Content uploaded by Shivam Gupta
Author content
All content in this area was uploaded by Shivam Gupta on Jan 24, 2023
Content may be subject to copyright.
Full Terms & Conditions of access and use can be found at
https://www.tandfonline.com/action/journalInformation?journalCode=tprs20
International Journal of Production Research
ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/tprs20
How to use no-code artificial intelligence to
predict and minimize the inventory distortions for
resilient supply chains
Sunil Kumar Jauhar, Shashank Mayurkumar Jani, Sachin S. Kamble, Saurabh
Pratap, Amine Belhadi & Shivam Gupta
To cite this article: Sunil Kumar Jauhar, Shashank Mayurkumar Jani, Sachin S. Kamble, Saurabh
Pratap, Amine Belhadi & Shivam Gupta (2023): How to use no-code artificial intelligence to predict
and minimize the inventory distortions for resilient supply chains, International Journal of Production
Research, DOI: 10.1080/00207543.2023.2166139
To link to this article: https://doi.org/10.1080/00207543.2023.2166139
Published online: 24 Jan 2023.
Submit your article to this journal
View related articles
View Crossmark data
INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH
https://doi.org/10.1080/00207543.2023.2166139
How to use no-code artificial intelligence to predict and minimize the inventory
distortions for resilient supply chains
Sunil Kumar Jauhar a, Shashank Mayurkumar Jani b, Sachin S. Kamble c, Saurabh Pratap d,
Amine Belhadi eandShivamGupta f
aOperations Management and Decision Sciences, Indian Institute of Management Kashipur, Kashipur, Uttarakhand, India; bH.L. College of
Commerce, Gujarat University, Ahmedabad, India; cEDHEC Business School, Roubaix, France; dDepartment of Mechanical Engineering, Indian
Institute of Technology (IIT BHU), Varanasi, India; eRabat Business School, International University of Rabat, Salé, Morocco; fDepartment of
Information Systems, Supply Chain and Decision Making, NEOMA Business School, Reims, France
ABSTRACT
Consumers’ dramatic demand has a pernicious effect throughout the supply chain. It exacer-
bates inventory distortion because of significant revenue loss caused by stock-level issues. Despite
the availability of several forecasting techniques, large organisations, manufacturing firms, and e-
commerce websites collectively lose around $1.8 trillion annually to inventory distortion. If this
problem is solved, sales may increase by 10.3 percent. The businesses are concerned about mitigat-
ing this loss. Artificial intelligence (AI) can play a significant role in building resilient supply chains.
However, developing AI models consumes time and cost. In this paper, we propose a No Code Artifi-
cial Intelligence (NCAI) enabling non-technical companies to build machine learning models based
on production quantity and inventory replenishment. The development of the NCAI model is fast
and inexpensive. However, little research deals with applying NCAI to operations and supply chain
problems. Addressing the existing gap, we show the application of NCAI in the retail industry.
ARTICLE HISTORY
Received 2 October 2021
Accepted 1 January 2023
KEYWORDS
Inventory distortion;
No-Code Artificial
Intelligence (NCAI);
overstocks; out-of-stocks;
supply chain resilience
1. Introduction
In today’s highly competitive market, organisations are
compelled to seize every opportunity to improve their
business processes. Customers’ dramatic demand for
high-quality products with shorter lead times is a long-
time struggle to fulll, resulting in low-prot margins
(Sustrova 2016;Rajetal.2022;Ratusny,Schier,andEhm
2022). Retailers must be attuned to their customers’ pref-
erences to navigate and evolve in real time (Cattapan and
Pongsakornrungsilp 2022). Additionally, manufacturers’
and retailers’ primary inventory management objective is
to match product supply to consumer demand (Fairhurst
and Fiorito 1990;Shrouty2019; Kumar and Shukla 2022).
As a result, many production units, e-commerce web-
sites, businesses, and retail stores are trying to meet their
customer demands with increased eciency (Shalan and
Abdul-Rahman 2022;Rydell2022)
However, the supply chain faces a massive inventory
distortion problem in this battle to meet the probabilis-
tic nature of demand (Rizqi and Khairunisa 2020). They
frequently run out of inventory when the demand uc-
tuations are high, referred to as an ‘out-of-stock’ (OOS)
CONTACT Sachin S. Kamble Sachin.kamble@edhec.edu EDHEC Business School, Roubaix, France
situation. As a result, customers purchase products from
competitors, which results in customer dissatisfaction
and lost sales (San-José et al. 2017; San José Nieto et al.
2022). Meanwhile, the late arrival of stocks faces lower
demand as the customers have already purchased them
fromothersuppliers.Thisexcessinventoryiscalled‘over-
stocks’ (OS). It typically occurs when product demand is
insucient. OS requires rms and retailers to sell their
excess inventory at a reduced price to avoid becoming
obsolete (Kumar and Shukla 2022). OOS and OS issues
frequently pose challenges to the companies (IHL Group
1b;IHLGroup2016;IHLGroup2021). As a result,
deciding the inventory size and storage location is a com-
plicated problem (Mekel, Anantadjaya, and Lahindah
2014).
Minimising inventory-related costs maximises sales
and prot (San-José et al. 2017). By addressing the OS
and OOS situations, businesses could boost their sales
by 10.3% (IHL Group 2020). Eective inventory manage-
ment is one of the critical success factors for companies
that keep operations in motion (Panigrahi 2022;Pan-
igrahi et al. 2022). Inventory management is a crucial
© 2023 Informa UK Limited, trading as Taylor & FrancisGroup
2 S. KUMAR ET AL.
decision in logistics and supply chains, which requires
a balance between high holding costs due to OS in
one location and high shortage costs due to OOS in
another (Rumetna, Renny, and Lina 2020). Excess inven-
tory aects the working capital, while shortage leads to
lost sales (Shrouty 2019; Kumar and Shukla 2022). The
main objective of this study is to make supply chains more
resilient by aligning production and stock levels with cus-
tomer demand. Reducing excess inventory (overproduc-
tion) and out-of-stock (also known as less production)
situations is possible. Visualising the future by predict-
ing potential inventory requirements is necessary to meet
market demand without OS and OOS. Reduced OS and
OOS for the companies mean increased prots, satis-
ed customers, and competitive advantage (Nikolić et al.
2018;Papetal.2022; Prajapati et al. 2022).
Economic Ordering Quantity (EOQ) model deter-
minestheoptimalordersize,whichismodiedover
time to accommodate the requirements of various types
of businesses. Goyal and Giri (2001) extended the clas-
sical EOQ model to include deteriorating items with
dierent demand variations. The EOQ model assumes
that consumer demand, ordering, and holding costs are
constant over time (Shrouty 2019). However, in a world
wherethingschangecontinuously,theEOQmodelfails
in uncertain demand environments (San-José et al. 2017).
It is an overly restrictive and inexible solution for con-
temporary businesses.
