are
analyzed as related work.
4.1. (RQ1) What Is the Map of Published Primary and Secondary Studies About AIoT-Based LSs?
Tables 31 and 32 show, respectively, the distribution of the 28 selected papers over the
years and their type as well. 26 papers are primary studies; moreover, 2 journal papers out
of 11 belong to the review category. The first occurrence of a paper emphasizing the impact
of the adoption of the AIoT in the development of smart LSs is quite recent (2019). It is
worth noticing that despite there being large agreement in academia and industry about the
relevance of the AIoT in the solution of real-life problems in manifold application domains,
Table 31 proves that the benefits that the AIoT can bring to the lighting domain are almost

Future Internet 2025,17, 90 33 of 50
unexplored. This aspect is emphasized by the consideration that only 28 studies (25.0%)
turned out to be pertinent to the aim of the present SLR out of the 112 papers dealing with
LSs that leverage the IoT technology. Seventeen primary studies out of the 26 analyzed
(i.e., 65.4%) are conference papers. This number tells us that the penetration of the AIoT in
the lighting domain is still in an infancy stage. The survey reported in [
97
] and available
online in August 2024 confirms the previous conclusion of our SLR. In fact, despite the
survey talks about IoT-based smart public street LSs, the advent of AIoT-based LSs is just
mentioned as a future trend.
Table 31. Distribution of the 28 papers over the years.
Year Papers Year Papers
2024 7 2021 1
2023 6 2020 4
2022 7 2019 3
Table 32. Aggregation of the 28 selected papers by publication venue.
Study Category Type Number
Primary study Journal Paper 9
Conference Paper 17
Secondary study Journal Paper 2
Conference Paper 0
4.2. (RQ2) What Are the Main Topics Addressed for AIoT-Based LSs?
Table 33 recaps the content of the column “Topic” in Table 28, while Table 34 presents
the topics in Table 33 against the selected 26 primary studies. The “+” in box <
i
,
j
> of
Table 34 denotes that primary study “
i
” deals with topic “
j
”. As we can see, the matrix
is very sparse, which denotes that the analyzed studies mostly focus on unrelated topics.
The top three investigated topics are, in decreasing order, energy efficiency (13/26), hu-
man activity recognition (7/26), and personalized service provision (6/26), while privacy
protection occupies the bottom position. Below, we elaborate a bit more on these points.
The present SLR:
•
Confirms what is largely reported in previous studies, namely that energy efficiency
is the most investigated research topic in connection with smart LSs [4,98].
•
Shows that HAR is playing an increasing role in the domain of smart LSs be-
cause it is a preliminary step towards the automation of a variety of functions in
smart homes/buildings for lighting control based on human occupancy and/or
human actions.
•
Reveals that personalized service provision is in the third position. Personalized
services are services tailored to individual users’ interests and preferences. In the case
of smart LSs, personalized services mean offering light conditions that meet daily
human habits in indoor/outdoor environments in order to increase their comfort.
In ref. [
4
], Putrada et al. state that high user comfort is still in the infancy stage. In fact,
reaching such an ambitious goal requires further progress in the field of HAR in
combination with optimized ML models.
Table 35 maps the 17 topics in Table 33 to well-known application domains, while in
Table 36 the latter are listed according to their number of occurrences.
4.3. (RQ3) What Are the Key ML Methods Enabling the Implementation of AIoT-Based LSs?
Table 37 offers a global view about the ML methods used in the analyzed 26 primary
studies to solve the problem they report on. The table is structured in terms of ML categories

Future Internet 2025,17, 90 34 of 50
(first column), ML sub-categories (second column), specific method name (third column),
and primary studies which adopt the latter method (fourth column). As we can see,
the supervised ML methods are the most investigated (12 papers out of 26 adopt them),
followed by the DL methods (11 papers), then the Ensemble Learning methods (5 papers),
and Reinforcement Learning methods (4 papers). Passing, in sequence, from Tables 31–34
and then to Table 35, it is possible to join a specific ML method to the topic where it has
been shown to be effective and then to the application domain that would benefit from
its adoption.
Table 33. Topic coding.
ID Topic
1 Anomaly detection (intrusion detection).
2 Detection of occupancy sensor signal anomalies
3 Energy-use behaviour
4 Energy efficiency
5 Horticultural lighting system
6 Human activity recognition
7 Personalized service provision
8 Highway digitalization
9 Smart lighting system for the learning context
10 Smart village micro-utilities
11 Voice-controlled lighting system
12 Edge Computing
13 Image processing
14 Model compression
15 Facial recognition
16 Classification performance of ML models
17 Privacy protection
Table 34. Cross link between the 17 topics in Table 33 (the columns) and the primary studies
(the rows).
Ref.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
[
70
]
✓ ✓ ✓
[
82
]
✓ ✓
[
71
]
✓ ✓
[
72
]
✓ ✓
[
73
]
✓ ✓
[
77
]
✓
[
80
]
✓
[
83
]
✓
[
75
]
✓
[
84
]
✓
[
86
]
✓
[
74
]
✓ ✓
[
78
]
✓ ✓
[
79
]
✓ ✓ ✓
[
76
]
✓
[
87
]
✓ ✓
[
85
]
✓ ✓
[
81
]
✓
[
88
]
✓ ✓
[
90
]
✓ ✓ ✓
[
91
]
✓ ✓
[
92
]
✓ ✓ ✓
[
93
]
✓ ✓
[
95
]
✓
[
94
]
✓
[
96
]
✓ ✓

Future Internet 2025,17, 90 35 of 50
Table 35. Mapping of the topics investigated in the 26 primary studies into well-known
application domains.
ID Topic Application Domain
1 Anomaly detection (Intrusion detection). Smart home/building, public safety
2Detection of occupancy sensor signal
anomalies Smart home/building, smart healthcare
3 Energy-use behaviour Smart home/building, smart city
4 Energy efficiency
Smart home/building, smart city, smart
transportation, smart classroom, smart
healthcare, smart farm
5 Horticultural lighting system Smart farm
6 Human activity recognition Smart home/smart building, smart
healthcare, smart classroom
7 Personalized service provision
Smart home/building, smart city, smart
transportation, smart healthcare, smart
classroom
8 Highway digitalization Smart transportation, public safety
9Smart lighting system for the learning
context Smart classroom
10 Smart village micro-utilities Smart village
11 Voice-controlled lighting system Smart home/building, smart healthcare
12 Edge Computing Smart home/building, smart city, smart
healthcare, smart transportation
13 Image processing Smart home/building, smart healthcare,
smart city, smart transportation
14 Model compression Smart city, smart transportation, smart
home/building, smart healthcare
15 Facial recognition Smart home/building, smart healthcare,
smart classroom, public safety
16 Classification performance of ML models Smart home/building, smart city, smart
transportation, smart healthcare
17 Privacy protection Smart home/building, smart classroom,
smart healthcare, smart city
Table 36. Ranking of the application domains.
Application Domain Number of Occurrences
Smart home/building 13
Smart healthcare 11
Smart city 8
Smart transportation 7
Smart classroom 6
Public safety 3
Smart farm 2
Smart village 1
4.4. (RQ4) Do the Selected Studies Through RQ3 Implement the EdgeML Computing Paradigm?
Sipola et al. [
28
] concluded their review on the AIoT ecosystem by stating that up
to the end of 2021 the EdgeAI ecosystem was still in its infancy. Our SLR confirms that
at the end of 2024 the trend in the smart LS domain is pretty much the same; in fact, the
Cloud-only computing paradigm is largely the most adopted one (Table 38). In numbers,
we found that just 7 works (out of 26) included in the SLR propose AIoT-based smart
LSs leveraging an actual cooperation between the Cloud and the edge, so they promote
EdgeML. As further confirmation of previous statements, it is worth noting that 5 studies
out of 7 are conference papers that usually report on ongoing projects.

