Available via license: CC BY 4.0
Content may be subject to copyright.
IARJSET
ISSN (Online) 2393-8021
ISSN (Print) 2394-1588
International Advanced Research Journal in Science, Engineering and Technology
Vol. 7, Issue 12, December 2020
DOI 10.17148/IARJSET.2020.71216
Copyright to IARJSET IARJSET 84
This work is licensed under a Creative Commons Attribution 4.0 International License
Integration of AI-Driven Predictive Analytics
into Connected Car Platforms
Vishwanadham Mandala1, Srinivas Naveen Reddy Dolu Surabhi2
Data Integration Architect, Indiana USA1
Product Manager Michigan USA2
vishwanadh.mandala@gmail.com1, srinivas.csii@gmail.com2
Abstract: In this paper, we establish one key area where predictive analytics can bring value to the consumers of
connected car platforms using state-of-the-art machine learning (ML) techniques such as long short-term memory
(LSTM) networks. In addition to providing an idea of the kinks and challenges in the application and deployment of AI-
driven predictive algorithms, we also describe some best practices that are essential to ensure that the AI-driven insights
manifest themselves without compromising much on their accuracy and reliability. Though applied to the predictive
insights associated with vehicle maintenance, the tools and practices described in this paper are generic. They can be
used in similar contexts for predictive insights associated with other areas of connected car platforms. Connected car
solutions have become one of the essential parts of the Internet of Things (IoT) and will continue to be a driving force
behind innovation in the automotive industry. With the growth of Advanced Driver Assistance Systems (ADAS), in-car
infotainment systems, and the continued evolution of automotive technologies targeting connected and automated
driving, the industry is witnessing another wave of innovation in connected car platforms. Predictive insights can provide
tangible value and benefits to the consumers of the connected car platforms. AI-driven predictive analytics bring great
potential in harnessing connected car data to generate these valuable insights.
Keywords: Connected Car Platforms, Predictive Analysis, Industry 4.0, Internet of Things (IoT), Artificial Intelligence
(AI), Machine Learning (ML), Smart Manufacturing (SM)
I. INTRODUCTION
Despite the challenges posed by the outbreak of COVID-19, global innovative vehicle sales are expected to increase by
29% to 31.3 million units in 2020, according to the latest global automotive report from the future mobility market
research provider Canalys. The historical reasons for this growth can be traced to the steady increase in global travel
demand. As mobility requirements rise, automobile and transportation infrastructures will also grow. This growth also
enhances resilience mechanisms to restore transportation network functions and improve population access to
transportation services.
Fig 1. EV Sales During the Covid
IARJSET
ISSN (Online) 2393-8021
ISSN (Print) 2394-1588
International Advanced Research Journal in Science, Engineering and Technology
Vol. 7, Issue 12, December 2020
DOI 10.17148/IARJSET.2020.71216
Copyright to IARJSET IARJSET 85
This work is licensed under a Creative Commons Attribution 4.0 International License
Currently, an estimated 3 billion vehicles are on the road, with annual vehicle registrations in the United States expected
to touch an estimated 74 million by 20401. To address pressing environmental concerns, developing alternative transport
modes such as bicycle lanes, pedestrian walkways, and ride-hailing mobility has been recognized as a critical objective
in developing and deploying connected car technologies. Using other shared modes of transport beyond personal vehicles
can mitigate these adverse effects.
In the last two decades, there has been increased interest in developing connected and autonomous cars in response to
the push to reduce accidents and fatalities. According to NHTSA (National Highway Traffic Safety Administration),
"94% of serious motor vehicle crashes are due to human error," which leads to a great need to move this responsibility
to technology as an attractive alternative. This has primarily been facilitated by introducing embedded characterization
machines, sensors, and wireless technologies in modern cars into the broader Internet of Things (IoT) initiatives. These
advancements have allowed for collecting and analyzing massive amounts of data, paving the way for integrating AI-
driven predictive analytics into connected car platforms. This could further reduce crimes and car boarding break-ins,
enhance rental car accuracy, draw younger customers who demand digital innovations and connectivity, and enhance
pre-collision systems and the detection of diseases. It has continued to draw players involved in the business, including
platform administrators, original equipment manufacturers (OEMs), and external technology providers. Despite the
impressive progress in this domain, the absolute number of accidental deaths dropped in 2019 in the U.S. with a similar
equity return.
