sustainability

Article
Predictive Analytics and Machine Learning for Real-Time
Supply Chain Risk Mitigation and Agility
Abeer Aljohani

Department of Computer Science, Applied College, Taibah University, Medina 42353, Saudi Arabia;
aahjohani@taibahu.edu.sa

Abstract: Supply chain agility has become a key success factor for businesses trying to handle up-
heavals and uncertainty in today’s quickly changing business environment. Proactive risk reduction
is essential for achieving this agility. To facilitate real-time risk prevention and improve agility, this
research study proposes an innovative strategy that makes use of machine learning as well as pre-
dictive analytics approaches. Traditional supply chain risk management frequently uses post-event
analysis as well as historical data, which restricts its ability to address real-time interruptions. This
research, on the other hand, promotes a futuristic methodology that uses predictive analytics to
foresee possible disruptions. Based on contextual and historical data, machine learning models can
be trained to find patterns and correlations as well as anomalies that point to imminent dangers.
Organizations can identify risks as they arise and take preventative measures by incorporating these
models into a real-time monitoring system. This study examines numerous predictive analytics
methods, showing how they can be used to spot supply chain risks. These methods include time
series analysis and anomaly detection as well as natural language processing. Additionally, risk
assessment models are continuously improved and optimized using machine learning algorithms,
assuring their accuracy and adaptability in changing contexts. This research clarifies the symbiotic
relationship among predictive analytics and machine learning as well as supply chain agility using a
synthesis of theoretical discourse and practical evidence. Case studies from various sectors highlight
the usefulness and advantages of the suggested strategy. The advantages of this novel technique
include improved risk visibility and quicker response times as well as the capacity to quickly modify
operations. The development of a holistic framework that incorporates predictive analytics and
Citation: Aljohani, A. Predictive
machine learning into risk management procedures, setting the path for real-time risk identification
Analytics and Machine Learning for
as well as mitigation, is one of the theoretical contributions. On the practical side, the case studies
Real-Time Supply Chain Risk
offered in this paper show the actual benefits as well as the adaptability of the proposed approach
Mitigation and Agility. Sustainability
2023, 15, 15088. https://doi.org/
across a wide range of businesses.
10.3390/su152015088
Keywords: machine learning; predictive analytics; real-time monitoring; supply chain agility; supply
Academic Editors: V. Ravi., Suresh
chain risk management
Subramoniam and Bijulal D.

Received: 20 August 2023

Revised: 8 October 2023
Accepted: 19 October 2023 1. Introduction
Published: 20 October 2023 The structure and operation of supply chains have grown more complex and inter-
connected in the modern setting of networked economies and global trade. However,
because of this increased complexity, supply networks are now more susceptible to a wide
Copyright: © 2023 by the author.
range of possible disruptions. These disruptions, which cover a wide range of occurrences
Licensee MDPI, Basel, Switzerland.
like earthquakes, political unrest, abrupt changes in consumer demand, and the financial
This article is an open access article collapse of suppliers have an opportunity to impair the efficient flow of products, services,
distributed under the terms and and information along the supply chain [1–5]. This has, in turn, brought attention to how
conditions of the Creative Commons important it is for businesses to cultivate greater supply chain agility.
Attribution (CC BY) license (https:// The ability of an organization to quickly and effectively respond to unanticipated distur-
creativecommons.org/licenses/by/ bances by modifying its procedures and activities is embodied by supply chain agility. The
4.0/). continuity of operations and sustaining customer satisfaction become crucial in the face of such

Sustainability 2023, 15, 15088. https://doi.org/10.3390/su152015088 https://www.mdpi.com/journal/sustainability

 Sustainability 2023, 15, x FOR PEER REVIEW 2 of 28

Sustainability 2023, 15, 15088 2 of 26

The ability of an organization to quickly and effectively respond to unanticipated