Future Internet 2025,17, 90 36 of 50
Table 37. A summary table linking primary studies and ML method.
ML Category ML Sub-Category Method Name Primary Studies
Supervised Learning Classification
DT [72,90]
NN [72,79,85]
ANN [88]
KNN [74,75,77,81,90,91,93]
SVM [75,90]
LR [75]
NB [77,88,90]
Regression LiR [72]
SR [87]
Ensemble Learning Bagging RF [72,73,75,76]
Boosting Gradient boosting [73,92]
XGBoost [73]
Deep Learning
Not specified [70,82]
DNN [86,91]
CNN [88,92,95,96]
Recurrent NN LSTM [27,75,94]
Reinforcement Learning Evolutionary computing
PSO [83,84]
RF [71]
DRL [80]
Table 38. Aggregation of the selected 26 primary studies by computing paradigm.
Edge Cloud-Based (EdgeML) Cloud-Based
[80] [86] [77]
[82] [88] [90]
[79] [81] [94]
[95] [96] [70]
[91] [84] [71]
[74] [75] [76]
[93] [85] [87]
[83] [78]
[72] [92]
[73]
Hereinafter, we recap how the EdgeML solution is implemented in these 7 primary
studies. The synopsis refines the information collected in Tables 28–30.
In [
80
], the authors adopt a Cloud-aided edge RL framework to support the download-
ing of the global consensus model from the Cloud and integrate it into the edge learning
process. The efficiency and effect of applying the downloaded pre-trained model are
boosted by applying an input-expansion strategy followed by an output-correction strategy.
The approach’s effectiveness and performance are shown through experiments on data
generated by the open software DAILux.
Singh et al. [
82
] suggest the adoption of the MEC paradigm to implement the smart
LS component of a digital highway. The IoT architecture is structured as a network of
wireless highway light controllers. Each controller comprises a sensor node and an edge
device-based vision node, both embedding a computing unit. The sensor nodes integrate
the sensors (e.g., LDR sensors, light dimmers, motion sensors). The edge device-based
vision node is responsible for processing the data coming from the sensors and the internal
camera by making use of DL algorithms. Then, accordingly, it sends information to the
computing unit of the sensor node that, in turn, uses it to adapt the light intensity on the
highway lane.

Future Internet 2025,17, 90 37 of 50
In ref. [
79
], Narasimharao et al. propose a low-cost solution to voice-activated LED
lighting. As hardware, Raspberry Pi Pico W, a microphone, an LED light bar, the power
supply, and the breadboard are used. Raspberry Pi Pico W captures audio inputs from
the microphone, processes the audio signals, and controls the LED lights. The LED light
bar provides visual feedback according to voice commands. Data collection, data pre-
processing, and model training are carried out through the Edge Impulse EON Tuner
platform (https://docs.edgeimpulse.com/docs/edge-impulse-studio/eon-tuner (accessed
on 10 Novemeber 2024)), which allows model design and optimization on edge devices.
Edge Impulse uses an NN classifier to perform voice command recognition and control
LED lights. The ML model is trained on a large dataset of voice commands. The NN
classifier is responsible for classifying the audios into different classes (e.g., start, stop,
noise, and other).
Prabowo et al. [
95
] propose a way to guarantee the preservation of occupants’ privacy
in smart LSs. Anonymization is brought through image pixelation. The images are elab-
orated by means of the CNN model. The adopted solution implements a camera-based
smart LS within the Edge Computing architecture. The edge environment is where the
images received by Raspicam are processed. Raspberry Pi 3B+ is used as an edge processor.
In ref. [
91
], the authors introduce an Edge Computing architecture for smart lighting
control (EdgeSL) based on the CIMA model presented in [
90
]. Moreover, they describe
a new distillation algorithm of the KNN model (DistilKNN) as a compression method
to deploy CIMA in a NodeMCU part of the EdgeSL architecture. Experiments running
DistilKNN on NodeMCU returned better performance than those obtained using Cloud
computing:
(a) best
accuracy compared to methods using quantization and pruning, and (b)
better average processing time.
The work by Putrada et al. [
74
] introduces a variant of the NearCount sampling method
(NearCount-PoS) as a compression strategy to reduce the number of samples needed to feed
the ML model to be run in an Edge Computing-based smart LS. The MQTT publish/subscribe
protocol is used to connect IoT edge devices over the Internet and with the Cloud. NearCount-
PoS returns better prediction accuracy than KNN at sample numbers above 9000.
Putrada et al. [
93
] adopt the quantization method to compress the well-known KNN
model to be able to run it in the NodeMCU as part of a smart LS. This is the first study that
investigated the performance of the quantization compression strategy on the KNN model
deployed on NodeMCU as part of an EdgeML-based smart LS.
Table 39 describes the previous seven studies through the tuple:
<paper’s reference, (used) ML model, (used) IoT devices, (used) hardware at
the edge>.
Table 39. A short description of the seven primary studies that promote EdgeML.
Paper ML Model IoT Devices Hardware at the Edge
[80] DRL
Light sensor, ultrasonic
sensor, and infrared
sensor
Not specified
[82] DRL LDR sensors and light
dimmers Not specified
[79] NN LED light bar and a
microphone Raspberry Pi Pico W
[91] CIMA + DistilKNN PIR sensor NodeMCU
[74] KNN Motion sensor and PIR
sensor NodeMCU
[93] KNN PIR sensor NodeMCU
[95] CNN Raspicam Raspberry Pi 3B+

Future Internet 2025,17, 90 38 of 50
5. Discussion
Figure 6depicts open research challenges in connection with EdgeML-based smart
LSs. They are discussed after a brief recap of the results presented in the previous section.
Figure 6. Open research challenges in EdgeML smart LSs.
The output of the research on Scopus, starting from the search string, was very
promising. Indeed, the Scopus engine returned 89 studies. Unfortunately, 68 of them
did not pass the filter criteria (i.e., IC/EC and quality assessment—Section 3). From this
stage of the present SLR, it emerges that only two secondary studies have appeared (up
to the end of October 2024) on the topics of this review. This finding is the consequence
of the very low number of published primary studies that we have found (i.e., 19). This
package of studies was increased with 7 more primary studies selected by carrying out the
forward snowballing.
In total, 7 primary studies propose the adoption of the EdgeML computing paradigm
to implement a new generation of smart LSs. This scarcity of research on the adoption of
EdgeML in the lighting domain confirms the conclusion in [
99
], a book chapter that focuses on
the state-of-the-art of IoT and ML applications in the smart city transportation domain. In the
work, city transportation is the umbrella that comprises street lighting. Sharma et al. state that
from the extant body of research it can be seen that ML is inadequately represented on smart
LSs, which means that nothing has changed since this deficiency was first reported in [89].
The answer to RQ2 reveals that energy efficiency is the most investigated research
topic. This finding confirms previous studies such as, for instance, [
4
,
100
]. As already
remarked in the introduction, the motivation that pushes the research on smart LSs in such
a direction is the huge energy consumption caused by them.
5.1. Predictive Maintenance
None of the 26 selected primary studies talks about predictive maintenance. Having
automatic support in carrying out the maintenance of LSs, especially those of public
interest such as hospitals, schools, universities, malls, city downtown, and so on, is an
emerging concern by the installers/maintainers of these systems since the effectiveness of
this activity has a direct repercussion on the quality of service. A low-quality value can
lead to the interruption of assistance contracts with serious economic and image damage
for the manager. This deficiency has been highlighted in previous sector studies, as briefly
reported on below.