As a result of these concerns, Articial Intelligence
(AI) appears to provide an optimal solution to inventory
replenishment and production quantity problems. Since
1956, when John Mccarthy coined the term, AI has been
a growing eld of research in all domains. However, our
ability to compete in this challenging market is limited
in many ways by modern technology (Hamilton 2018).
Adopting AI is time-consuming and complex (Enholm
et al. 2021). Due to changes in organisational structure,
continuous improvement and skill gaps alter employees’
roles beyond their daily routines. Therefore, employees
cannot navigate this intricate technology simultaneously.
For successful AI deployment, companies require a team
of dedicated technical workforce, a technology service
provider, IT engineers, data scientists, and analysts. Com-
panies should exercise caution when confronted with
rising labour costs and their risks. Big data and AI knowl-
edge are intangible assets acquired through education
and practice (Bag et al. 2021;SijuandHasan2022). Suc-
cessful AI adoption requires employees to have strong
coding and data analytics skills. Employees will need to
be trained and re-trained to deal with these short-term
problems,resultinginmorecostsandalongertimeto
implement new ideas (Ciner 2020). Businesses are nd-
ing ways to use AI without incurring high technological
workforce costs to achieve signicant savings in inven-
tory distortion loss.
However, ‘No Code Articial Intelligence’ (NCAI) can
help overcome AI’s high costs and other limitations.
NCAI oers a high-level, easy-to-use interface with sev-
eral pre-built templates that shorten adoption times and
simplify the adoption of analytics in the industry (Red-
chuk and Walas Mateo 2022). These no-code platforms
do not require deep programming knowledge. Currently,
NCAI provides drag-and-drop functionality to fully
automated ML services suitable for non-technical users
and software developers requiring little mathematical
orstatisticalknowledge.NCAIeliminatesAI’stime-
consuming nature and tasks related to software develop-
ment and algorithms, leading to faster results with lower
complexity (García-Ortiz and Sánchez-Viteri 2021).
Through NCAI, it is possible to adopt AI quickly with-
out coding skills or programming knowledge. NCAI is
a No-Code platform component, enabling businesses to
create more accessible, reliable, and aordable applica-
tions without writing a single line of code. As a result, the
costs reduce due to eliminating the need for IT experts
(Redchuk and Walas Mateo 2022). The NCAI can assist
organisations in sorting and analyzing data more quickly.
The NCAI – machine learning models are trained to
recognisepatternsandobjectsandtomakerapidpredic-
tions in a short period, resulting in increased eciency
and accuracy for businesses.
The present study calls attention to one of the world’s
most serious problems: inventory distortion (OS and
OOS) (San-José et al. 2017;IHLGroup2021,Statista2B
2022; Kumar and Shukla 2022). The study provides a
practical and deployable solution to this colossal and hor-
rifying issue confronting the retail and manufacturing
industries via a newer technology – NCAI. Our goal is to
ndwaystoalignproduction,inventory,andstocklev-
els with potential customer demand so that businesses
arenotsubjecttoOSandOOS.Inthisresearch,wetried
to nd out what quantities companies should replenish
or produce so that they will not have issues with being
over and out of stock. We also answer its aligned problem:
what quantities companies should create and manufac-
ture to reduce production distortion. In this study, we
attempttondamethodthatsolvesalltheseissuesata
low cost and as fast as possible to save precious corporate
timeandexpenses.WefoundAItobethebestmethod,
but it has become too costly and time-consuming due to
several constraints. So, we developed NCAI to use in such
situations.
The remaining of the paper is structured as follows.
Section 2discusses the literature on inventory distor-
tion and the various AI solutions. This section positions
our research within the existing literature and justies
INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH 3
its contributions. The problem denition based on a
retail store is developed in Section 3. Section 4proposes
asolutionframework.Section5presents the ndings
and analysis. Section 6presents research implications in
three parts: theoretical, methodological, and managerial.
Section 7provides the conclusions and limitations and
suggests future research directions.
2. Literature review
In this section, we present the review on four distinct
areas: (1) Supply chain and logistics management in the
industry 4.0 era, (2) Inventory distortion and predic-
tions, (3) Dynamic applications of AI addressing inven-
tory distortion and predictions (4) NCAI applications on
inventory distortion and predictions.
2.1. Supply chain and logistics management in
industry 4.0 era
Data-driven technologies have become the core of Indus-
try 4.0 supply chains (Batta et al. 2021). According to
Deepu and Ravi (2021), Inter-Organizational Informa-
tion Systems (IOIS) can plan, coordinate, collaborate,
and integrate supply chains to gain a competitive advan-
tage. Sarkar and Shankar (2021) developed an eight-level
hierarchy framework for Industry 4.0 implementation.
The study used ‘Total Interpretive Structural Modelling’
(TISM) and stakeholder theory to devise an algorithm to
plan routes and schedules in port logistics, signicantly
cutting distribution costs and CO2 emissions.
According to Grover, Kar, and Dwivedi (2020), AI-
enabled systems will strengthen decision-making across
the supply chain (Awan et al. 2021). Utilising AI in supply
chains can help managers boost organisational eciency,
quality,andcustomersatisfaction(Pournaderetal.2021).
However, in the initial stage, organisations face many
challenges in AI implementation (Chatterjee, Ghosh,
and Chaudhuri 2020). The synchronisation between AI
and employees is necessary as both rely on each other,
and the success of AI lies in mutual understanding (Siju
and Hasan 2022). Moreover, the speed of innovation evo-
lution and adoption, lack of clarity, and delay in taking
essential managerial and strategic decisions for adopt-
ing data-driven technologies require further research in
supply chain management (Enholm et al. 2021). A sig-
nicant supply chain issue, namely inventory distortion,
remains unsolved, resulting in lost revenues. Innovation
is vital to democratising AI and reducing inventory dis-
tortion (Jauhar et al. 2021b; Jauhar, Zolfagharinia, and
Amin 2022b;Pratapetal.2022b). Figure 1shows how
supply chain players presently deal with uncertainty and
how they should deal with it.
The industry reports reveal that inventory distortion
is increasing signicantly (IHL Group 2021;Statista2A
2022;Statista2B2022). The framework (gure 1)shows
that manufacturers often make mistakes in procuring raw
materials. They may procure large quantities of a low-
demand item and smaller amounts of a high-demand
item creating distortions. Over-production at a time of
less demand creates the need for more space and fur-
thers other problems. Moreover, the low production rate
during high-demand periods leads to customer dissatis-
faction. Retailers also face similar issues, often making
mistakes in inventory procurement (Kumar and Shukla
2022;Rajetal.2022).