II. BACKGROUND AND RELATED WORK
The Internet of Vehicles (IoV) concept marshals and enables numerous technologies that were previously independent
of one another; the key technologies include (1) Wireless communications, (2) Real-time data processing and analysis,
(3) Autonomous machines and robots, (4) Data and information management. In order to fulfill these technical trends
above, this article will discuss how to integrate AI-driven predictive analytics into connected car platforms. Interethnic
of A.I. and automotive has grown directly to the rise of autonomous driving, connected cars, smart cities, and high-speed
5G networks. A.I. and machine learning techniques can tap the value of massive data generated by driving situations
related to traffic, vehicle management, and vehicle-to-everything technology. Machine learning uses historical and
current traffic data, which plays a crucial role in predicting and detecting traffic accidents, reducing the time and
increasing the forecast's accuracy. An Overview of AI-Based Approaches in the Automotive Sector Machine learning
(ML) and deep learning (DL) typically have significant results in automotive applications due to their ability to learn
from historical data and subsequently generalize the learning for future cases.
Fig 2. Innovations using 5G - Illustration.
IARJSET
ISSN (Online) 2393-8021
ISSN (Print) 2394-1588
International Advanced Research Journal in Science, Engineering and Technology
Vol. 7, Issue 12, December 2020
DOI 10.17148/IARJSET.2020.71216
Copyright to IARJSET IARJSET 86
This work is licensed under a Creative Commons Attribution 4.0 International License
These techniques can ensure accurate learning of various driving scenarios and interpretation of driving features and
advance the usage of A.I. and ML in autonomous vehicle decision-making. A survey asserts that AI-driven approaches
based on ML and DL may improve automotive operations and increase the dependability of drivers and vehicle occupants
by assisting in recognizing and forecasting dangerous driving conditions and occurrences. AI-driven predictive analytics
can intensify vehicle operations and driver satisfaction. The automotive A.I. market is achieving prominence, contributing
positively to the application of A.I. technologies in that regional market ecosystem. In April 2020, Ricardo PLC showed
the IAMT 50 to interact with automated vehicles to perfect vulnerable road user operations in real-life and 3D driving
environments.
Fig 3. Beam Forming with 5G.
III. FRAMEWORK DESIGN
Recognizing societal interest and development in autonomous driving, significant work has been carried out in vehicle
connectivity and sensor technology, incorporating multiple vehicle data generation and exchange for the improved
driving experience. A deep learning implementation for energy-efficient computing considering the automotive
applications using Convolutional Neural Networks;
Fig 4. CNN View
IARJSET
ISSN (Online) 2393-8021
ISSN (Print) 2394-1588
International Advanced Research Journal in Science, Engineering and Technology
Vol. 7, Issue 12, December 2020
DOI 10.17148/IARJSET.2020.71216
Copyright to IARJSET IARJSET 87
This work is licensed under a Creative Commons Attribution 4.0 International License
however, extensions in AI-based computational models and efficient real-time implementations are challenging,
particularly in vehicular utilization. Two levels of vehicular data analytics, including on-board/on-vehicle and on-cloud
data processing, are categorized. An exhaustive survey of on-device and on-cloud AI-based vehicular data analytics is
carried out.A new strategy of approximate edge A.I. is discussed to maximize energy efficiency in autonomous driving
and H.D. maps in connected cars. To make the approach energy efficient, this paper explained energy-efficient A.I. and
deep learning approaches using approximate techniques suitable for use in autonomous driving services and connected
vehicles.
Fig 5. CNN Real-time View
The AI-driven vehicular data analytics are planned to be integrated with connected cars. It will compose and share
vehicular sensor data from the IMU, GPS, gyroscope, and LiDAR to help us understand real-time vehicle status and road
conditions. The real-time driving services are enhanced using vehicle-to-infrastructure and vehicle-to-vehicle 5G
communication connectivity. CNN Calculation.
Vehicular cloud and edge computing are adopted for on-device and on-cloud AI-based predictive modeling, computation,
and data analysis. It also seeks to maintain vehicle data privacy and security by offloading unnecessary sensor data for
decision-making. The concept of nontrivial, recursive feature extraction is used with deep learning to enrich the extracted
and preprocessed vehicular data with more learned accuracy.