Future Internet 2025,17, 90 39 of 50
Alahi et al. [
101
] highlight the primary role of AIoT in implementing predictive
maintenance of LSs in the context of future city management.
Galatanu et al. [
102
] state that public LSs are maintained via periodical inspections
in situ of luminaires. During the inspection, the lamps that are likely to stop running are
replaced, based on the only one parameter that is monitored, namely their running time.
To overcome such an unsatisfactory practice, they propose the adoption of an imaging
method suitable for analyzing the degradation in time of the LS parameters, on the basis of
which predictive maintenance activities can be scheduled.
Singh et al. [
82
] cite six studies about highway LSs published between 2016 and 2020.
None of them talk about the “fault detection and diagnosis system” that, in contrast, authors
consider an imperative component in future applications for highways to prevent outages
of the LS due to faults.
The review reported in [
4
] does not even mention the words “predictive maintenance”.
However, authors talk about “predictive control” as the precondition to predict the proper
maintenance time for street luminaires.
In 2024, Pasolini et al. [
2
] describe the general architecture and functionality of city
lighting infrastructure. The latter leverages environmental sensors and a wireless communi-
cation network that connects a thousand smart, remotely controlled streetlights. In addition,
the research discusses two lighting infrastructures deployed, respectively, in Italy and Viet-
nam. In the current version of the system, functional parameters (i.e., lamp temperature,
electrical parameters, and energy consumption) of each luminaire are remotely controlled
and stored. As future work, authors plan to expand the functionality of the infrastructure
by implementing ML algorithms that learn from this huge data asset in order to offer an
accurate predictive maintenance service to the managers of the infrastructure. In summary,
predictive maintenance is a software functionality that should be featured by AIoT-based
LSs in any application domain among those listed in Table 36.
5.2. Protection of Personal Data
It has been remarked that because future indoor/outdoor LSs become smarter, they
need to collect more data about people and their daily activities [
58
]. This point increases
the data privacy concern which at present is a big issue for all IoT systems that handle
personal information. In May 2018, the EU put into effect the General Data Protection
Regulation. GDPR is a privacy and security law that imposes obligations and fixes penalties
for organizations that collect data about people in the EU (a guide to the GDPR may be
found here: https://gdpr.eu/ (accessed on 22 November 2024)). Because of the advent
of the GDPR and subsequent privacy regulations in other countries all over the world,
the adherence to the following principles are considered a precondition for actual protection
of personal data: data minimization; storage limitation; purpose limitation; accuracy;
lawfulness, fairness, and transparency; integrity and confidentiality; and accountability
(https://gdpr.eu/what-is-gdpr/ (accessed on 22 November 2024)).
The present SLR on AIoT-based smart LSs points out that, in the set of the selected
26 primary studies, there is just one conference paper (namely, [
95
]) about the privacy
preservation of personal data (Table 34). In the study, the authors propose a solution that
addresses the data-minimization principle mentioned above. According to the GDPR, this
principle states that only as much data as is really necessary for the purposes specified are
involved in the processing.
5.3. Need of High-Quality Real-World Datasets
The answer to RQ3 reveals that the categories of ML algorithms mostly investigated
in the domain of smart LSs are, in decreasing order of relevance, supervised ML methods,

Future Internet 2025,17, 90 40 of 50
DL methods, Ensemble Learning methods and Reinforcement Learning methods. It is
worth repeating that the number of studies that have explored the adoption of ML methods
in the LS domain is low. An objective cause behind this delay emerges from the very
recent study conducted by Shao et al. [
103
]. In the study, the authors explore strategies
for prototyping an IoT- and BIM-based efficient LS of a library, as an alternative to that
based on the adoption of ML algorithms. The motivation behind this choice lies in the lack
of high-quality real-world data and the lack of procedural understanding which prevent
ML models from producing trustworthy findings. Concerns about the quality and size of
datasets used to train ML methods are reported in most studies that adopt the AI technology
to make predictions [81,104].
5.4. Security
AIoT-based LSs are vulnerable to cyber-attacks than can cause a system’s outage.
In ref. [
4
], the authors mention the issue as a new recent research topic in the domain.
Three years later, the situation is almost the same. Besides the manifold attacks already
studied in connection with IoT networks, EdgeML-based smart LSs are exposed to further
potential threats strictly related to the credibility of ML predictions. Poisoning integrity
attacks against ML systems is one of them [
105
]. They are the potential adversarial control
of the training dataset in order to modifying the model predictions and, hence, adversely
affect the operation of the LS. Through the present SLR, we found just one conference paper
touching on the security of smart LSs [
81
]. The study introduces a data-centric anomaly-
based detection system for the identification of poisoning integrity attacks. Ref. [
104
] is an
authoritative study that deeply analyzes the security vulnerabilities that may originate from
the absence of trustworthy human supervision during the collection process of datasets to
be used for the training of ML classifiers. In simple words, it may happen that the training
datasets are manipulated to control and degrade the downstream behaviours of learned
models. This study classifies a large range of dataset vulnerabilities, then it focuses on
ways to defend against the threats; eventually, it lists open problems in the area. The latter
represents a research opportunity for scholars investigating EdgeML-based smart LSs.
5.5. Leveraging TinyML
Among the 26 primary studies included in the present SLR, only [
79
] explored the
feasibility and effectiveness of adopting a TinyML-based voice-activated LED LS. It has been
remarked in Section 2.7 that there are huge expectations on the benefits that the TinyML
paradigm can bring to EdgeML-based systems. So, much more attention should be devoted
to exploring the impact of TinyML on the smart LS domain. As pointed out in [
32
], there
are a lot of open challenges in TinyML research. Key points which require prompt attention
include the capacity of TinyML to implement DNNs at the edge and the assessment of
the power consumption versus performance of the ML model. A complementary research
opportunity consists in evaluating EdgeML-based LSs with respect to the four metrics
identified as relevant for evaluating TinyML systems, namely accuracy, power consumption,
latency, and memory requirements [32].
5.6. Compression of ML Models
The higher the degree of compression of the ML models, the more possible it
will be to perform the inference phase on the edge. Recent research is moving in this
direction [74,90,91],
but further efforts are needed. Two popular DL methods for compres-
sion in Edge Computing are CNN and DNN. Model compression is the other face of the
coin of the memory requirements metric mentioned in [
32
] to compare the performance of
EdgeML-based LSs against the classical Cloud-based architecture. Ref. [
91
] selected in the
present SLR explores such a research direction.

Future Internet 2025,17, 90 41 of 50
5.7. A Federated Learning Architecture
Figure 7, inspired by one in ref. [
10
], shows the architecture that, considering what
emerged from the present SLR, we envision as suitable for the deployment of EdgeML-
based smart LSs. Before going into detail about the previous claim, hereafter a brief
description of Figure 7follows, moving from bottom to top.
Figure 7. A three-layer architecture of an EdgeML-based LS.
The first layer concerns smart IoT devices in the sense given in [
106
], that is, devices
made up of one or more sensors directly connected, generally in Bluetooth or wireless
mode, to an MCU, the latter equipped with memory and computing capacity, albeit to a
limited extent. This layer can host an arbitrary number of smart devices among those listed
in Table 1. The number of devices and their type depend on the addressed use case.
The intermediate level in Figure 7is composed of a certain number of nodes, each
equipped with memory and computation capacity that is significantly greater than the
level offered by the MUCs. Solutions based on the use of Raspberry Pi family hardware
are widely reported in the literature [
10
]. These nodes can take charge of the training
phase of any ML algorithm among those listed in Table 37, in addition to carrying out the
task of sending the code of the corresponding inference algorithm to the smart devices
connected to them. Figure 7also highlights that “horizontal” collaborations are possible
between the nodes of the intermediate level, useful, for example, to build the training
model of the chosen ML algorithm, also drawing on data present on multiple nodes of the
intermediate level.
The Cloud layer is the network node where in a sense unlimited computing and data
storage resources are concentrated; therefore, it is the natural location for the permanent
storage of “valuable” data and their processing. Figure 7ignores the application layer
mentioned in Section 2.2 because it is not relevant at this stage of the study.
An important feature of the architecture in Figure 7is that it allows one to experiment
with the use of federated ML algorithms. This implies the collaboration of the Cloud node
with the nodes of the intermediate level, thus reducing the resources required at the edge
layer, keeping at the same time the sensed data decentralized and hence secure. We envision
that the adoption of this kind of architecture might be appropriate for the deployment
of EdgeML-based smart LSs because of the findings in ref. [
27
]. In that study, Zhou et al.
present an Edge Cloud architecture that leverages FL among layers by referring to the
transportation domain as the use case. The authors proved that the adopted architecture