Retailers may procure large quantities of low-cus-
tomer-demand products and stock smaller amounts for
high-customer-demand products. The decisions may
lead to lost sales, low customer satisfaction, high holding
costs, operational costs, and obsolete goods (Fitzsimons
2000;San-Joséetal.2017;Shrouty2019,Francia-Arias
et al. 2020). There is a need for technology to assist in
solving these issues. NCAI oers predictive solutions to
addresstheproblemsmentionedabove.Itcangiveaccu-
rate predictive analysis for future procurement, reduc-
ing the uncertainty in the supply chain and leading to
robust decision-making. NCAI can solve all these issues.
However, the present study focuses on retailers xing
inventory problems (OS and OOS).
2.2. Inventory distortion and predictions
The retail and manufacturing sectors are transforming
by adopting cyber-physical-production systems and the
Internet of things (IoT) technologies. For a smooth tran-
sition, a high priority is to resolve issues using contempo-
rary technologies to encourage their continued growth.
Even though retail is a high revenue-generating indus-
try,retailcompaniesandwholesalersreportthehigh-
est quarterly sales declines, resulting in increased debt.
High inventories are more expensive for businesses. Con-
sequently, stock problems occur with low inventories
and inaccurate information, with the companies fre-
quently struggling to meet dynamic consumer require-
ments (Kumar and Shukla 2022). As a result, failure to
determine demand is a signicant issue, resulting in a
loss of customer satisfaction (Fitzsimons 2000; San-José
et al. 2017). As a result, eective operations management
requires careful planning and control of the right amount
of inventory. Numerous research studies have examined
inventory optimisation but have given a lack of consid-
eration to the risk associated with inventory inaccuracy,
which results in stock-outs. Companies argue that OOS
occursduetosuppliersandweather,whichresultsincon-
sumer complaints (San José Nieto et al. 2022). Companies
4 S. KUMAR ET AL.
Figure 1. NCAI in the supply chain solves three significant issues through one solution.
are unaware of the consequences of stock-outs, which
can devastate internal processes and result in market
share loss (Kumar and Shukla 2022). For PT Combiphar,
a pharmaceutical manufacturing company, the products
gooutofstockatleastonceamonth,leadingtomany
customer complaints and criticism, aecting sales and
revenue (San-José et al. 2017; Mekel, Anantadjaya, and
Lahindah 2014).
In cases of excessive inventory, it will cost the retail-
ers more in warehouse rent, maintenance, employee
wages, and loss from obsolete goods (Kumar and Shukla
2022). Based on the research by Pani and Kar (2011),
the importance of product quality is the highest in the
decision-making process for an e-procurement scenario.
Asaresult,ahighstocklevelincreasesoperationalcosts,
whereas a low or no stock level results in lost sales and
customer dissatisfaction. A stock surplus can maximise
customer satisfaction, but the holding costs result in
revenue loss (Shrouty 2019). Kmavi EIRL, a Peruvian
retail company, acquires and sells ve dierent products
(collectibles, books and magazines, leisure goods, acces-
sories, and home goods). Due to their lower inventory
turnover, they are forced to sell goods at bulk discounts
to avoid becoming obsolete (Francia-Arias et al. 2020).
Table 1details the methodologies used in previous stud-
ies to address inventory distortion.
Data-driven inventory prediction models can assist
supply chain managers in decision-making. Inven-
tory forecasting optimises inventory across the supply
chain, increasing retailers’ nancial eciency (McKin-
sey&Company2020). Inventory replenishment refers
to the stock required to cover sales based on inven-
tory objectives, supply, and demand. As a result,
the need for precise forecasting is increasing. How-
ever, AI will become more accessible to businesses
andindividualswithreducedornocodingrequire-
ments. Inventory decision-making is contingent upon
demand planning, which can be achieved by analyz-
ing historical company data, addressing seasonality, and
existing logistical infrastructure (San José Nieto et al.
2022). Accurate forecasting ensures that businesses have
optimal inventory to meet customer orders and not
result in OS or OOS (San-José et al. 2017;Hamilton
2018).
Retailers should mitigate potential losses by care-
fully considering stock issues and developing plans for
future sales periods. Through AI-enabled precise pre-
dictions, businesses can strike a balance between invest-
ing in stock and maintaining an adequate inventory to
meet consumer requirements and the company’s nan-
cial goals (Awan et al. 2021; Kumar and Shukla 2022).
AI has already unfolded its potential in almost all sectors
INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH 5
Tab le 1. Previously used methodologies in solving inventory
distortion.
Authors Methodologies
Aburto and Weber
(2007)
Used hybrid intelligent forecasting
developed from ARIMA-type approaches
and MLP-type neural networks on system
demand forecasting that helps improve
the supply chain and efficiently manage
the retail industry.
Hachicha (2011) Deployment of meta-modeling simulation
meeting with the lot-sizing problem in the
supply chain
Paul and Azeem (2011) Artificial neural network-based model in
the determination of inventory level
of finished goods in a manufacturing
industry
He (2013) convergent BP neural network model
for inventory level prediction of an
automotive parts company
Mekel, Anantadjaya,
and Lahindah (2014)
Utilization of Demand forecasting using the
Double Exponential Smoothing method in
Combiphar pharmaceutical company to
reduce OOs and OS
Sustrova (2016) AI model for forecasting in f inventory
optimisation in lot-sizing problems.
Rumetna, Renny, and
Lina (2020)
The waterfall method for developing the
system and the exponential smoothing
method for forecasting to reduce OOS in
Samsung Partner Plaza
Rizqi and Khairunisa
(2020)
Material classification through ABC Analysis,
determination of stock levels through the
min-max method, and evaluation through
Monte Carlo simulation.
Francia-Arias et al.
(2020)
Executed Change Management, SCOR Model,
ABC analysis, and Online Sales Channel in
Kmavi E.I.RL., a small retail company
(Jauhar, Amin, and Zolfagharinia 2021a;Jauharetal.
2022a;Murugesan,Jauhar,andSequeira2021;Pratap
et al. 2022a).
2.3. Dynamic applications of AI in addressing
inventory distortion and predictions
We know that articial intelligence’s ubiquitous nature
enables its application in nearly every industry, includ-
ing nance, marketing, consulting, accounting, manu-
facturing, e-commerce, automotive, robotics, agriculture,
education, gaming, entertainment, transportation, logis-
tics, supply chain, social media, government agencies,
urbanisation, environmental sustainability, virtual assis-
tance, and astute decision-making (Chien et al. 2020;
Rodríguez-Espíndola et al. 2020;Wuestetal.2020;Oost-
huizen et al., 2021;Leeetal.2018;Lee,Singh,andAzam-
far 2019;Rizviandcolleagues2021). In today’s competi-
tive market, AI has established itself as an academic and
scientic discipline worthy of exploration. Table 2shows
how AI has been used in the past to make supply chains
more ecient and resilient.
In addition, AI systems have become an integral
part of digital transformation, signicantly inuencing
human decision-making (Duan, Edwards, and Dwivedi
2019). Evolving industry 4.0 technologies have increased
agility, productivity, visualisation, interoperability, accu-
racy, provenance, time management, resource manage-
ment, waste reduction, and sustainable performance
(Belhadi et al. 2021;Olanetal.2021). Using the
enterprise’s internal and external data (Mishra et al.