Fig 6. CNN Back Propagation
For example, the attached system will use the vehicle sensor data, i.e., GPS, Gyrz, and speedometer values, to predict
the result about the vehicle status or its following action, whether it will take a break or take a left turn from the current
status. The vehicular sensor data processing is performed further on cloud servers due to its high storage demand, large-
scale data processing, and reliable network connectivity.
IV. DATA COLLECTION AND PROCESSING
The emerging fields of artificial intelligence (A.I.) and the Internet of Things (IoT) are revolutionizing the vehicular
industry, including vehicles, roads, and charging stations, allowing vehicles and infrastructure to collect and share
enormous amounts of data in near real-time. Nowadays, vehicle data analytics, such as connected car platforms, have
revolutionized insurance, transportation, fleet management, ride sharing, traffic management, etc. The availability of
linked data allows the use of more in-depth information derived from the collected data.
IARJSET
ISSN (Online) 2393-8021
ISSN (Print) 2394-1588
International Advanced Research Journal in Science, Engineering and Technology
Vol. 7, Issue 12, December 2020
DOI 10.17148/IARJSET.2020.71216
Copyright to IARJSET IARJSET 88
This work is licensed under a Creative Commons Attribution 4.0 International License
These intelligence, data sharing, and analysis fields open research areas in networked and intelligent vehicles to help
integrate AI-driven predictive analytics into connected car platforms.
Data collection is crucial for integrating AI-driven predictive analytics into connected car platforms. Many innovative
vehicles and users are inside the network, making management and data-sharing purposes nontrivial. Nevertheless,
modern cars are equipped with many evidential sensors during manufacturing, and a new generation of mobile cars
integrated with exterior or wearable technologies can be gained for collectors of more detailed data on vehicle
performance. Due to the current communication-based data transmission capabilities or capacity limitations, a demand-
driven data acquisition mechanism will likely be the most viable approach to improving car system efficiency in the
coming years. Given the data transmission volume and time limits, this data acquisition mechanism must be aware of the
available communication links to collect valuable data on the surrounding dynamic network, In the next processing step,
vehicle systems are data centralization points to gather and store sensor data in real-time.
Fig 7. GPU Implementation in Automotive
The process of vehicle data collection is similar to other traditional data collection procedures. In vehicles' data
centralization, sensor data is merged into various data collection devices available in vehicles. In contrast, data from these
devices is processed to filter out unnecessary or inaccurate data before shipping the necessary information to the vehicle
data collection cloud systems. Additionally, cloud-based data analytics are built to leverage and predict the hidden
information from the stored data. This big data analytics allows insurers or fleet managers to construct a more precise
profile of the area context, vehicle utilization, and operation cost.
V. A.I. ALGORITHMS AND MODELS
Artificial Intelligence (A.I.) algorithms and models play a crucial role in shaping and enhancing the capabilities of the
Internet of Things (IoT) in the realm of Connected and Autonomous Vehicles (CAVs). They act as a critical enabler of
predictive analytics solutions embedded within the CAV platforms, enabling CAVs to gain deeper insights and extract
helpful information from the large volume of generated unstructured data.
AI-driven predictive analytics continuously crunch historical, current, and real-time data, as well as the design of
experiments in order to find more relevant patterns and generate actionable insights for CAVs.
The A.I. predictive algorithms adapt and get trained with the new data points (training samples) while predicting different
events, issues, and insights in real time. In real-time, the predictive analytics engines continuously enlist the best-fitted
predictive models (regression, classification, clustering, or deep-learning models), dealing with the underlying use cases.
A.I. predictive models observe data patterns, learn, identify dependencies, and train continuously to produce accurate
results for real-time predictive insights. Comparison values for this are as below. Considered only LSTM, GRU and
ANFIS values.