Future Internet 2025,17, 90 42 of 50
delivers high-quality service to AIoT devices, while protecting the user privacy, so it
overcomes the limitations of traditional Cloud computing ones.
At the beginning of 2022, Putrada et al. [
4
] stated that the smart lighting technology
was still in the proof-of-concept stage. The investigation of current trends coming from the
manufacturers’ side is outside the scope of the present SLR.
Table 40 recaps the open issues discussed in this section. Each of them represents a
research opportunity both for academia and industry.
Table 40. Open issues related to future EdgeML smart LSs.
Open Issue
Lack of methods for the automatic support of predictive maintenance polices of these systems
Protection of personal data of users of these systems
Lack of high-quality open-real-world datasets
Lack of protection methods against attacks on the integrity of training datasets
Leveraging TinyML in the development of these systems
Reducing the size of ML models without reducing their accuracy
Exploring the usage of federated learning architectures
Investigating the current trends coming from the manufacturers’ side
6. Related Works
The two secondary studies selected through the present SLR constitute the related
work. They are recapped below. Ref. [
89
] is a 23-page-length review which references
74 papers. Neither the period covered by the study, nor the scientific databases queried
by authors are declared. It has accumulated 425 citations, up to the end of October 2024.
This study summarizes the extant AIoT techniques for smart transportation, where LSs are
one of the six components. Specifically, the study talks about smart streetlights. The pillar
topics within the frame of the study are road lighting and energy efficiency. Ref. [
4
] is a
remarkable SLR that has accumulated 47 citations up to October 2024. The 42-page-length
study covers the time interval from 1993 up to 2021 and references 434 papers taken from
Google Scholar. The authors declare that 332 studies (out of 434) are specific to answering
their RQs. However, it is worth noticing that the actual number of studies where there is a
confluence of IoT and AI/ML in the advancement of LSs goes down to 196, since the first
studies belonging to this category date back to 2019 [76,83].
The investigation is carried out by answering three RQs (Table 23) whose aim is to
learn, first, the topics discussed in studies about smart LSs and then the ML algorithms
implemented to improve user comfort. Based on the results of text mining on the keywords
of the retrieved studies, [
4
] classifies them into four main issues (user comfort, light control,
lighting network, and energy-consumption reduction), four main implementation domains
(smart city, smart home, smart building/office, and smart street lighting), and six main
technologies (sensors, LED lights, IoT, intelligence, energy harvesting, and renewable
energy). Table 41 links the topics in Table 33 to the previous four main issues. As we can
see, 13 topics (out of 17) relate to light control, while 12 (out of 17) relate to user comfort.
The present SLR:
•
Investigates the state-of-the-art in the adoption of AIoT technology in the development
of the next generation of smart LSs leveraging the EdgeML paradigm. The latter term
is not mentioned in [
4
], a topic to which the same authors are currently devoting a lot
of effort [74,91,93,95].
•
Is orthogonal to [
4
], in the sense that the findings from our study cross different applica-
tion domains (smart home/building, smart healthcare, smart city, smart transportation,
smart classroom, public safety, smart farm, smart village—Table 36), while [
4
] focuses
on the use of the AIoT to control LSs to increase people’s comfort.

Future Internet 2025,17, 90 43 of 50
Table 41. Topics in Table 33 against the issues in [4].
ID Topic Main Issue, [4]
1 Anomaly detection (intrusion detection) User comfort
2Detection of occupancy sensor signal
anomalies User comfort, light control
3 Energy-use behaviour User comfort, light control, energy
consumption reduction
4 Energy efficiency Energy consumption reduction
5 Horticultural lighting system Light control
6 Human activity recognition User comfort, light control
7 Personalized service provision User comfort, light control
8 Highway digitalization User comfort, light control
9Smart lighting system for the learning
context User comfort, light control
10 Smart village micro-utilities User comfort, light control
11 Voice-controlled lighting system User comfort, light control
12 Edge Computing Light control, lighting network
13 Image processing User comfort, light control, energy
consumption reduction
14 Model compression Light control
15 Facial recognition User comfort, light control, energy
consumption reduction
16 Classification performance of ML models Light control
17 Privacy protection User comfort
7. Threats to Validity
Construct validity concerns the risk of losing relevant studies during the searching
stage. Consequently, some concepts, definitions, case studies, and so on may not have
appeared in the study results. This threat was mitigated by carrying out the selection stage
(Figure 5) in independent groups, and by performing regular internal meetings to reach
a consensus on which studies would be included. Moreover, the “IoT” word was added
to the search string, so that the Scopus engine could retrieve those papers that in the title,
abstract, and keywords only used “IoT” instead of “Internet of Things”. Refs. [
73
,
79
] are
two papers that would not have been identified by the Scopus engine that, instead, are
pertinent to the SLR reported in this paper. Lastly, the forward snowballing activity was
carried out by making recourse to the Scopus engine. Through this further effort, seven
more papers were added to the final pool of studies to be analyzed.
Internal validity is the extent to which the design and conduct of the study are likely
to avoid systematic errors [5]. The protocol of Figure 5guided us in avoiding this threat.
External validity refers to the degree of applicability, outside of this SLR, of the
observed results [
5
]. We can say that the more the selected publications were published in
relevant peer-reviewed venues, the more our findings are applicable. To prevent this threat,
we queried the Scopus scientific repository.
Conclusion validity refers to threats that can impact the reliability of the conclusions.
In this regard, a potential threat might be caused by a wrong interpretation of the results of
the collected papers. To limit this threat, all the articles returned by the Scopus engine were
reviewed by all the authors and no decision was taken individually, but collectively.
8. Conclusions
The authors carried out an SLR to answer the following RQs:
(RQ1) What is the map of published primary and secondary studies about AIoT-based LSs?
(RQ2) What are the main topics addressed for AIoT-based LSs?
(RQ3) What are the key ML methods enabling the implementation of AIoT-based LSs?
(RQ4) Do the selected studies through RQ3 implement the EdgeML computing paradigm?

Future Internet 2025,17, 90 44 of 50
Overall, 28 publications were selected for analysis in depth. Twenty-one of them came
from the Scopus database, while 7 were found by carrying out forward snowballing.
From the analysis, we learnt the following:
• The penetration of the AIoT in the lighting domain is still in an infancy stage.
•
The top three investigated topics are, in decreasing order, energy efficiency, human
activity recognition, and personalized service provision.
•
The top three application domains are smart home/building, smart healthcare,
and smart city.
•
The Supervised ML methods are the most investigated, followed by the DL methods,
then the Ensemble Learning methods.
•
The Cloud-only computing paradigm is largely the most adopted one. We found that
just 7 primary studies included in the SLR propose AIoT-based smart LSs, leveraging
an actual cooperation between the Cloud and the edge, so they promote EdgeML.
About the future research necessary to promote the EdgeML paradigm in the LS
domain, it should focus, primarily, on methods suitable for (a) supporting predictive
maintenance, (b) protecting the personal data of the users of these systems, (c) generating
high-quality and trustable open-real-world datasets necessary to training ML models,
(d) raising the level of protection of these systems against cyber-attacks, (e) promoting
the adoption of TinyML-based architectures, (f) reducing the size of ML models without
reducing their accuracy, (g) being deployed on a federated learning architecture.
Author Contributions: Conceptualization, P.D.F. and G.P.; methodology, P.D.F. and G.P.; validation,
A.P., R.P. and J.C.; formal analysis, P.D.F.; investigation, P.D.F., G.P., A.P., R.P. and J.C.; data curation,
F.P.; writing—original draft preparation, P.D.F. and F.P.; visualization, R.P. and F.P.; supervision, F.P.
and G.P.; funding acquisition, G.P. All authors have read and agreed to the published version of
the manuscript.
Funding: This research received no external funding.
Data Availability Statement: Not applicable.
Conflicts of Interest: Gaetanino Paolone, Jacopo Camplone and Andrea Piazza were employed by the
company B2B S.r.l.; Romolo Paesani and Francesco Pilotti were employed by the company Gruppo SI
S.c.a.r.l. The authors declare no conflicts of interest.
Abbreviations
The following abbreviations are used in this manuscript:
Acronym Definition
AI Artificial Intelligence
AIoT Artificial Intelligence of Things
ANN(s) Artificial Neural Network(s)
AR Activity Recognition
CNN Convolution NN
CPS(s) Cyber Physical Systems
CV Computer Vision
DNN Deep Neural Network
DRL Deep Reinforcement Learning
DT Decision Tree
EC Edge Computing
EL Ensemble Learning
FL Federated Learning
HAR Human Activity Recognition
HHMM Hierarchical Hidden Markov Model