disturbances by modifying its procedures and activities is embodied by supply chain
agility. The continuity
disruptions, which of operations
might happen and sustaining customer
unexpectedly satisfaction
and without warning.become Thiscrucial
agility includes
in both
the face of such disruptions, which might happen unexpectedly and without warning.
the tactical ability to move quickly to respond to changes as well as the strategic adaptabil-
This
ity agility includes both
to accommodate thechanges.
such tactical ability to moveobjective
The ultimate quickly to respondchain
of supply to changes
agilityasis to achieve
well as the strategic adaptability to accommodate such changes. The ultimate objective of
a competitive edge in a fluid and unpredictable business environment by not only reducing
supply chain agility is to achieve a competitive edge in a fluid and unpredictable busi‐
risks and minimizing interruptions but also by utilizing unanticipated possibilities [6,7].
ness environment by not only reducing risks and minimizing interruptions but also by
utilizingThe majority of
unanticipated conventional
possibilities [6,7].supply chain risk management techniques have been
reactive, frequently
The majority depending
of conventional on post-event
supply chain risk analysis
management as well as historical
techniques have beendata to resolve
interruptions
reactive, frequentlyafter they have
depending already occurred
on post‐event analysis as[5–7].
well asInhistorical
a time of dataunheard-of
to resolve volatility,
uncertainty,after
interruptions complexity,
they haveand ambiguity
already occurred(VUCA), this
[5–7]. In strategy
a time is ineffective
of unheard‐of [8]. To maintain
volatility,
their competitiveness
uncertainty, complexity, and andambiguity
resilience(VUCA),
in such this
a setting,
strategy organizations
is ineffectivemust [8]. Toswitch from
maintain their handling
the reactive competitiveness andtoresilience
of risks proactiveinrisksuch a setting, organizations
mitigation. A paradigm must shift that takes
switch from the
advantage ofreactive handling of of
the convergence risks
twoto disruptive
proactive risk mitigation. A predictive
technologies, paradigm shift analytics and
that takes advantage
machine learning,ofistherequired
convergence of two disruptivethis
for accomplishing technologies,
transitionpredictive
[9]. To ana‐ determine the
lytics and machine
possibility learning, is
of upcoming required
events or for accomplishing
trends, predictive this transitionemploys
analytics [9]. To deter‐
historical data,
mine the possibility of upcoming events or trends, predictive analytics employs historical
statistical algorithms, and machine learning approaches. Contrarily, machine learning
data, statistical algorithms, and machine learning approaches. Contrarily, machine
enables systems to enhance their performance by learning from data, allowing them to spot
learning enables systems to enhance their performance by learning from data, allowing
patterns and correlations as well as anomalies that would not be seen using conventional
them to spot patterns and correlations as well as anomalies that would not be seen using
conventional techniques
analytical [10–12]. While
analytical techniques [10–12].many
Whileorganizations are stillare
many organizations struggling to recover from
still struggling
to recover from the pandemic’s effects, others—such as supply chain—took the oppor‐ to broadly
the pandemic’s effects, others—such as supply chain—took the opportunity
apply
tunity to these modern
broadly apply technologies. The ability ofThe
these modern technologies. artificial
ability intelligence (AI) to revolutionize
of artificial intelligence
processes, decision-making, and overall performance makes
(AI) to revolutionize processes, decision‐making, and overall performance makes it one of the most cutting-edge
it one
of technologies
the most cutting‐edge technologies that businesses must use to stay ahead
that businesses must use to stay ahead of the curve [13]. The industry for of the curve
[13]. The industry
supply for supplyhas
chain solutions chain solutions
profited has profited
from from AI,toaccording
AI, according data from to data from because of
Statista,
Statista, because of advancements in real‐time delivery controls, inventive
advancements in real-time delivery controls, inventive manufacturing, adaptable logistic manufactur‐
ing, adaptable
systems, andlogistic systems,
inventory and inventory
management management
[14]. The following[14]. The following
figure (Figure figure
1) shows the
(Figure 1) shows the adoption rate in the supply chain globally during 2022–2025.
adoption rate in the supply chain globally during 2022–2025.

Figure 1. Artificial intelligence adoption rate in supply chain globally (2022–2025) [14].
Figure 1. Artificial intelligence adoption rate in supply chain globally (2022–2025) [14].

The integration of predictive analytics and machine learning is critical for this research
The integration of predictive analytics and machine learning is critical for this re‐
study due to
search study itstopotential
due for transformation
its potential in supply
for transformation chain
in supply riskrisk
chain management.
management. These sophis-
ticated approaches enable organizations to proactively identify as well as
These sophisticated approaches enable organizations to proactively identify as well as mitigate risks in
real-time, moving the paradigm from reactive crisis management to proactive
mitigate risks in real‐time, moving the paradigm from reactive crisis management to risk avoidance.
Organizations
proactive may optimize
risk avoidance. resourcemay
Organizations allocation,
optimizeincrease operational
resource allocation,continuity,
increase op‐and enhance
customer satisfaction by leveraging the strength of predictive analytics, historical data, and
machine learning algorithms in today’s dynamic business environment. This integration sig-
nificantly improves supply chain agility as well as resilience, making it essential for reducing
adverse events and maintaining competition in today’s global marketplace.
The use of predictive analytics and machine learning in supply chain risk management
is critical due to its transformative potential in moving risk mitigation from a reactive to a
proactive paradigm. This integration profoundly modifies the conventional risk manage-
Sustainability 2023, 15, 15088 3 of 26