2018;Ben-Daya,Hassini,andBahroun2019), intelligent,
resilient, and sustainable industry 4.0 technologies can
easily translate customer needs into product develop-
ment (Ben-Daya, Hassini, and Bahroun 2019).
High-quality data can be collected using AI professional
skills, algorithms, and models, resulting in signicant
Tab le 2. AI in improving OM efficiency and resiliency.
Authors Application Technology-Industry 4.0 Industry Country
Lee et al. (2018) Industrial AI in manufacturing systems Manufacturing
Magistretti, Dell’Era, and
Petruzzelli (2019)
Implementation of AI using the IBM Watson product design
Lee, Singh, and Azamfar
(2019)
Application of Artificial industrial intelligence in
lighthouse manufacturing companies
Light-house-Manufacturing USA
Duan, Edwards, and
Dwivedi (2019)
Application of AI in decision-making with the help of
big data
Wang et al. (2020) AI and additive manufacturing in product design Additive – 3D-Manufacturing China
Oosthuizen et al. (2021) AI in retail for enabled value chain improvement Retail, supply chain
Woo (2020) Covid-19 contact tracing by Unqork software Covid Mitigation USA
Wuestetal.(2020) AI to mitigate the impact of the covid-19 pandemic on
Manufacturing and SC
Manufacturing, Circular
economy, and Supply chain
Networks
USA
Burström et al. (2021) AI in incumbents Manufacturing Sweden
Bag et al. (2021) Big data analytics powered AI in the role of institutional
pressures and resources
Manufacturing and circular
economy
South Africa
Rizvi et al. (2021) AI and industry 4.0 technologies in manufacturing Manufacturing India
Haefner et al. (2021) AI in innovation management
Sucharitha and Chary
(2021)
AI to predict the effect of covid-19 using Prophet
Model.
Covid Mitigation India
How et al. (2021) Low code/no code AI for social good Social good-worldbank
Current study Application of NCAI in companies for Resilient
Operations Management
Retail and India Manufacturing
6 S. KUMAR ET AL.
energy savings (Chien et al. 2020;Rizvietal.2021;
Rodrguez-Espndola et al., 2020). Prediction-enabled
inventory management enables accurate forecasting and
inventory optimisation throughout the supply chain,
increasing revenue and cost savings. As retailers want
to ensure there are enough stocks for everyone, they
need to keep improving their sales forecasts to meet the
needs of potential customers through machine learn-
ing and deep learning. Better inventory management
ensuresahigherlevelofcustomersatisfactionwithout
investing in unnecessarily large inventories that result in
increased operating expenses (Kumar and Shukla 2022;
Shrouty 2019). Machine learning helped Otto Group,
an e-commerce retailer, cut its alarming stocking rate
by80%,makingitmoreprotableandbetterableto
adapt to changes in the market (Oosthuizen et al.,
2021).
The researchers asserted that traditional chains are
inecient and are on the verge of extinction due to
AI-enabled newcomers delivering superior value to con-
sumers. Retailers need to get smaller, more agile, and
more creative to stay alive in this very competitive
and always-changing world. According to Burström
et al. (2021), successfully implementing, integrating, and
maintaining AI technology within incumbents can trans-
form the enterprise within the industrial ecosystem,
resulting in value creation, delivery, and capture.
The AI-inspired business model can oer a competi-
tive edge and cost advantage through robust analysis of
algorithm-generated reports, as well as identify short-
ages and bottlenecks before they could occur in the
real world, avoiding signicant disruptions and provid-
ing the benets of increased connectivity, transparency,
and visibility (Baryannis et al. 2019; Pournader et al.
2021). The classication framework for AI’s application
in the eld of supply chain management is shown in
Table 3.
Industry 4.0 has resulted in extraordinary disruptions
(Queiroz et al. 2020). The rapid pace of this revolution
hasforcedustorethinksectorsthatrelyoninnovation.
Figure 2shows the evolution of dierent technologies
under the industry 4.0 revolution.
Tab le 3. Classification of current research for AI application.
Authors Objectives Highlights
Lee et al. (2018) Industrial AI for industry 4.0-based manufacturing
systems, the perspective of the industry 4.0
paradigm.
An AI-based ecosystem can meet unmet needs such as Self-aware,
Self-compare, Self-predict, Self-optimize, and resilience, also
including four leading enabling technologies, Data Technology (DT),
Analytic Technology (AT), Platform Technology (PT), and Operations
Technology (OT). These are enablers for success in Connection,
Conversion, Cyber, Cognition, Configuration, or 5C.
Lee, Singh, and
Azamfar (2019)
For sustainable and smart manufacturing practices
and circular economy capabilities, value creation for
lighthouse manufacturing
New value-creation opportunities for smart manufacturing in a
dynamic and uncertain environment achieving the 3W’s in smart
manufacturing: Work-reduction, Waste-reduction, and Worry-free
manufacturing.
Magistretti, Dell’Era,
and Petruzzelli
(2019)
Digitalization, internal development, external
collaboration
AI-enabled DBT reshapes product development and improvement in
product development processes.
Duan, Edwards, and
Dwivedi (2019)
Prepositions for future research on AI AI systems are becoming an embedded element of digital systems, and
more specifically, they profoundly impact human decision-making.
Bag et al. (2021) Sustainable manufacturing practices and circular economy capabilities
Rizvi et al. (2021) To know the current status of AI in manufacturing
companies in India
Reducing a significant operational expense in human error is
completely nullified, resulting in substantial energy savings.
Haefner et al. (2021) To know the effect of the covid pandemic on various
industries, mitigating the impact of covid-19 with
AI-based digital technologies on manufacturing and
supply networks
Recognition, discovery, creation, and development of various
innovative ideas, opportunities, and solutions through AI, using
across optimisation of battery components and solar cells,
discovering new materials in pharmaceutical research and
development, speeding up the process of protein engineering,
identifying treatments for disease
Sucharitha and Chary
(2021)
Predicting the number of confirmed cases, deaths
recovering patients in India through AI
To validate patients from afar to save time and safety precautions
effectively, to evaluate disease, to modify surveillance systems and
observational data simulations besides epidemiology, to recognise
and optimize treatment strategies in government health policies
Burström et al. (2021) Value creation, delivery and capturing, business
transformation in the industrial ecosystem
Obtaining competitive advantage, bringing cheaper and more
innovative solutions to the market through can auto-generate
reports
Current study Make operations management resilient by reducing
inventory distortion (Out-of-Stocks and Over-stocks,
Less production and overproduction) via deploying
newer technology-NCAI determine the Inventory
(/Production) Balance Point.