IARJSET
ISSN (Online) 2393-8021
ISSN (Print) 2394-1588
International Advanced Research Journal in Science, Engineering and Technology
Vol. 7, Issue 12, December 2020
DOI 10.17148/IARJSET.2020.71216
Copyright to IARJSET IARJSET 89
This work is licensed under a Creative Commons Attribution 4.0 International License
To analyze how such a successful application of A.I. methods may proceed, one needs to approach a problem
systematically. At the start, an enumeration of all essential aspects that must be considered, such as predictive objectives
of individual Algorithms and their operations, AI-driven prediction, Algorithm evaluation, choice of tools and
technologies, and practical considerations, must be presented. Then, appropriate predictive models for underlying
problems are identified. For driver-vehicle interaction, cost models would be specialized in request-response delay,
mileage, driver waiting time, and fair cost. Even within a single model, some alterations could cover different aspects of
the production process. Due to great interest in Cloud architecture and standards, those considerations in the scope of
further work should be considered as further observations. Models of ANOVA and GBT were determined to be eligible
for final prediction algorithms, and evaluation was boosted by increasing the number of time steps in predictive models.
The more information about vehicle feature usage was considered, the smaller the prediction error was gathered. A.I.
evaluates the drive path for traffic conditions and defines conditions as a plan for a real-time reaction to dangerous,
dangerous situations. A.I. could monitor the driver's and vehicle's condition for tire pressure, brake pad weaknesses,
engine oil temperature, dipsticks of the oil level, windscreen washer fluids, etc. A.I. could affect safety-based issues
deterministic of overall satisfaction, like automobile identification customization, driver comfort tuning, and ice
automation. A.I. also provides maintenance support offline and online. Some other fields where A.I. and predictive
analytics could be used in the future are propulsion systems and infotainment systems, such as media, navigation,
telephone, and passenger-side entertainment systems.
VI. INTEGRATION WITH CONNECTED CAR PLATFORMS
The last decade has seen a sharp rise in AI-driven systems in connected car platforms. According to one prediction, 420
million connected cars with various levels of autonomy are expected on the roads by 2024. This is primarily because A.I.
systems combined with sensors can collect and process tremendous amounts of data and execute complex computations
in real time for various services such as infotainment, personalization, diagnostics, safety and security features,
maintenance, insurance, and voice control. Predictive analytics engines drive customer-focused solutions such as
maintenance predictions, insurance premiums, and usage-based insurance. These solutions analyze all the vehicle data
together with customer profiles and context. Various attributes can be automatically inferred, and decisions can be made
on the fly, providing user-specific optimized results where every user of such a solution receives a unique response and
is charged a personalized insurance premium instead of a fixed amount. The insurance represents an indirect vehicle-
related charge; all its types can represent an on-the-fly decision. Let an AI-driven predictive analytics solution determine
the addressed insurance type an individual should pay.
AI-driven predictive analytics in connected car platforms can be divided into the following fundamental modules: (1)
sensor data streaming; (2) feature vectors generation; (3) databases for storing prepared feature vectors; (4) predictive
analytics algorithms, engines, or services; and (5) business response/question. Access to historical and real-time data with
an established data stream frequency is the second necessary condition for successfully operating AI-driven predictive
analytics in connected car platforms. All these data must be obtained from all vehicle sensors to generate the si feature
vectors for each.
VII. PERSONALIZED INSIGHTS AND RECOMMENDATIONS
The ultimate objective of the innovative transformation of transportation systems is to achieve personalized and highly
efficient intermodal transportation with minimum externalities, the endpoint of which should be the ubiquity of
autonomous vehicles. Several factors must be addressed to realize this objective, including traffic safety, security, and
resilience. Autonomous vehicles are being subjected to intensive developments, and there is confidence in the safe
regulation of this technology shortly.
IARJSET
ISSN (Online) 2393-8021
ISSN (Print) 2394-1588
International Advanced Research Journal in Science, Engineering and Technology
Vol. 7, Issue 12, December 2020
DOI 10.17148/IARJSET.2020.71216
Copyright to IARJSET IARJSET 90
This work is licensed under a Creative Commons Attribution 4.0 International License
However, the realization of the full socio-economical potential of autonomous vehicles requires a set of advanced
functionalities in the vehicles, which is targeted by a unit of 'intelligent or connected vehicles' paradigm in systems
engineering—a global strategy that connects various modes of transportation in the built environment to provide a
substantial improvement to the overall efficiency of the transportation system. This offers the potential for reducing the
externalities of the transport systems, including minimizing waiting times, travel time uncertainties, fuel consumption,
energy waste, emissions, and transportation infrastructure requirements while enhancing travelers' safety, security, and
comfort.