Future Internet 2025,17, 90 45 of 50
IIoT Industrial Internet of Things
IoT Internet of Things
KNN K-Nearest Neighbor
LED Light Emitting Diode
LDR Light Dependent Resistor
LiR Liner Regression
LoR Logistic Regression
LS(s) Lighting System(s)
LSTM Long Short-Term Memory
MCU Microcontroller Unit
MEC Mobile Edge Computing
ML Machine Learning
NB Naïve Bayes
NN(s) Neural Network(s)
PIR Passive Infrared
PIR Passive Infrared
PSO Particle swarm optimization
RF Random Forest
RL Reinforcement Learning
RNN(s) Recurrent Neural Network(s)
SML Supervised ML
SpR Speech Recognition
SME(s) Small and Medium-sized Enterprise(s)
SVM Support Vector Machine
SVR Support Vector Regression
SyR Symbolic Regression
VR Voice Recognition
WSN Wireless Sensor Network
References
1.
de Kort, Y.; Michalos, A.C. Light and Quality of Life. In Encyclopedia of Quality of Life and Well-Being Research; Springer:
Berlin/Heidelberg, Germany, 2014; pp. 3615–3620.
2.
Pasolini, G.; Toppan, P.; Toppan, A.; Bandiera, R.; Mirabella, M.; Zabini, F.; Bonata, D.; Andrisano, O. Comprehensive Assessment
of Context-Adaptive Street Lighting: Technical Aspects, Economic Insights, and Measurements from Large-Scale, Long-Term
Implementations. Sensors 2024,24, 5942. [CrossRef] [PubMed]
3.
Soheilian, M.; Fischl, G.; Aries, M. Smart Lighting Application for Energy Saving and User Well-Being in the Residential,
Environment. Sustainability 2021,13, 6198. [CrossRef]
4.
Putrada, A.G.; Abdurohman, M.; Perdana, D.; Nuha, H.H. Machine Learning Methods in Smart Lighting Toward Achieving User
Comfort: A Survey. IEEE Access 2022,10, 45137–45178. [CrossRef]
5.
Kitchenham, B.; Charters, S. Guidelines for Performing Systematic Literature Reviews in Software Engineering; Technical Report
EBSE-2007-01; EBSE: Delhi, India, 2007. Available online: https://www.researchgate.net/publication/302924724_Guidelines_
for_performing_Systematic_Literature_Reviews_in_Software_Engineering (accessed on 25 August 2024).
6.
Kuhrmann, M.; Fernández, D.; Daneva, M. On the pragmatic design of literature studies in software engineering: An experience-
based guideline. Empir. Softw. Eng. 2017,22, 2852–2891. [CrossRef]
7.
Fabiano, G.; Marcellusi, A.; Favato, G. Channels and processes of knowledge transfer: How does knowledge move between
university and industry? Sci. Public Policy 2020,47, 256–270. [CrossRef]
8.
Cazón-Martin, A. Transferring Knowledge from Academia to the Companies: A New Method. In Materialising the Future: Research
for Development; Ferraro, V., Kääriäinen, P., Morer-Camo, P., Pasold, A., Eds.; Springer: Cham, Switzerland, 2023; pp. 65–80.
9.
dos Santos, V.; Iwazaki, A.Y.; Felizardo, K.R.; de Souza, E.F.; Nakagawa, E.Y. Sustainable systematic literature reviews. Inf. Softw.
Technol. 2024,176, 107551. [CrossRef]
10.
Merenda, M.; Porcaro, C.; Ier, D. Edge Machine Learning for AI-Enabled IoT Devices: A Review. Sensors 2020,20, 2533. [CrossRef]
11.
Debauche, O.; Mahmoudi, S.; Mahmoudi, S.A.; Manneback, P.; Lebeau, F. A new edge architecture for AI-IoT services deployment.
Procedia Comput. Sci. 2020,175, 10–19. [CrossRef]

Future Internet 2025,17, 90 46 of 50
12.
Sung, T.W.; Tsai, P.W.; Gaber, T.; Lee, C.Y. Artificial intelligence of things (AIoT) technologies and applications. Wirel. Commun.
Mob. Comput. 2021,2021, 9781271. [CrossRef]
13.
Achouch, M.; Dimitrova, M.; Ziane, K.; Karganroudi, S.S.; Dhouib, R.; Ibrahim, H.; Adda, M. On predictive maintenance in
industry 4.0: Overview, models, and challenges. Appl. Sci. 2022,12, 8081. [CrossRef]
14.
Vemuri, V.P.; Priya; Naik, V.R.; Chaudhary, V.; RameshBabu, K.; Mengstie, M. Analyzing the use of internet of things (IoT) in
artificial intelligence and its impact on business environment. Mater. Today Proc. 2022,51, 2194–2197. [CrossRef]
15.
Syed, A.S.; Sierra-Sosa, D.; Kumar, A.; Elmaghraby, A. IoT in smart cities: A survey of technologies, practices and challenges.
Smart Cities 2021,4, 429–475. [CrossRef]
16.
Herath, H.M.; Mittal, M. Adoption of artificial intelligence in smart cities: A comprehensive review. Int. J. Inf. Manag. Data
Insights 2022,2, 100076. [CrossRef]
17.
Firouzi, F.; Farahani, B.; Barzegari, M.; Daneshmand, M. AI-driven data monetization: The other face of data in IoT-based smart
and connected health. IEEE Internet Things J. 2022,9, 5581–5599. [CrossRef]
18.
Sivalingam, S.M.; Thisin, S. A new framework to enhance healthcare monitoring using patient-centric predictive analysis. Int. J.
Electr. Comput. Eng. (IJECE) 2024,14, 3295–3302. [CrossRef]
19.
Lombardi, M.; Pascale, F.; Santaniello, D. Internet of Things: A General Overview between Architectures, Protocols and
Applications. Information 2021,12, 87. [CrossRef]
20.
Rong, G.; Xu, Y.; Tong, X.; Fan, H. An edge-cloud collaborative computing platform for building AIoT applications efficiently. J.
Cloud Comput. Adv. Syst. Appl. 2021,10, 36. [CrossRef]
21. Singh, R.; Gill, S.S. Edge AI: A survey. Internet Things Cyber-Phys. Syst. 2023,3, 71–92. [CrossRef]
22.
Gkonis, P.; Giannopoulos, A.; Trakadas, P.; Masip-Bruin, X.; D’Andria, F. A Survey on IoT-Edge-Cloud Continuum Systems:
Status, Challenges, Use Cases, and Open Issues. Future Internet 2023,15, 383. [CrossRef]
23.
Shahraki, A.; Ohlenforst, T.; Kreyß, F. When Machine Learning meets Network Management and Orchestration in Edge-based
networking paradigms. J. Netw. Comput. Appl. 2023,212, 103558. [CrossRef]
24.
Khoei, T.T.; Kaabouch, N. Machine Learning: Models, Challenges, and Research Directions. Future Internet 2023,15, 332.
[CrossRef]
25.
Sakr, F.; Bellotti, F.; Berta, R.; De Gloria, A. Machine Learning on mainstream microcontrollers. Sensors 2020,20, 2638. [CrossRef]
26.
Zhao, Z.; Wang, K.; Ling, N.; Xing, G. EdgeML: An AutoML Framework for Real-Time Deep Learning on the Edge. In Proceedings
of the International Conference on Internet of Things Design and Implementation, Charlottesville, VA, USA, 18–21 May 2021;
pp. 133–144. [CrossRef]
27.
Zhou, J.; Pal, S.; Dong, C.; Wang, K. Enhancing quality of service through federated learning in edge-cloud architecture. Ad Hoc
Netw. 2024,156, 103430. [CrossRef]
28.
Sipola, T.; Alatalo, J.; Kokkonen, T.; Rantonen, M. Artificial intelligence in the IoT era: A review of edge AI hardware and software.
In Proceedings of the 31st Conference of Open Innovations Association (FRUCT), Helsinki, Finland, 27–29 April 2022; pp. 320–331.
[CrossRef]
29.
McEnroe, P.; Wang, S.; Liyanage, M. A Survey on the Convergence of Edge Computing and AI for UAVs: Opportunities and
Challenges. IEEE Sens. J. 2022,9, 15435–15459. [CrossRef]
30.
Tu, D.V.; Quang, P.M.; Nghi, H.P.; Thinh, T.N. An Edge AI-Based Vehicle Tracking Solution for Smart Parking Systems. In Intel-
ligence of Things: Technologies and Applications, Proceedings of the International Conference on Intelligence of Things (ICIT 2023), Ho
Chi Minh City, Vietnam, 25–27 October 2023; Dao, N.N., Thinh, T.N., Nguyen, N.T., Eds.; Lecture Notes on Data Engineering and
Communications Technologies; Springer: Cham, Switzerland, 2023; Volume 187, pp. 234–243.
31. Dutta, D.L.; Bharali, S. TinyML meets IoT: A comprehensive survey. Internet Things 2021,16, 100461. [CrossRef]
32.
Kallimani, R.; Pai, K.; Raghuwanshi, P.; Iyer, S.; López, O.L. TinyML: Tools, applications, challenges, and future research directions.
Multimed. Tools Appl. 2024,83, 29015–29045. [CrossRef]
33.
Oliveira, F.; Costa, D.G.; Assis, F.; Silva, I. Internet of Intelligent Things: A Convergence of Embedded Systems, Edge Computing
and Machine Learning. Internet Things 2024,26, 101153. [CrossRef]
34.
Capogrosso, L.; Cunico, F.; Cheng, D.S.; Fummi, F.; Cristani, M. A Machine Learning-oriented survey on tiny Machine Learning.
IEEE Access 2024,12, 23406–23426. [CrossRef]
35. Bhardwaj, S.; Syed, A.A.; Ozcelebi, T.; Lukkien, J. Power-managed smart lighting using a semantic interoperability architecture.
IEEE Trans. Consum. Electron. 2011,57, 420–427. [CrossRef]
36.
Kim, T.Y.; Park, N.H. Design of path prediction smart street lighting system on the Internet of Things. J. Chosun Nat. Sci. 2019,
12, 14–19.
37.
Northeast Group. The Benefits of LED & Smart Street Lighting. A Performance Benchmark of US Cities, Northeast Group and
CityLab Insights. 2021. Available online: https://northeast-group.com/services-2/ (accessed on 10 November 2024).
38.
Putri, M.; Abdullah, I.J.; M. Syahruddin, C.; Guci, P.S.; Sembiring, E.A.B.; Evalina, N. Control and Monitoring System for Lighting
Through Timing Based on The Internet of Things. Int. J. Appl. Res. Sustain. Sci. (IJARSS) 2023,2, 781–792.