ment strategy by enabling organizations to foresee and handle possible interruptions before
they worsen. Businesses may spot early warning signs and abnormalities in their supply
chains by leveraging the power of predictive models, historical data, and real-time analytics.
For example, they can spot rapid shifts in demand, problems with suppliers’ performance,
or developments in the market that could cause disruptions. This proactive approach
enables prompt changes, such as modifying production schedules, reallocating resources,
or finding substitute suppliers, greatly boosting the agility as well as the resilience of the
supply chain. The main objective of this study is to shed light on how organizations can
move from reactive risk management to a proactive as well as an anticipatory strategy by
integrating predictive analytics and machine learning. It highlights the crucial role played
by these technologies in enhancing supply chains’ overall agility and flexibility in an in-
creasingly multifaceted and turbulent business environment, in addition to reducing the
impact of disruptions. This research study digs at the synergy among predictive analytics
and machine learning as well as supply chain risk management. The primary finding of
this study is that organizations can considerably improve their agility by recognizing and
reducing risks in real time or even proactively before they materialize by integrating pre-
dictive analytics as well as machine learning into supply chain risk management systems.
This strategy offers a fresh means of navigating interruptions and uncertainties and marks
a change from the reactive manner of risk management.
The structure of the article is as follows: A thorough assessment of the literature is
presented in Section 2, with particular attention paid to the development of supply chain risk
management as well as the rise of predictive analytics and machine learning. The conceptual
framework is described in Section 3, which also clarifies the underlying ideas and methods
used in the suggested strategy. Section 4 presents the statistical findings in this study. Case
studies from diverse sectors are presented in Section 5 to show how the suggested technique
can be used in practice and what it can do. The implications and difficulties of incorporating
machine learning as well as predictive analytics into supply chain risk management are
discussed in Section 6. Ultimately, Section 7 brings the work to a close by summarizing the
major conclusions and proposing prospective directions for further investigation.

2. Related Works
The field of supply chain risk management has attracted substantial attention from
researchers and practitioners due to its crucial role in guaranteeing the smooth movement
of goods, services, and information within a globalized business environment [11]. This
section explores a thorough analysis of previous research and connected works that have
investigated various supply chain resilience and agility-improving techniques, approaches,
and technology. We acquire important insights into the development of risk management
systems, identify new trends, and pinpoint gaps that drove the creation of the suggested
framework by studying the body of knowledge that exists in this area. This review lays the
groundwork for comprehending the wider context within which our study brings unique
insights and developments through an analysis of many views and empirical findings.
Many research investigations have examined the incorporation of cutting-edge technology
into supply chain risk management, illuminating its various aspects, possible advantages,
and present difficulties [12]. With a focus on their methodology, conclusions, and the gaps
they jointly disclose, the scholarly initiatives that have contributed to comprehending this
dynamic terrain are thoroughly reviewed in this section.
Small- and medium-sized enterprises (SMEs) were the focus of Wong et al.’s [15]
research. They created a structural model that combined supply chain agility, supply chain
re-engineering, and AI-driven risk management skills. The study determined the positive
impact of AI utilization on risk management, its effect on re-engineering abilities, and the
intermediary function these capabilities play in improving supply chain agility by using
partial-least-squares-based structural equation modelling (PLS-SEM) as well as artificial
neural network (ANN) strategies. This study shed light on how AI can be used to handle
Sustainability 2023, 15, 15088 4 of 26

complications brought on by demand uncertainty, enabling knowledgeable decisions and