Creation of AI algorithmic model without writing one line of code,
practical application of NCAI in solving real-world challenges,
solution of unresolved enormous problem inventory distortion
through creating machine learning model making via a custom-
tailored best algorithm based on data’s properties for streamlining
picking, packing and replenishment quantities
INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH 7
Figure 2. Technological evolution of Industry 4.0 technologies.
2.4. NCAI applications on inventory distortion and
predictions
While traditional AI methods require extensive manual
work, data engineering, and a high level of statistical
prociency, NCAI makes predictions simple, aordable,
and accessible to anyone, even non-technical sta. Users
can easily create predictive models from any row of data
anddeploytheminanysetting.ThisNCAI-drivennew
business model can transform a traditional production
processintoadigitalone,easingtheimplementationpro-
cedure in the industry and boosting industrial operators’
knowledge (Redchuk and Walas Mateo 2022). NCAI is
widely used in the nancial sector through its workows
to improve security, sort out the customer experience,
and uncover how credit card fraud costs businesses bil-
lions of dollars each year. While multinational banks and
other large nancial institutions employ expert AI teams
to develop fraud prediction models, these eorts are pro-
hibitively expensive and time-consuming for thousands
of SMEs. NCAI can use fraud detection models to send
SMS messages to customers when a fraudulent transac-
tion is found (Akkio 2021).
NCAI is used eectively in healthcare by developing
machine learning models tailored to their specic needs,
from patient care to billing. This trained model can scan
8 S. KUMAR ET AL.
thousandsofimagesinseconds,identifyingmalignancy
growth signs and saving doctors valuable time (Tariq
2021). Additionally, marketers and business owners use
NCAI to transform raw data into actionable insights to
forecast accurate revenue ow and optimise business pro-
cesses. Manufacturing and supply chain companies can
signicantly use NCAI to achieve: consistency in quality,
lower costs, improve customer relationship management,
and increase revenue (Redchuk and Walas Mateo 2022).
NCAI requires data and selects a column for predic-
tions based on the user’s requirements. NCAI creates
a custom-tailored machine-learning algorithm based on
the data’s properties and provides an accurate predic-
tion report in seconds. By deploying NCAI, businesses
of all sizes and sectors can incorporate robust insights
into their daily operations, aiding them in achieving
their business objectives (Tariq 2021). NCAI can also be
applied to inventory replenishment.
Numerous NCAI platforms are available, enabling
individuals without programming skills to develop algo-
rithmic reports rapidly. However, proper data analy-
sis skills are required (García-Ortiz and Sánchez-Viteri
2021). In this research, we address the inventory distor-
tion problem using NCAI technology. NCAI identies
data properties based on parameters such as the num-
ber of rows and columns, features (discrete, continuous,
date/time, etc.), variance, sparseness, and 35 +additional
properties. Second, it will examine thousands of dierent
algorithms and choose the best ve based on the proper-
ties of the data properties. NCAI is capable of predicting
both discrete and continuous target variables. Figure 3
illustrates the decision-making process using historical
data through the deployment of NCAI.
NCAI creates unique hyper-parameters that result in
customised algorithms for our data. To be sure, hyper-
parameters have the potential to aid the algorithm in
producing accurate predictions. It takes the ve best
algorithms shortlisted (based on higher accuracy) and
numerous hyper-parameters (dierent combinations of
settings), which will go across 10,000 +dierent algo-
rithms.Eachalgorithmistestedforaccuracy,andthebest
ones are executed for prediction in less than a minute
(Obviously.ai.6th). Each industry has variables and val-
ues that NCAI users must consider when forecasting
demand. These are used to clean the data. NCAI chooses
thebestalgorithmbasedonthisdatatomakeaccurate
predictions and have a low RMSE (Root Mean Square
Error), which leads to excellent results.
NCAI is a relatively new domain for researchers, evi-
denced by the scarcity of research papers on NCAI
applications. According to How et al. (2021), individuals
not in the information technology or computer science
domains can also generate actionable predictive insights
using low-code or no-code AI. It enables rapid devel-
opment of predictive simulations via drag-and-drop,
allowing for complete customisation of variables and AI
Figure 3. The NCAI-based decision-making framework.
INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH 9
algorithms, creating models for global citizenship educa-
tion, social good, sustainable development, malnutrition
mitigation, and nancial involvement. The NCAI tech-
nology has the potential to democratise AI by augment-
ing human-centric insights with user-friendly predictive
modeling.
Garca-Ortiz et al. (2021) proposed using a no-code or
low-code AI tool to facilitate student learning and educa-
tion. They used WEKA (a collection of machine learning
algorithms for data preparation, classication, regression,
clustering, association rule mining, and visualisation) to
identify student learning issues. Iyer et al. (2021)applied
the Trinity platform for spatial datasets with image rep-
resentations to various cases in the geospatial domain.
Trinity is Apple’s NCAI platform, enabling non-technical
domain experts to experiment independently with dier-
ent signals or datasets to solve problems. It can potentially
standardise the formulation of disparate issues for ease of
consumption.
Redchuk and Walas Mateo (2022)implementedthe
Canvass AI-NCAI platform in a steel manufacturing
company and made forecasting, anomaly detection, opti-
misation, simulation, failure prediction, and defective
part prediction 12 times faster than other solutions and
approaches. They could shorten the implementation time
from several months to a few months. By executing
insights into the operations, they could increase the qual-
itylevel,optimiseenergysupplytoplantdemand,opti-
mise fuel sources and fuel costs, and reduce waste in
raw material processing, steelmaking and casting, and
hot rolling, leading to fast sales cycles and obtaining a
competitive advantage. Table 4compares the NCAI to
conventional machine learning on selected parameters.
However, each solution has a set of inherent con-
straints in the platform’s creature. While NCAI is easy
and fast to use, it also has some boundary conditions
andlimitations.Here,‘nocode’doesn’tmean‘noneed
for technical knowledge at all.’ It still requires funda-
mental knowledge, such as the principles of A.I., the
datasets, the specic applications, and integration with
other technologies, to eectively utilise these platforms.
While NCAI is gaining attention and democratising
AI, we need to ensure that the full potential of AI is
realised (Redchuk and Walas Mateo 2022). Managers will
have to learn how to deploy these platforms optimally. So
they can get the best possible outcomes. Another notable
thing is that companies must know their goals from
NCAI. Because some allures, such as the development
of dierent use cases, lower implementation charges, etc.,
may inuence them to buy it, but they may not be able to
get the model according to their requirements. For exam-
ple, some platforms are eective for image recognition,
whereas some other tools are adequate for analytics.
Tab le 4. Comparison between Conventional ML and NCAI
through different parameters.