Future Internet 2025,17, 90 47 of 50
39.
Polepaka, S.; Swami Das, M.; Ram Kumar, R.P. Internet of things and its applications: An overview. Lect. Notes Electr. Eng. 2020,
643, 67–75.
40.
Zandbergen, D. The Unfinished Lampposts: The (anti-) Politics of the Amsterdam Smart Lighting Project. City Soc. 2020,
32, 135–156. [CrossRef]
41.
Kanthi, M.; Dilli, R. Smart streetlight system using mobile applications: Secured fault detection and diagnosis with optimal
powers. Wirel. Netw. 2023,29, 2015–2028. [CrossRef]
42.
Nusrat, M.A.; Paul, S.; Bhushan, B. Practicle Coordination and Aspect of IoT for Smart Cities and Healthcare System. In Proceed-
ings of the 12th International Conference on System Modeling and Advancement in Research Trends, SMART,
Moradabad, India,
22–23 December 2023; pp. 280–287. [CrossRef]
43. Chiradeja, P.; Yoomak, S. Development of public lighting system with smart lighting control systems and internet of thing (IoT)
technologies for smart city. Energy Rep. 2023,10, 3355–3372. [CrossRef]
44. Mouaadh, Y.; Bousmaha, B.; Mhamed, R. Intelligent Control and Reduce Energy Consumption of Smart Street Lighting System.
Int. J. Power Electron. Drive Syst. (IJPEDS) 2022,13, 1966–1974. [CrossRef]
45.
Nigel, F.T.; Longe, O.M. Smart energy efficient lighting system for smart buildings. In Proceedings of the IEEE PES/IAS
PowerAfrica, Nairobi, Kenya, 23–27 August 2021. [CrossRef]
46.
Yu, T.; Yang, X.; Sang, P. Design strategy of green intelligent building using deep belief network. Int. J. Syst. Assur. Eng. Manag.
2023,14, 196–205. [CrossRef]
47.
Widartha, V.P.; Ra, I.; Lee, S.Y.; Kim, C.S. Advancing Smart Lighting: A Developmental Approach to Energy Efficiency through
Brightness Adjustment Strategies. J. Low Power Electron. Appl. 2024,14, 6. [CrossRef]
48.
Vinh, P.V.; Dung, P.X. Designing a Smart Lighting System for Illuminating Learning Experiences. In From Smart City to Smart
Factory for Sustainable Future: Conceptual Framework, Scenarios, and Multidiscipline Perspectives; Lecture Notes in Networks and
Systems, 1062 LNNS; Springer: Cham, Switzerland, 2024; pp. 296–305.
49.
Bernardo, M.; Fajardo, A.; Medina, R. End-Product of Solar-Sharing Smart Lighting Artificial Intelligence Driven Platform for
High-Valued Crops (Lactuca Sativa) on Indoor Hydroponics System. In Proceedings of the IEEE 10th Conference on Systems,
Process and Control, ICSPC, Malacca, Malaysia, 17 December 2022; pp. 160–165. [CrossRef]
50.
Bernardo, M.S. DLI and PPFD throughput of Solar and AI-Based Smart Lighting Apply on Illumination Stratums. In Proceedings
of the IEEE 11th Conference on Systems, Process and Control, ICSPC, Malacca, Malaysia, 16 December 2023; pp. 171–176.
[CrossRef]
51.
Ferrari, A.; Micucci, D.; Mobilio, M.; Napoletano, P. On the Personalization of Classification Models for Human Activity
Recognition. IEEE Access 2020,8, 32066–32079. [CrossRef]
52.
Matveev, I.; Karpov, K.; Iushchenko, M.; Dugaev, D.; Luzianin, I.; Siemens, E.; Chmielewski, I. Comparative Analysis of Object
Detection Methods in Computer Vision for Low-Performance Computers Towards Smart Lighting Systems. In Progress in Advanced
Information and Communication Technology and Systems; Lecture Notes in Networks and Systems; Springer:
Cham, Switzerland
,
2023; Volume 548, pp. 203–215.
53.
Laad, M.; Maurya, R.; Saiyed, N. Unveiling the Vision: A Comprehensive Review of Computer Vision in AI and ML. In Proceed-
ings of the International Conference on Advances in Data Engineering and Intelligent Computing Systems (ADICS), Chennai,
India, 18–19 April 2024; pp. 1–6. [CrossRef]
54. Vaidya, M.; Chatterjee, P.; Bhopatkar, K. Energy Efficient Smart Lighting System for Rooms. Stud. Big Data 2022,92, 107–125.
55.
Tandel, N.H.; Prajapati, H.B.; Dabhi, V.K. Voice Recognition and Voice Comparison using Machine Learning Techniques: A
Survey. In Proceedings of the 6th International Conference on Advanced Computing and Communication Systems (ICACCS),
Coimbatore, India, 6–7 March 2020; pp. 459–465. [CrossRef]
56.
Chandolikar, N.; Joshi, C.; Roy, P.; Gawas, A.; Vishwakarma, M. Voice Recognition: A Comprehensive Survey. In Proceedings
of the International Mobile and Embedded Technology Conference (MECON), Noida, India, 10–11 March 2022; pp. 45–51.
[CrossRef]
57.
Sarbast, H.M.; Abdulazeez, M. Voice Recognition Based on Machine Learning Classification Algorithms: A Review. Indones. J.
Comput. Sci. (IJCS) 2024,13, 4414–4431. [CrossRef]
58.
Parise, G.; Kermani, M.; Zissis, G.; Cumberbatch, T. A Comprehensive Exploration of Smart Lighting Aspects: Area of Use,
Methodologies and Purposes. In Proceedings of the IEEE Industry Applications Society Annual Meeting, IAS,
Nashville, TN, USA,
29 October–2 November 2023. [CrossRef]
59.
Wohlin, C.; Kalinowski, M.; Felizardo, K.; Mendes, E. Successful combination of database search and snowballing for identification
of primary studies in systematic literature studies. Inf. Softw. Technol. 2022,147, 106908. [CrossRef]
60.
Wohlin, C. Guidelines for Snowballing in Systematic Literature Studies and a Replication. In Proceedings of the 18th International
Conference on Evaluation and Assessment in Software Engineering, EASE 2014, London, UK, 13–14 May 2014. [CrossRef]
61.
Sakshi, K.V. Machine Learning and non-Machine Learning methods in mathematical recognition systems: Two decades’ systematic
literature review. Multimed. Tools Appl. 2024,83, 27831–27900. [CrossRef]