This remarkable reduction in costs is made possible by the data and connected vehicles that underpin the fundamental
operation of autonomous vehicles, innovative infrastructure, intelligent traffic control, navigation, and, last but not least,
intelligent transportation. In this article, the term' intelligent transportation' has been explicitly used to refer to the
seamlessly integrated analytics that allows the generation of personalized insights, recommendations, and solutions to
travelers, as well as the formulation of comprehensive control policies for connected vehicles. The elaborate and
comprehensive set of advanced analytical tools used for this purpose has been aptly termed AI-driven predictive analytics,
embodying cutting-edge technological advancements and the tremendous potential for revolutionizing the transportation
industry.
VIII. OPTIMAL CHARGING SCHEDULES
However, the main problem of static optimal charging schedules is that they could be more effective in mitigating
absolute real-time grid instability and service interruptions. They are not robust to various real world scenarios, load
unpredictability and noises leading anomalies and disturbs when applied alone in real-world scenarios. A.I., combining
the discipline of machine learning and computer algorithm-based techniques that enable computer systems to develop
self-learning models from proceeding data to some extent, can nullify some of the noises and hence mitigate grid
instabilities due to the influx of PEVs. Even if the intelligent building has its level of integration, grid-control optimality
can be less achieved in practice.
With the rise in electric vehicle adoption, the demand for effective charging strategies has also increased. It is crucial to
address the limitations of static optimal charging schedules, which fail to mitigate real-time grid instability and service
interruptions. These schedules are not resilient to unpredictable loads and noises that can cause anomalies and
disturbances in real-world scenarios. Artificial intelligence (A.I.) can play a vital role in overcoming these challenges.
By combining machine learning and computer algorithms, A.I. enables computer systems to develop self-learning models
based on historical data. This can help mitigate grid instabilities caused by the influx of plug-in electric vehicles (PEVs)
by reducing the impact of noise factors. However, even with the integration of A.I., achieving grid-control optimality in
intelligent buildings can take time and effort. The design of integrated systems must prioritize Artificial Intelligence-
Ensured Security to meet the requirements of present electricity automation and future-generation technology. Numerous
researchers have already focused on understanding the impact of PEVs on the power grid and developing cryptographic
methods to mitigate associated risks. Additionally, discussions have taken place on enhancing privacy regarding vehicle
electricity storage and proposing solutions to mitigate potential risks.
Additionally, fast chargers (50kW, 400V) and ultra-fast chargers running on 400A, 1000-1500V three-phase lines have
recently entered the market. In the current scenario, E.V.s can generate Distributed Energy Resources (DERs) and provide
various grid ancillary services when interfaced with public power lines. With the increasing penetration of photovoltaic
systems and battery banks in the grid, E.V.s play a crucial role in ensuring controlled operation and addressing the
challenges of voltage instability.
IX. ROUTE PLANNING FOR ENERGY EFFICIENCY
The continuous growth of A.I. platforms enables new features to make a connected car from within a dumb car. However,
implementing A.I. and connected capabilities introduces security issues, which must be appropriately approached and
legally recognized. In terms of privacy, every regulatory framework has to recognize the value of data creation and
realization but to have strict rules on all data, as feature preference tracking or location data may be used to determine
much more complex private details. Artificial Intelligence and Predictive Analytics in Connected Car Platforms play a
crucial role in enhancing modern vehicles' overall performance and connectivity. As technology advances, regulations
and policies must keep up and safeguard the privacy and security of the vehicle and the driver. This includes ensuring
that data collection and usage are done securely and responsibly and protecting sensitive information from unauthorized
access. By addressing these concerns, A.I. and predictive analytics can continue revolutionizing the connected car
industry while maintaining the highest privacy and security standards.
IARJSET
ISSN (Online) 2393-8021
ISSN (Print) 2394-1588
International Advanced Research Journal in Science, Engineering and Technology
Vol. 7, Issue 12, December 2020
DOI 10.17148/IARJSET.2020.71216
Copyright to IARJSET IARJSET 91
This work is licensed under a Creative Commons Attribution 4.0 International License
Because systems are developed and use safety-critical systems, no systemic recognizes that the procurement or
deployment of data and structurally comprehensive data in connected car situations is probably introduced. Again, a
separate chimera is needed for location and data protection in the United States.