Parameters Conventional ML NCAI
Installation team Team of data scientists
required
Only a single employee
required
Skills Excellent coding skills,
long-term coding
experience
No coding skills
required
Model-making
duration
4Months 1min
(For the data of 1000
Rows and five
columns)
(Time increases
with increasing
numbers of rows
and columns)
(Time increases
with increasing
numbers of rows
and columns)
Cost >100 man-days <2 man-days
Manual model
selection
Mandatory to go with a
particular model
Faci lity to go wi th
different models
White label Embed This model is not
accessible to others
This model can be
easily accessible by
permitted viewers
Automatic Model
improvement
No facility for
Automatic Model
improvement
Automatic model
improvement is
possible
Integrations Cannot integrate with
other company tools
and apps
Easy to integrate
with several other
company tools and
apps
Risk of Investment Too high Too low
(Due to the cost of
hiring a technical
employee for
months)
(Due to anytime
cancelation without
extra charges)
That’s why it is better to do a need assessment before
using NCAI. Furthermore, these platforms are based on
models that are easy to grasp and execute but oer less
exibility.
Moreover, the NCAI platforms have limited function-
ality as the solutions aren’t entirely customised (García-
Ortiz and Sánchez-Viteri 2021). Generally, they can cover
a specic problem, and running complicated models is
challenging.Employeesmayhavetoconsulttheplatform
before running complex models restricting the perfor-
mance of complicated processes. However, the success of
NCAI depends on its use case and level of understanding.
Another important aspect is security. Some platforms
may not be successful in creating access protocols. It is
a concern for companies where security is the priority. It
wouldbebettertobecautiousinexaminingthetermsand
conditions to understand how and where the data will be
processed (SuperAnnotate 2021).
3. Problem identification
As the population and demand for products grow, the
need for robust supply chains with substantial inven-
tory management increases. Inventory management’s
primary objective is to acquire products and manage
inventory ow according to supply and demand to
10 S. KUMAR ET AL.
Figure 4. Worldwide loss due to inventory distortion (OS and OOS) issue.
meet customer requirements and the company’s nan-
cial objectives (Oosthuizen et al. 2021). Massive revenue
loss is also becoming a challenging issue due to the
enormous revenue loss. Numerous supply chains face
huge losses due to inconsistent management and retailer
conservatism. Innumerable businesses frequently make
a mistake in product replenishment decisions. Inven-
tory distortion occurs as a result of these mismanage-
ment issues. Inventory distortion has become a curse
for retailers, sucking up a signicant portion of their
prot. Figure 4showstheglobalandregionallossof
inventory distortion in ve major regions: Latin America,
North America, Asia/Pacic, Europe, the Middle East,
and Africa (EMEA).
According to IHL Group’s 2015 study, retailers world-
wide lose $1.1 trillion in revenue due to OS and OOS
issues. OS accounts for 3.2 percent of the average retail
seller’s revenue loss, while OOS accounts for 4.1 per-
cent. The loss from OS is estimated to cost retailers
$471.9 billion and OOS $634.10 billion annually in
North America alone. Over time, inventory distortion
has become an ever-increasing challenge, exacerbated
by the COVID-19 pandemic. Retailers face signicant
inventory management challenges as the visibility of
goods increases and decreases by a considerable amount.
As a result, billions of dollars eroded the entire sector’s
value.
Inadequate stock availability results in lost sales and
nancial corrosion, while large stock surpluses result in
heavy discounting of products and losses from outdated
and expired-dated stock. Stock surplus issues generate
more revenue losses on average than poor stock issues.
People often think that this is a problem unique to the
fashion industry. Nonetheless, e-commerce has become
a primary trading focus for many brick-and-mortar
retailers, particularly during the Covid period. Due to
the COVID-19 pandemic, administering a stock pool
across multiple chains and maintaining optimal avail-
ability in each warehouse has become more challenging
than ever.
According to a report by international logistics com-
pany Advanced Supply Chain Group (ASCG), retailer
prot warnings (lower than expected prot) between
June 2018 and January 2020 account for an average of
22% of (reduced) retailers’ total market value totaling
£1.6 billion. Based on the initial pandemic situation, it
was estimated that goods worth USD 8 trillion were
placed on hold from sale due to a halt in operations,
demonstrating the severe eect of overstocking. Manu-
facturers in the United Kingdom alone have incurred a
loss of nearly $88 billion due to stock management issues
caused by the COVID disruption. The data demonstrate
unequivocally that the world is facing a massive prob-
lem. In 2019, e-commerce retail sales increased to 14%
of total retail sales globally, and for only e-commerce,
inventory distortion has resulted in a $150 billion loss in
revenue. E-commerce sales are expected to account for
21.8 percent of total sales by 2024 (InVia Robotics 2020).
These complications arise as a result of the primary issue
– inventory replenishment. Inventory replenishment is
producing, moving, or obtaining necessary stock in the
warehouse/supply chain to meet customer demand.
However, according to the IHL, inventory distortion
(OOS & OS) amounted to 1.8 trillion dollars worldwide
in 2020, exceeding Canada’s GDP (Gross Domestic Prod-
uct). Even though the IT spending was higher in the
North American region, with more advanced solutions,
the region is expected to lose $42.2 billion in revenue
INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH 11
due to planning issues with overstocking. The inven-
tory distortion problem in EMEA (Europe, Middle East,
and Africa) costs $501 billion, or about 10.4 percent of
Schwarz’s retail sales or the combined revenues of Lidl,
Aldi, Tesco, Edeka, Ahold, and Auchan. If this signicant
issue were resolved, the same stores’ sales could increase
by 10.3 percent.
While sales are the primary focus of retail businesses,
optimal inventory management is a critical sales com-
ponent. Inventory management is a crucial function of
manufacturing and commercial enterprises, signicantly
impacting their overall performance. In today’s highly
competitive environment, coupled with high consumer
expectations for high-quality products, inventory dis-
tortion results in revenue loss for GEMA Novelty retail
stores. Managing inventory is a job that covers every part
of the store’s work, from buying and storing goods to
selling them.
One of the primary constraints that a business may
face is the delayed sale of products due to low sales,
which increases stock levels. OS and OOS frequently
occur when controlling inventory, which makes deter-
mining the optimal point extremely dicult. It is crit-
ical to resolve this issue. For GEMA, OS resulted in a
lot-sizing problem, increased holding costs, substandard
service, high rental warehouse charges, employee salaries,
maintenance costs, a high probability of damage, loss
from expired goods, and decreased prot. While OOS
has caused high shortage costs, insucient inventory to
meet customer demand, loss of market competitiveness,
reduced revenue, customers shopping at other retailers,
and decreased customer loyalty and satisfaction, resolv-
ingthisissueiscritical.Thisisarealisticapproach,con-
sidering general inventory distortion issues and eects.
4. Solution approach
Figure 5shows the owchart of how we did our research.
It shows how we chose a domain, a problem, a severity
level, and how we chose a method and a technology, how
weprocesseddata,andhowwechoseatargetcolumn.
Finally, we looked at the model and used the excellent
results to see the future.