Future Internet 2025,17, 90 48 of 50
62.
Kitchenham, B.; Brereton, P.O.; Budgen, D.; Turner, M.; Bailey, J.; Linkman, S. Systematic literature reviews in software engineering:
A systematic literature review. Inf. Softw. Technol. 2009,51, 7–15. [CrossRef]
63.
Yang, L.; Zhang, H.; Shen, H.; Huang, X.; Zhou, X.; Rong, G.; Shao, D. Quality Assessment in Systematic Literature Reviews: A
Software Engineering Perspective. Inf. Softw. Technol. 2021,130, 106397. [CrossRef]
64.
Cadavid, R.H.; Andrikopoulos, V.; Avgeriou, P. Architecting systems of systems: A tertiary study. Inf. Softw. Technol. 2020,
118, 106202. [CrossRef]
65.
Singh, J.; Singh, P.; Gill, S.S. Fog computing: A taxonomy, systematic review, current trends and research challenges. J. Parallel
Distrib. Comput. 2021,157, 56–85. [CrossRef]
66.
Iftikhar, S.; Gill, S.S.; Song, C.; Xu, M.; Aslanpour, M.S.; Toosi, A.N.; Du, J.; Wu, H.; Ghosh, S.; Chowdhury, D.; et al. AI-based fog
and Edge Computing: A systematic review, taxonomy and future directions. Internet Things 2023,21, 100674. [CrossRef]
67.
Budgen, D.; Brereton, P.; Drummond, S.; Williams, N. Reporting systematic reviews: Some lessons from a tertiary study. Inf. Softw.
Technol. 2018,95, 62–74. [CrossRef]
68.
Pérez, J.D.L.R.B.; Vera, R.A.A. Quality Assessment for Selected Primary Studies in a Systematic Literature Review: A Case Study.
In Proceedings of the IEEE Mexican International Conference on Computer Science (ENC), Xalapa, Mexico,
24–26 August
2022;
pp. 1–6. [CrossRef]
69.
Ivarsson, M.; Gorschek, T. A method for evaluating rigor and industrial relevance of technology evaluations. Empir. Softw. Eng.
2011,16, 365–395. [CrossRef]
70.
Rafsanjani, H.N.; Ghahramani, A.; Nabizadeh, A.H. iSEA: IoT-based smartphone energy assistant for prompting energy-aware
behaviors in commercial buildings. Appl. Energy 2020,266, 114892. [CrossRef]
71.
Sun, B.; Zhang, Q.; Cao, S. Development and implementation of a self-optimizable smart lighting system based on learning
context in classroom. Int. J. Environ. Res. Public Health 2020,17, 1217. [CrossRef] [PubMed]
72.
Mohagheghi, A.; Moallem, M. An Energy-Efficient PAR-Based Horticultural Lighting System for Greenhouse Cultivation of
Lettuce. IEEE Access 2023,11, 8834–8844. [CrossRef]
73.
Natarajan, A.; Krishnasamy, V.; Singh, M. Design of a Low-Cost and Device-Free Human Activity Recognition Model for Smart
LED Lighting Control. IEEE Internet Things J. 2024,11, 5558–5567. [CrossRef]
74.
Putrada, A.G.; Abdurohman, M.; Perdana, D.; Nuha, H.H. NearCount for Model Compression on Edge Computing-Based Smart
Lighting with Product-of-Sum Function. In Proceedings of the 2024 International Conference on Smart Computing, IoT and
Machine Learning (SIML), Surakarta, Indonesia, 6–7 June 2024; pp. 13–18. [CrossRef]
75.
Hussaini, A.; Qian, C.; Guo, Y.; Lu, C.; Yu, W. Digital Twins of Smart Campus: Performance Evaluation Using Machine Learning
Analysis. In Proceedings of the IEEE/ACIS 21st International Conference on Software Engineering Research, Management and
Applications, SERA, Orlando, FL, USA, 23–25 May 2023; pp. 132–137. [CrossRef]
76.
Violatto, G.; Pandharipande, A.; Li, S.; Schenato, L. Anomalous occupancy sensor behavior detection in connected indoor lighting
systems. In Proceedings of the IEEE 5th World Forum on Internet of Things, WF-IoT, Limerick, Ireland, 15–18 April 2019;
pp. 335–339. [CrossRef]
77.
Putrada, A.G.; Ramadhan, N.G.; Al Makky, M. An Evaluation of Activity Recognition with Hierarchical Hidden Markov Model
and other Methods for Smart Lighting in Office Buildings. ICIC Express Lett. 2022,16, 91–100. [CrossRef]
78.
Zhu, J.; Tan, J.; Wu, W. Data-Driven End-to-End Lighting Automation Based on Human Residential Trajectory Analy-
sis. In Proceedings of the International Conference on Smart Applications, Communications and Networking, SmartNets,
Harrisonburg, VA, USA, 28–30 May 2024. [CrossRef]
79.
Narasimharao, M.; Swain, B.; Nayak, P.P.; Bhuyan, S. Assessing the Feasibility and Efficacy of TinyML Based Voice-Activated
LED Lighting System for Smart Village Micro-Utilities. In Proceedings of the IEEE 3rd International Conference on Sustainable
Energy and Future Electric Transportation, SEFET, Bhubaneswar, India, 9–12 August 2023. [CrossRef]
80.
Na, J.; Zhang, H.; Deng, X.; Zhang, B.; Ye, Z. Accelerate personalized IoT service provision by cloud-aided edge Reinforcement
Learning: A case study on smart lighting. In Service-Oriented Computing, Proceedings of the 18th International Conference, ICSOC
2020, Dubai, United Arab Emirates, 14–17 December 2020; Lecture Notes in Computer Science (including subseries Lecture Notes in
Artificial Intelligence and Lecture Notes in Bioinformatics), 12571 LNCS; Springer: Cham, Switzerland, 2020; pp. 69–84.
81.
Bugeja, J.; Persson, J.A. A Data-Centric Anomaly-Based Detection System for Interactive Machine Learning Setups. In Proceedings
of the 18th International Conference on Web Information Systems and Technologies-WEBIST, Valletta, Malta, 25–27 October 2022;
pp. 182–189. [CrossRef]
82.
Singh, R.; Sharma, R.; Vaseem A.S.; Gehlot, A.; Buddhi, D.; Malik, P.K.; Arya, R. Highway 4.0: Digitalization of highways for
vulnerable road safety development with intelligent IoT sensors and Machine Learning. Saf. Sci. 2021,143, 105407. [CrossRef]
83.
Maharani, T.; Abdurohman, M.; Putrada, A.G. Smart Lighting in Corridor Using Particle Swarm Optimization. In Proceedings of
the 2019 4th International Conference on Informatics and Computing, ICIC, Semarang, Indonesia, 16–17 October 2019. [CrossRef]