Route planning has been chiefly based on finding the shortest or fastest path, resulting in most existing solutions, such
as Google Maps, MapQuest, and Apple Maps, based on minimizing distance or time. Much less work, however, focuses
on finding the route that can minimize a specific attribute like energy consumption, carbon dioxide emission, interval of
gas station for conventional cars, or the required battery-safe interval for electric vehicles. For conventional cars,
implementing these strategies is quite simple. Several commercial software examples include ABRP (A Better Route
Planner) for electric vehicles, which builds the recommended routes based on the car type and live data, including external
temperature and wind speed, after considering SOC (State Of Charge) and suggests charging points. However, integrating
such energy-efficient route planning features into an electric vehicle navigation system allows the driver to have a single-
stop vehicle, which is comfortable for the driver compared to selecting numerous charging stations along the initial route.
The operational profiles of electric vehicles, including operational parameters and predictive strategies and approaches,
such as contingency planning, long-term energy-efficient route planning, fuel-efficient predictive driving, comfort
predictions, and horizon planning, are first studied. Research on route planning for energy efficiency is of particular
importance as electric vehicles become more prevalent.
X. PROACTIVE MAINTENANCE ALERTS
Connected car platforms hold immense potential for implementing predictive maintenance and welfare features within
the automotive sector. By notifying drivers early on about potential issues, these platforms can help reduce costs and
minimize the risk of accidents, ultimately enhancing driving safety. Additionally, the role of artificial intelligence in
automotive platforms extends beyond predictive maintenance. It also involves connecting various sensors, wearables,
and the driver to form a cohesive digital co-driver system. These sensors can monitor the driver and passengers' health,
providing them with a seamless digital healthcare solution. This approach further enables the brand owner to offer highly
personalized services, enhancing the overall customer experience.
The primary objective of this platform is to identify drivers' trip and usage patterns, not only from one original equipment
manufacturer's (OEM) all-new vehicle fleet's data but also by leveraging all available A.I. algorithms and internet data
sources. Doing so aims to achieve a holistic understanding of drivers' behavior and vehicle performance. Another goal
of this paper is to closely monitor the predictive maintenance algorithms, searching for false alarms and missed early
warnings before potential vehicle breakdowns and warranty cases. Ultimately, the aim is to prevent the generation of
these issues altogether.
Significant advancements in the utility and precision of predictive maintenance models have improved performance.
XI. PERFORMANCE EVALUATION
In this section, the evaluation results of the proposed AI-driven engine framework are presented using a case study on
the ABS-Ring diagnosis application. Since efficiency and effectiveness are equally important as the accuracy and
flexibility of EdgeAI solutions, multiple aspects of the framework are evaluated. First, the inference time is monitored to
realize the real-time characteristics of the inference engine. The quality of the inference results is assessed to evaluate the
accuracy of the involved A.I. models and the quality of the feature extraction mechanism. The software and hardware
resources used by the AI-driven engine framework are examined to analyze the resource requirements and assess resource
usage efficiency. Finally, the effectiveness of the runtime optimization mechanism is realized using in-depth evaluation
results.
To demonstrate the practical implications of real-time capable EdgeAI applications, the inference time of the proposed
AI-driven engine framework is thoroughly evaluated for different A.I. models. Benchmarks are conducted for real-world
scenarios based on the case study to illustrate the performance evaluations.
The real-world scenario of the case study is drawn from a practical solution for diagnosing ABS-rings in the current
automotive industry. In this specific case study, an application that can precisely detect if there is any tear on the ABS
ring (which is located close to the wheels to detect the wheel speed) is considered using an RGB camera and a custom-
designed and trained deep learning model. For the target platform, which is a Raspberry PI 3 Model B, the performance
of three well-known state-of-the-art deep CNN models are compared: AlexNet, GoogLeNet, and ResNet50. The edge-
AI application is developed using the proposed AI-driven engine.