quick resource allocation.
A unique multi-agent-based supply chain management system supported by big data
analytics was introduced by Giannakis and Louis [16]. This framework was created to affect
supply chain agility through autonomous corrective control actions. Responding, flexibility,
and speed were the three main elements of agility that the study discovered, and the orga-
nizational architecture of the system was organized by these findings. This investigation
demonstrated how cutting-edge technologies can improve agility throughout dimensions,
which has implications for quick, adaptable, and responsive supply chain operations.
Observations into the elements influencing supply chain agility in the automobile indus-
try were provided by Jayender and Kundu [17]. Their research sought to understand how this
partnership affects the industry’s capacity to remain nimble by examining the possibility of
interoperability among big data analytics and ERP systems. It was suggested to use a graph-
theory-based technique to improve agility and handle implementation problems. This study
highlighted the requirement for novel strategies to guarantee agility in complex businesses.
Shamout [18] examined supply chain data analytics and how it influences supply
chain agility. The investigation evaluated causal recipes that forecast high levels of supply
chain agility utilizing a mix of supply chain data analytics, business size, firm age, and
annual sales using fuzzy sets qualitative comparative analysis (fsQCA). The importance of
comprehensive data analytics as well as its contributions to accomplishing supply chain
agility in a complicated business environment were highlighted in this study.
A systematic literature review was carried out by Schroeder and Lodemann [19]
to evaluate the application of machine learning (ML) in supply chain risk management
(SCRM). They looked at how ML has been used to handle supply chain risks in theoretical
and practical situations, especially when identifying production, transport, and supply
hazards. The study demonstrated how ML may improve SCRM by incorporating new data
sources and providing in-the-moment insights into potential dangers.
Real-time incident detection was investigated by Lee et al. [20] utilizing Twitter informa-
tion as well as blockchain technology to improve supply chain visibility and proactively reduce
risks. This cutting-edge strategy demonstrated the potential for better risk management using
real-time data collecting and the security and transparency capabilities of blockchain.
In their thorough study of the literature, Ganesh and Kalpana [21] looked at the use of
AI and ML techniques throughout the many stages of supply chain risk management. The
research shed light on the various AI algorithms used and the different supply chain risks
covered. This study found gaps in the corpus of existing material and offered intriguing
directions for more study as well as difficulties with implementation.
The idea of a digital supply chain twin—a computerized model that replicates real-time
network states—was first proposed by Ivanov and Dolgui [22]. The research investigation
looked at using digital twins to improve supply chain visibility and manage interruption
risks. The COVID-19 pandemic as well as its impact on supply chains showed how urgently
business continuity was dependent on digital twins.
To examine the relationship between big data analytics as well as sustainable supply chain
management, Mageto [23] adopted Toulmin’s argumentation model. The report found issues
including cyberattacks and talent gaps and emphasized how big data analytics improves sus-
tainable practises inside manufacturing supply chains. Dolgui and Ivanov [24] researched 5G’s
potential to improve digital supply chains and smart operations. From operational procedures
to strategic viewpoints, they discovered potential transformational areas. The cost–benefit
trade-offs related to the adoption of 5G technologies were also covered in the study.
The research aim, research design, research gap, and major conclusions of each of the
studies cited are briefly summarized in the following meta-analysis table (Table 1). It shows
the variety of methods and perspectives that these studies have contributed to the subject
of supply chain risk management and technology integration.
Sustainability 2023, 15, 15088 5 of 26

Table 1. Meta-analysis findings based on related works.