Future Internet 2025,17, 90 49 of 50
84.
Hsueh, Y.F.; Kuan, C.X.; Ho, K.C.; Liu, Y.H.; Cheng, Y.S.; Kung, C.J. Indoor automatic dimming system based on parti-
cle swarm optimization. In Proceedings of the 2024 IEEE 33rd International Symposium on Industrial Electronics (ISIE),
Ulsan, Republic of Korea, 18–21 June 2024; pp. 1–6. [CrossRef]
85.
Intarungsee, I.; Thararak, P.; Jirapong, P.; Pengwon, K.; Kaewwong, S. Intelligent Internet of Things Using Artificial Neural
Networks and Kalman Filters for Energy Management Systems. In Proceedings of the International Electrical Engineering
Congress (iEECON), Khon Kaen, Thailand, 9–11 March 2022. [CrossRef]
86.
Aizono, Y.; Sakamoto, T.; Ochiai, H.; Esaki, H. Building Automation with Vision Transformer Using Synthetic Indoor Images
for Room Light Control. In Proceedings of the 16th International Conference on Knowledge and Smart Technology—KST,
Krabi, Thailand, 28 February–2 March 2024; pp. 40–44. [CrossRef]
87.
Zdankin, P.; Matkovic, V.; Schiele, G.; Weis, T. Towards Predicting the Perceived Brightness in a Smart Home Through Symbolic
Regression. In Proceedings of the IEEE International Conference on Pervasive Computing and Communications Workshops and
other Affiliated Events, PerCom Workshops, Biarritz, France, 11–15 March 2024; pp. 457–460. [CrossRef]
88.
Banerjee, T.; Chowdhury, P.; Ghosal, S.; Saha, S.K.; Bandyopadhayay, R. An integrated domestic sensing and control system with
supervisory check and real-time data acquisition. Adv. Intell. Syst. Comput. 2020,1112, 645–657.
89.
Zantalis, F.; Koulouras, G.; Karabetsos, S.; Kandris, D. A review of Machine Learning and IoT in smart transportation. Future
Internet 2019,11, 94. [CrossRef]
90.
Putrada, A.G.; Abdurohman, M.; Perdana, D.; Nuha, H.H. CIMA: A Novel Classification-Integrated Moving Average Model for
Smart Lighting Intelligent Control Based on Human Presence. Complexity 2022,2022, 4989344. [CrossRef]
91.
Putrada, A.G.; Abdurohman, M.; Perdana, D.; Nuha, H.H. EdgeSL: Edge-Computing Architecture on Smart Lighting Control
with Distilled KNN for Optimum Processing Time. IEEE Access 2023,11, 64697–64712. [CrossRef]
92.
Zhang, J.; Chen, Z.; Wang, A.; Li, Z.; Wan, W. Intelligent Personalized Lighting Control System for Residents. Sustainability 2023,
15, 15355. [CrossRef]
93.
Putrada, A.G.; Abdurohman, M.; Perdana, D.; Nuha, H.H. Q8KNN: A Novel 8-Bit KNN Quantization Method for Edge
Computing in Smart Lighting Systems with NodeMCU. In Intelligent Systems and Applications (IntelliSys 2023); Arai, K., Ed.;
Lecture Notes in Networks and Systems; Springer: Cham, Switzserland, 2024; Volume 824.
94.
Putrada, A.G.; Abdurohman, M.; Perdana, D.; Nuha, H.H. Recurrent Neural Network Architectures Comparison in Time-Series
Binary Classification on IoT-Based Smart Lighting Control. In Proceedings of the 10th International Conference on Information
and Communication Technology, ICoICT, Virtual, 2–3 August 2022; pp. 391–396. [CrossRef]
95.
Prabowo, S.; Putrada, A.G.; Oktaviani, I.D.; Abdurohman, M. Camera-Based Smart Lighting System that complies with
Indonesia’s Personal Data Protection Act. In Proceedings of the ICADEIS 2023—International Conference on Advancement in
Data Science, E-Learning and Information Systems (ICADEIS), Bali, Indonesia, 2–3 August 2023; pp. 1–6. [CrossRef]
96.
Hadi, A.A.; Pahlevi, R.R.; Putrada, A.G. Office Room Smart Lighting Control with Camera and SSD MobileNet Object Localization.
In Proceedings of the International Conference on Advanced Creative Networks and Intelligent Systems: Blockchain Technology,
Intelligent Systems, and the Applications for Human Life, Bandung, Indonesia, 23 November 2022. [CrossRef]
97.
Khemakhem, S.; Krichen, L. A comprehensive survey on an IoT-based smart public street lighting system application for smart
cities. Frankl. Open 2024,8, 100142. [CrossRef]
98.
Mora, H.; Peral, J.; Ferrández, A.; Gil, D.; Szymanski, J. Distributed Architectures for Intensive Urban Computing: A Case Study
on Smart Lighting for Sustainable Cities. IEEE Access 2019,7, 58449–58465. [CrossRef]
99.
Sharma, V.; Srivastava, D.; Kumar, L.; Payal, M. A novel study on IoT and Machine Learning-based transportation. In Machine
Learning Techniques and Industry Applications; IGI Global: Hershey, PA, USA, 2024; pp. 1–28. [CrossRef]
100.
Sung, W.T.; Devi, I.V.; Hsiao, S.J. Smart Lamp Using Google Firebase as Realtime Database. Intell. Autom. Soft Comput. 2022,
33, 967–982. [CrossRef]
101.
Alahi, M.E.E.; Sukkuea, A.; Tina, F.W.; Nag, A.; Kurdthongmee, W.; Suwannarat, K.; Mukhopadhyay, S.C. Integration of IoT-
Enabled Technologies and Artificial Intelligence (AI) for Smart City Scenario: Recent Advancements and Future Trends. Sensors
2023,23, 5206. [CrossRef]
102.
G˘al˘atanu, C.D.; Haba, C.G.; Petrisor, D.; Breniuc, L. Imaging measurements for public lighting predictive maintenance. In Pro-
ceedings of the 11th International Symposium on Advanced Topics in Electrical Engineering (ATEE), Bucharest, Romania, 28–30
March 2019; pp. 1–5.
103.
Shao, Z.; Li, Y.; Huang, P.; Abed, A.M.; Ali, E.; Elkamchouchi, D.H.; Abbas, M.; Zhang, G. Analysis of the opportunities and costs
of energy saving in lightning system of library buildings with the aid of building information modelling and Internet of things.
Fuel 2023,352, 128918. [CrossRef]
104.
Goldblum, M.; Tsipras, D.; Xie, C.; Chen, X.; Schwarzschild, A.; Song, D.; M ˛adry, A.; Li, B.; Goldstein, T. Dataset Security for
Machine Learning: Data Poisoning, Backdoor Attacks, and Defenses. IEEE Trans. Pattern Anal. Mach. Intell. 2023,45, 1563–1580.
[CrossRef]

Future Internet 2025,17, 90 50 of 50
105.
Oprea, A.; Singhal, A.; Vassilev, A. Poisoning Attacks Against Machine Learning: Can Machine Learning Be Trustworthy?
Computer 2022,55, 94–99. [CrossRef]
106.
Paolone, G.; Pilotti, F.; Piazza, A. Evolution of the concept of device moving from Internet of Things to Artificial Intelligence of
Things. Int. J. Electr. Comput. Eng. (IJECE) 2024,14, 7236–7243. [CrossRef]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.