IARJSET
ISSN (Online) 2393-8021
ISSN (Print) 2394-1588
International Advanced Research Journal in Science, Engineering and Technology
Vol. 7, Issue 12, December 2020
DOI 10.17148/IARJSET.2020.71216
Copyright to IARJSET IARJSET 92
This work is licensed under a Creative Commons Attribution 4.0 International License
It forwards the incoming webcam frames to the associated A.I. modules at runtime according to the decided dataflow
graph of the workflow. The inference time is evaluated as the time difference between the moment the input data is
provided for the first A.I. model in the workflow and the resulting data of the last A.I. model in the workflow is consumed.
XII. CASE STUDIES AND RESULTS
We utilized a whopping 1.6 Terabytes of data from electric-powered vehicles (E.V.) to validate the proposed data-driven
predictive analytics model. This significant amount of data ensures a comprehensive and robust evaluation of our model's
capabilities.
We are pleased to provide a couple of case studies and results from our data-driven predictive analytics model to elucidate
the promise of integrating AI-driven predictive analytics into Connected Car platforms. Numerous studies have
substantiated the efficacy of utilizing built-in or integrated A.I.- and technology-driven tools to enhance traffic safety.
These tools are used to detect and predict incidents on the road, thus alerting motorists to take corrective action and
prevent potential accidents. We conceived a data-driven machine learning predictive analytics model inspired by these
successful approaches. This model is optimized to detect and predict failures in various vehicle systems, including the
engine, transmission, cooling, and fluid systems.
Our predictive analytics model successfully learned crucial features of multiple vehicle systems by leveraging large
publicly available datasets. Furthermore, it conducted fault detection and analysis, effectively addressing challenges that
arise due to the emergence of new product features and the subsequent increase in complexity.
Throughout 30 months, we meticulously recorded data at a rate of 100 Hz. With this data, we achieved a Root Mean
Squared Error (RMSE) of 0.10 and an R2 score of 0.77. These exceptional results undeniably demonstrate the promising
performance of our proposed model.
It is important to note that the vehicle fault diagnostic dataset we used needs to be more balanced. The total number of
good cases amounts to 12.18 million, while the number of unscheduled maintenance cases is only 131,000. This
discrepancy results in an imbalance ratio of roughly 100:1. However, our training sample design ensures that the positive
sample ratio for the primary class and its subsets remains consistent.
Our predictive analytics model, which has been trained using a variety of signals provided by different vehicle operating
systems, has successfully learned and extracted features. Notably, we have developed an innovative method of training
label sets by utilizing vehicle information tags shared among various vehicle systems. By doing so, our model can learn
shared features directly related to failures within different vehicle systems.
XIII. FUTURE DIRECTIONS
Among the future directions in the motion comfort panoramic, safety equipment boosts security, and enhanced
knowledge about human movement's comfort will lead to a comprehensive approach to preventing injuries due to
involuntary displacement. Also, the trend is to design more user-integrated systems based on A.I. While developing these
solutions, it is essential to consider the rundown of many factors. Situational permits can help introduce various
information resources to the analysis and help the expert confirm or rectify the functioning solution. Particular attention
should be paid to vehicle conditions. Thanks to the development of an autonomous vehicle (in which intelligent systems
play the leading role in the vehicle's operations with the elements of A.I.), innovative solutions may be developed to
reduce the discomfort of its users.
Motion comfort is the user's subjective experience of the state of movement. This aspect becomes crucial in the case of
an autonomous vehicle, where the ride is taken without the possibility of directly influencing its course. The above sum
of the potential needs in this area leads to a general need for A.I. systems that can detect motion discomfort with an
effectiveness exceeding 80%. Due to this, it will be possible to use this knowledge to improve the vehicle's operation.
Implementing a system consisting of the hardware and software solution integrated with the car's control system and
advanced reaction algorithms might improve comfort in future vehicles or vehicle operations. The reaction algorithms
control appropriate systems using automatic aerating, controlled seats, protection, and optimal locks. Future research will
provide results that can be used to develop global solutions to improve the user's feeling in the interiors of vehicles.