4.1. Domain selection and Problem identication
He (2013) states that inventory is a signicant aspect
of the retail world and is a hot topic of discussion and
debate. There is a considerable research gap regarding the
use of AI to solve inventory management problems. To
increase inventory management eciency and close this
enormous gap, we researched plans to improve inven-
tory management. GEMA frequently encounters OS and
OOS issues. Excessive inventory investment consumes
working capital and increases the likelihood of prod-
ucts becoming low-valued or obsolete inventory, whereas
poor investment results in OOS. These issues can result
in delayed delivery and customer dissatisfaction, and we
have already seen the severity of these issues. We chose
inventory distortion as our research problem, consider-
ing these considerations.
4.2. Best methodology for inventory replenishment
To meet today’s requirements, we discovered that inven-
tory predictions provide a more accurate method for
determining the quantity of inventory replenishment. It
is the only robust solution for striking the right balance
between prot and customer satisfaction by balancing
adequate inventory levels against customer demand. It
performs better in a competitive market using technol-
ogy rather than traditional statistical methods. The loss
data unequivocally demonstrates the absence of a robust
processforoptimalinventoryplanning.
4.3. Selection of AI for predictions
Spreadsheets provide forecasting functionality, but they
are limited to small businesses with a limited number of
products to sell and are prone to human error. Numerous
Enterprise Resource Planning (ERP) software packages
include forecasting capabilities. However, appalling data
demonstratethatthereisstillaneedforrobustinven-
tory forecasting technology and improved statistical
methods.
A precise inventory forecast is priceless, even more
so in today’s world of dramatic changes in consumer
demand. Retailers should leverage the AI algorithmic
model’s advantage to obtain accurate forecasts to increase
their inventory productivity and maintain adequate lev-
els. However, applying AI to inventory management
problems is a signicant research gap. As a result, we
chose AI to address inventory distortion.
4.3.1. Go with newer and faster technology
We compared innovations in detail to make a more
informed choice. As previously stated, AI has some lim-
itations that prevent businesses from gaining insights.
These constraints can be overcome using cutting-edge
technology such as NCAI, enabling companies to gain
valuable insights without writing a single line of code.
Current NCAI deployments in inventory management
using data sources can drive lower stock levels and ana-
lyzepredicteddemandtoprocureandkeeppopularitems
stocked on shelves in assortments, thus beneting from
lower working capital requirements.
12 S. KUMAR ET AL.
Figure 5. Flowchart of Research Approach.
4.4. Selection of best NCAI tool
Tobeginwith,wereviewedtheNCAIplatforms.We
identied AI as the best tool to use according to our
requirements for inventory distortion, particularly for
non-technical users. It is simple to use and can pro-
vide valuable insights in minutes without requiring the
data scientist to write any code, signicantly reducing
the hours and eort needed and making data science
eortless. NCAI provides a drag-and-drop solution that
automates the entire data science process, removing the
requirement for consulting on data science projects with-
out coding skills (García-Ortiz and Sánchez-Viteri 2021).
This machine learning approach is based on a consistent,
repeatable time-series classication method that excels in
the presence of incomplete and irregular data and leads
to real-time actions. It can also overcome some of the
challenges of real-time data analytics. This method has
much potential in traditional industries that have not yet
reached the full potential of industry 4.0 and, in most
cases, have just started learning about data analytics and
machine learning (Redchuk and Walas Mateo 2022).
4.5. Data collection
The data for the research came from various sources,
including interviews, the analysis of research papers, and
the examination of newspaper articles. In this study, we
INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH 13
usedsalesdatafromKaggle’sonlinestoreandtheprevi-
ous year’s novelty retail sales data CSV le. We looked at
how the NCAI could be used to make inventory replen-
ishment decisions.
4.6. Data uploading and selection of target column
and algorithm
Theselectionofthetargetcolumniscriticaltothis
process because the tool will display the eect of other
factors(columns)onthetargetedcolumn.Thetargetcol-
umn should be chosen based on future requirements.
Our requirements specify the quantities an organisation’s
stock should be maintained. As a result, we’ve selected
the quantity column, which displays the total number of
sales on a particular date. We did this using the 14-day
free trial of Obviously AI. The tool determines the opti-
mal and most custom-tailored algorithm for our model
based on the properties of our data. It customised a ran-
dom forest regressor algorithm based on the properties
of our data.
5. Result and analysis
An in-depth analysis of the results and analysis of this
study has been covered in the following eight sections.
5.1. Attributive information of data
First, we used the recent two years’ sales data to make pre-
dictions. We uploaded this CSV data le and chose the
quantitycolumnasthepredictiontarget.Salesdatacon-
sisted of 19873 rows and 11 columns containing detailed
information about sales transactions. The invoice date
column indicates the transaction date. The month col-
umn held the rst quarter’s months, while the year col-
umn represented 2018 and 2019. The description column
contains the product names, while the quantity column
contains the total number of product sales on a particu-
lar date. The start-end of the month or quarter provides
information on the date of the transaction and whether
the month started or ended. Following that, the tool ini-
tialises the server and performs data cleaning. NCAI
sorted the entire data set based on numerical values, texts,
date, and time to create a correlation between attributes
before using the best algorithm. Table 5displays our data
attributes and type, which it handles while creating the
prediction columns in our dataset.
This process took less than 30 s compared to the
months it typically takes when using the coding app-
roach. Models created using conventional methods are
highly time-consuming, expensive, and labour-intensive
to produce. For example, traditional machine learning
Tab le 5. Attributes of the data set assigned to create the model.
ID Number
Invoice Date Date/Time
Description Text
Month Number
Month-Start Text
Month-End Text
Quarter-Start Text
Quarter-End Text
Yea r-S tar t Text
Year-End Text
Quantity Number
takes four months to develop a predictive model for
1,000 rows and ve columns of data, which entails hir-
ing more than 100 person-days of expensive professional
programmers (Obviously.ai.6th). This tool takes less than
a minute to make the best predictions, and the monthly
charges are less than two person-days.
5.2. Analysis of the model
5.2.1. Model metrics
Table 6summarises model metrics based on the model’s
various measurable model metrics values. The root mean
squared error (RMSE) is considered the best measure
of accuracy, with the standard deviation of the residu-
als (distance of the data points from the regression line)
serving as the general-purpose model metric for numer-
ical predictions and R2acting as a goodness-of-t metric
for linear regression models. The RMSE is a commonly
used metric for estimating the rate of experimental out-
comes in forecasting and regression analysis. The model
with the lowest number of errors (RMSE) has the most
signicant inuence on the outcome or is the most rel-
evant (Nikoli et al. 2018). Superior results are indicated
by a lower RMSE and a higher R2score (greater than or
equal to 0.75). The low MAE and MSE values suggest that
the proposed method is more accurate. The low error val-
uesindicatethatthemodelissignicantinpredictingthe
product quantity (Pap et al. 2022).