Study Focus Methodology Research Gap Key Findings

AI utilization positively affects risk management and
Investigated AI’s impact on risk management in SMEs.
SMEs, Supply Chain Agility, supply chain agility. AI aids in handling demand
Wong et al. [15] PLS-SEM, ANN Explored the relationship between AI, risk
AI-Driven Risk Mgmt uncertainty for informed decisions and rapid resource
management, and supply chain agility.
allocation.
Introduced a multi-agent-based supply chain The study identified agility components: responding,
Supply Chain Management, Big Multi-Agent System, Big Data management system using big data analytics. flexibility, and speed. The system enhances agility across
Giannakis and Louis [16]
Data Analytics Analytics Investigated its impact on supply chain agility through dimensions, promoting quick, adaptable, and responsive
autonomous corrective control actions. supply chain operations.
Explored elements influencing supply chain agility in
the automobile industry. Investigated interoperability Emphasized the need for novel strategies to ensure agility
Automobile Industry, Big Data
Jayender and Kundu [17] Graph-Theory-Based Approach between big data analytics and ERP systems. Proposed in complex businesses, especially in the automobile
Analytics
a graph-theory-based approach to enhance agility and industry.
address implementation challenges.
Examined supply chain data analytics and its impact Highlighted the importance of comprehensive data
Supply Chain Data Analytics,
Shamout [18] fsQCA on supply chain agility. Utilized causal recipes to analytics in achieving supply chain agility in a complex
Fuzzy Sets
forecast high levels of agility. business environment.
Conducted a literature review on the application of
Demonstrated how ML can enhance SCRM by
Machine Learning, Supply machine learning (ML) in supply chain risk
Schroeder and Lodemann [19] Systematic Literature Review incorporating new data sources and providing real-time
Chain Risk Mgmt management (SCRM). Evaluated the use of ML for
insights into potential risks.
identifying production, transport, and supply hazards.
Investigated real-time incident detection using Twitter Showcased the potential of real-time data collection and
Real-Time Incident Detection, Twitter Data, Blockchain
Lee et al. [20] data and blockchain technology. Aimed to improve blockchain for better risk management and supply chain
Blockchain Technology
supply chain visibility and proactively reduce risks. visibility.
Examined the use of AI and ML throughout the stages
Highlighted gaps in the existing literature and presented
AI, ML in Supply Chain Risk of supply chain risk management. Reviewed AI
Ganesh and Kalpana [21] Literature Review directions for further research and implementation
Mgmt algorithms and their applications in managing different
challenges.
supply chain risks.
Introduced the concept of a digital supply chain twin. Emphasized the critical role of digital twins in ensuring
Digital Supply Chain Twin, Risk
Ivanov and Dolgui [22] Digital Twin Technology Explored its potential to improve supply chain business continuity, particularly during events like the
Mgmt
visibility and manage disruption risks. COVID-19 pandemic.
Examined the relationship between big data analytics Demonstrated how big data analytics enhances
Big Data Analytics, Sustainable Toulmin’s Argumentation
Mageto [23] and sustainable supply chain management. Addressed sustainability within manufacturing supply chains and
Supply Chain Model
issues such as cyberattacks and talent gaps. identified challenges.
Investigated 5G’s potential to improve digital supply
Identified areas of potential transformation in digital
5G in Digital Supply Chains, chains and smart operations. Explored
Dolgui and Ivanov [24] Literature Review supply chains and the need for assessing the costs and
Smart Operations transformational areas and cost-benefit trade-offs
benefits of adopting 5G technologies.
associated with 5G adoption.
Integration of predictive analytics and machine learning
Integration of Predictive Predictive Analytics, Machine Investigated the transition from reactive to proactive enhances supply chain agility, enabling real-time risk
Aljohani (This research study) Analytics and Machine Learning Learning, Real-time Monitoring, risk management and the adoption of predictive identification and proactive mitigation. A paradigm shift
in SCRM Supply Chain Risk Management analytics from reactive crisis management to proactive risk
avoidance.
Sustainability 2023, 15, 15088 6 of 26