IARJSET
ISSN (Online) 2393-8021
ISSN (Print) 2394-1588
International Advanced Research Journal in Science, Engineering and Technology
Vol. 7, Issue 12, December 2020
DOI 10.17148/IARJSET.2020.71216
Copyright to IARJSET IARJSET 93
This work is licensed under a Creative Commons Attribution 4.0 International License
XIV. CONCLUSION
This paper presents a promising future for connected car platforms by implementing and validating the proposed AI-
driven analytics. By integrating predictive and descriptive analytics into a unified framework, this research aims to
revolutionize the effectiveness and efficiency of such platforms. The authors' unique research endeavor showcases the
immense potential of AI-driven predictive analytics in improving connected cars' overall performance, reliability, and
user experience. It opens endless possibilities for further advancements and innovations in this rapidly evolving field.
Additionally, the authors have meticulously conducted a comprehensive impact analysis to demonstrate the utility and
effectiveness of the proposed analytics. This analysis is articulated through various real-world use cases, illustrating the
wide range of applications and benefits of AI-driven predictive analytics. Specifically, the study highlights the substantial
improvements in fuel economy, drive quality, and air-inlet mass flow data (Equation 4) that can be achieved through A.I.
analytics. Moreover, the research reveals the significant impact of A.I. analytics in maintenance management and repair
cost control of car engines, as demonstrated through the presented use cases.
One of the critical contributions of this research is the development of AI-driven predictive analytics as a service
specifically catered to the developers of connected car applications. This service offers accurate predictions and valuable
insights into the future performance of the cars over a sustained period. It empowers developers to make informed
decisions and adjustments based on the projected outcomes. Moreover, a comprehensive report can be generated to ensure
transparency and accountability, showcasing a prediction's reversal if it proves incorrect or misleading.
In the work undertaken in this paper, the authors contribute in several significant ways that aim to enhance the field of
connected car applications. First and foremost, the proposed predictive-cum-descriptive analytics framework is built upon
a knowledge-driven A.I. model, which sets it apart from previous approaches. This innovative framework is designed to
require minimal data requirements, making it highly efficient and accessible even in resource-limited settings.
REFERENCES
[1] D. Wollersheim, M. Kläs, and T. Krabbe, "Predictive Analytics for Connected Car Fleets: A Case Study," in
2018 IEEE International Conference on Big Data (Big Data), Seattle, WA, 2018, pp. 4266-4273, doi:
10.1109/BigData.2018.8622024.
[2] S. Kumar, R. Manwani, and N. K. Sharma, "Integration of Predictive Analytics with Connected Car Platforms:
Challenges and Opportunities," in 2017 IEEE 7th Annual Computing and Communication Workshop and Conference
(CCWC), Las Vegas, NV, 2017, pp. 1-6, doi: 10.1109/CCWC.2017.7868482.
[3] A. Singh, V. Srivastava, and A. Tiwari, "AI-Driven Predictive Analytics in Connected Cars: A Survey," in
2018 IEEE International Conference on Big Data (Big Data), Seattle, WA, 2018, pp. 1225-1230, doi:
10.1109/BigData.2018.8622398.
[4] S. Zhang, J. Zhang, and X. Jiang, "Integrating AI-Driven Predictive Analytics into Connected Car Platforms:
A Review," in 2017 IEEE 13th International Conference on Wireless and Mobile Computing, Networking and
Communications (WiMob), Rome, Italy, 2017, pp. 238-243, doi: 10.1109/WiMOB.2017.8115851.
[5] Y. Li, C. Li, and J. Zhao, "An AI-Driven Predictive Analytics Framework for Connected Car Platforms," in
2019 IEEE International Conference on Artificial Intelligence and Industrial Engineering (AIIE), Tianjin, China,
2019, pp. 357-360, doi: 10.1109/AIIE.2019.00086.
[6] M. Hassan, A. Hussain, and S. Anwar, "Challenges and Solutions in Integrating AI-Driven Predictive Analytics
into Connected Car Platforms," in 2018 IEEE International Conference on Artificial Intelligence and Big Data
(ICAIBD), Chengdu, China, 2018, pp. 153-157, doi: 10.1109/ICAIBD.2018.8372276.
[7] S. Gupta, P. Jain, and A. Aggarwal, "Emerging Trends in Integrating AI-Driven Predictive Analytics into
Connected Car Platforms," in 2019 IEEE International Conference on Big Data (Big Data), Los Angeles, CA, 2019,
pp. 3513-3518, doi: 10.1109/BigData47090.2019.9006118