Tab le 6. Differenterrorvaluesandmetricsofourmodel.
Algorithm Name Random Forest Regressor
Training RMSE 17.6
Validation RMSE 17.96
Testing RMSE 18.15
Loss Function Least Squares Regression
Min Absolute Error 0
Max Absolute Error 344.3931
R2 Score 0.8143
Mean Absolute Error 37.0162
Mean Squared Log Error 0.0085
Mean Squared Error 329.4225
Root Mean Squared Percentage Error 2355693.864
Mean Absolute Percentage Error 299648.8109
14 S. KUMAR ET AL.
The NCAI identied random forest regressor (Hyper-
parameter Combination A) as the most suitable model
with an 18.15 root mean square error rate and an R2
score of 0.8143, indicating that the model is both valid
and ecient. Based on the analysis of the values of dier-
ent model metrics, we can say that the identied model
is an excellent and deployable solution for supply chain
inventory distortion-reduction models.
5.2.2. Random forest regressor
The random forest regressor is the best ensemble learn-
ing (combination predictions from multiple machine
learning algorithms). It eciently uses a combination of
numerous decision trees rather than individual decision
trees to determine the outcomes, resulting in more accu-
rate predictions than any other popular ensemble learn-
ing algorithm (Breiman 1996;Bakshi2020;Zhangetal.
2022). Furthermore, the random forest regressor creates
an ensemble of regression trees using disparate bootstrap
samplesofthetrainingdata.Severalresearchstudieshave
conrmed that the random forest regressor has a high
noise tolerance and is eective at avoiding over-tting.
It also lowers the average error of the tree used (Zhang
et al. 2014).Thestructureoftherandomforestregressor
algorithm is depicted in Figure 6, which generates a large
number of trees. Instead of just one decision tree, it takes
the average of all of them to make accurate predictions
(Redchuk & Walas Mateo 2022).
5.2.3. Performance Evaluation of the Proposed
Method
We can re-run the predictive model using other algo-
rithmic models that were shortlisted. For predictions,
the NCAI platform presented a choice of ve addi-
tional models that used another algorithm based on the
data properties. The algorithm’s loss function, RMSE,
R2,MAE,andMSE(MeanSquaredError)arelistedin
Table 7.Wecanseethattherandomforestregressor
model outperforms the other ve models on all metrics.
The random forest regressor (hyper-parameter combina-
tion B) produces a model with a root mean square error
of 18.69 and an R2of 0.8143, which is less accurate than
ourmodelbutstillvalidforanalysis.
The NCAI generated various models using the best
custom-tailored algorithms, including gradient boosting
regressor (hyper-parameter combination A), gradient
Figure 6. Random forest regressor predictions-creation procedure.
Tab le 7. Top-five model metrics.
Model Algorithm Loss Function RMSE R2 MAE MSE
Random Forest Regressor (Hyper-parameter
Combination B)
Random Forest
Regressor
Least Absolute
Deviation
18.69 0.8143 37.0162 349.3161
Gradient Boosting Regressor (Hyper-parameter
Combination A)
Gradient Boosting
Regressor
Least Absolute
Deviation
39.85 0.6046 81.2896 1588.0225
Gradient Boosting Regressor (Hyper-parameter
Combination B)
Gradient Boosting
Regressor
Least Squares
Regression
41.04 0.6046 416.247 1684.2816
Elastic Net (Hyper-parameter Combination A) Elastic Net Elastic Net 68.95 0.0417 140.6579 4754.1025
Elastic Net (Hyper-parameter Combination B) Elastic Net Elastic Net 71.02 0.0417 140.6579 5043.8404
INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH 15
boosting regressor (hyper-parameter combination B),
elastic net (hyper-parameter combination A), and elas-
tic net (hyper-parameter combination B), with RMSE of
39.85, 41.04, 68.95, and 71.02, respectively, and R2scores
of 0.6046, 0.6046, and 0.0417, respectively. We found
that the random forest regressor model (hyper-parameter
combination A) and the random forest regressor model
(hyper-parameter combination B) were better suited for
inventory replenishment needs.
When we look at the gradient boosting regres-
sor (hyper-parameter combination A) and the gradient
boosting regressor (hyper-parameter combination B), it
provides a valid predictive model but at a lower level.
The elastic nets (hyper-parameter combinations A) and
elastic nets (hyper-parameter combinations B) are not
signicant predictive models because of the high RMSE
and low R2scores. The application of this precise ran-
dom forest regressor (Hyper-parameter Combination A)
model can signicantly reduce inventory distortion.
5.3. Validation
Validation was done to scientically judge the method-
ology by evaluating the model’s predictive performance
(Pham et al. 2019; Khozani et al. 2019). In this research,
we have validated the model based on several parame-
ters such as root mean square error (RMSE), the Mean
Absolute Error (MAE), R2, MSE, graphical analysis, scat-
terplot,MaxAbsoluteError,MinAbsoluteError,Mean
Squared Log Error, Mean Squared Error, Mean Abso-
lute Percentage Error, etc. Table 8depicts the study’s
validation metrics.
All these measures have given better values, which
supports the reliability of the study. For the random forest
Tab le 8. Validation metrics.
Training RMSE 17.6
Validation RMSE 17.96
Testing RMSE 18.15
Min Absolute Error 0
Max Absolute Error 344.3931
R2 Score 0.8143
Mean Absolute Error 37.0162
Mean Squared Log Error 0.0085
Mean Squared Error 329.4225
Root Mean Squared Percentage Error 2355693.864
Mean Absolute Percentage Error 299648.8109
regressor (Hyper-parameter Combination A) model, we
got the following results: Training RMSE-17.6, Vali-
dation RMSE-17.96, Testing RMSE-18.15, Min Abso-
lute Error-0, Max Absolute Error-344.3931, R2-0.8143,
Mean Absolute Error-37.0162, Mean Squared Error-
37.0162, Mean Squared Log Error-0.0085, Mean Squared
Error-329.4225, Root Mean Squared Percentage Error-
2355693.864, Mean Absolute Percentage
Error-299648.8109.
5.4. Graphical analysis
In Figure 7, the ‘x’ axis represents data points, and the ‘y’
axis represents quantity. The blue indicates the amount,
and the red indicates the predicted amount.
The graph illustrates the dierence between the actual
and predicted quantities through overlapping. Predicted
quantity overlaps actual quantity, meaning the amount
predicted covers the actual quantity. More red overlaps
with blue indicate that algorithms have covered more
data in making predictions. The graph demonstrates a
high degree of overlap, explaining that the random forest
regressor (Hyper-parameter Combination A) model cov-
ers enough data to make accurate predictions. Figure 8
shows model performances through a scatter graph.
The graph displays data points on the X-axis, quan-
tity through blue dots, and percent error through red
crossesontheY-axis.Thequantityandpercenterroris
plotted on the graph. If the percent error is prominent
on the gr