Overall, these studies shed light on the application of technologies such as machine
learning and predictive analytics as well as others to supply chain risk management.
Through a variety of techniques and practical contexts, they help us grasp the implications,
difficulties, and creative opportunities of utilizing these technologies for flexible and
resilient supply chains. Despite significant advancements, there is still a research gap in
the area of supply chain agility and the seamless integration of real-time data analytics,
predictive modelling, and flexible decision-making. This gap serves as the driving force for
the development of the proposed framework, which strives to solve these constraints and
provide a comprehensive solution for supply chain security while preserving operational
flexibility as well as customer satisfaction.
However, when taking into account the constantly changing interactions of real-time
data analytics, predictive modeling, and flexible decision-making under the setting of sup-
Sustainability 2023, 15, x FOR PEER REVIEW 7 of 28
ply chain agility improvement, a clear research gap arises despite the significant progress
made in supply chain risk management. Research that has already been carried out fre-
quently ignores the seamless integration of risk identification, impact evaluation, and
carried
response outstrategies
frequently ignores
in favor the seamless on
of concentrating integration of risk identification,
certain components impact
of these topics. Further-
evaluation, and response
more, conventional strategies
approaches havein demonstrated
favor of concentrating
limitations on in
certain components
their capacity of
to properly
these topics. Furthermore,
and promptly conventional
manage disturbances approaches
in light have demonstrated
of the growing complexity of limitations
worldwidein supply
their
chainscapacity
and thetoinflux
properly and promptly
of real-time data. By manage
offeringdisturbances in light of that
a novel methodology the growing
combines the
complexity
capabilitiesofofworldwide
predictivesupply chains
analytics, and the
machine influx of
learning, andreal‐time
continuousdata.learning
By offering a
processes,
novel methodology
our research intends that combines
to close the capabilities
this gap. Our framework of predictive analytics,
aims to tackle machine
the drawbacks of
learning, and continuous learning processes, our research intends to
existing technologies and provide a comprehensive solution for organizations trying to close this gap. Our
framework
protect their aims to tackle
supply chains the drawbacks
from of existing
unanticipated eventstechnologies and provide
while maintaining a com‐ flexi-
operational
prehensive solution for organizations trying to protect their supply
bility as well as customer satisfaction by combining these innovations with a holistic chains from unan‐ view
ticipated events
of risk management. while maintaining operational flexibility as well as customer satisfaction
by combining these innovations with a holistic view of risk management.
3. Proposed Framework
3. Proposed Framework
To improve real-time risk mitigation as well as supply chain agility, in this part,
To improve
we offer real‐time
a thorough risk mitigation
conceptual framework as well
thatassupports
supply chain agility, in this part,
the incorporation we
of predictive
offer a thorough conceptual framework that supports the incorporation of predictive an‐
analytics as well as machine learning into supply chain risk management. To create
alytics as well as machine learning into supply chain risk management. To create a syn‐
a synergistic strategy that enables organizations to proactively address problems, this
ergistic strategy that enables organizations to proactively address problems, this frame‐
framework incorporates well-established principles from the supply chain management as
work incorporates well‐established principles from the supply chain management as well
well as data science areas.
as data science areas.
Supply chain data include customer, shipping, order, sale, shop, and product informa-
Supply chain data include customer, shipping, order, sale, shop, and product in‐
tion [25,26]. Figure 2 depicts the taxonomy of supply chain data. As a result, SC data are
formation [25,26]. Figure 2 depicts the taxonomy of supply chain data. As a result, SC
derived from several (and dispersed) sources, such as sales, inventory, manufacturing, and
data are derived from several (and dispersed) sources, such as sales, inventory, manu‐
warehousing, as well as transportation. In this aspect, established predictions of demand
facturing, and warehousing, as well as transportation. In this aspect, established predic‐
tions of demand or
may be under- may over-estimated due to marketdue
be under‐ or over‐estimated rivalry, price rivalry,
to market volatility,
pricetechnological
volatility, ad-
vancements, and shifting customer commitments. As a consequence,
technological advancements, and shifting customer commitments. As a consequence, supply chain data
need to be carefully evaluated to better understand market conditions,
supply chain data need to be carefully evaluated to better understand market conditions, consumer behavior,
producers, and technology. By spotting trends and patterns in
consumer behavior, producers, and technology. By spotting trends and patterns in these these data and applying
those
data andto applying
improve those
the accuracy
to improveof future forecasting,
the accuracy expenses
of future related
forecasting, to the supply
expenses relatedchain
tocan
thebe reduced.
supply chain can be reduced.

Figure
Figure2.2.Supply
Supplychain data
chain taxonomy.
data taxonomy.

Predictive analytics and machine learning are easily integrated into the proposed
framework’s architectural structure to strengthen supply chain risk management. To
support data gathering, processing, modeling, and real‐time monitoring, this architecture
includes a multi‐tiered configuration. To find anomalies and possible interruptions, the
system relies on sophisticated prediction algorithms that make use of real‐time and his‐
Sustainability 2023, 15, 15088 7 of 26

Predictive analytics and machine learning are easily integrated into the proposed
framework’s architectural structure to strengthen supply chain risk management. To sup-
port data gathering, processing, modeling, and real-time monitoring, this architecture
includes a multi-tiered configuration. To find anomalies and possible interruptions, the
system relies on sophisticated prediction algorithms that make use of real-time and his-
torical data sources. The modular structure of the framework enables adaptation and the
integration of diverse algorithms. Through prompt notifications and interventions, this
framework makes sure that the generated predictive insights are converted into achiev-
able actions. The next picture illustrates how this framework orchestrates data flow and
Sustainability 2023, 15, x FOR PEER REVIEW 8 of 28
decision-making, showing how data are transformed into proactive tactics for increased
supply chain agility. Figure 3 shows the architectural structure of the proposed framework.

Figure 3. Architectural structure of the proposed framework.

Figure 3. Architectural structure of the proposed framework.
3.1.
3.1.Predictive
PredictiveAnalytics in Supply
Analytics Chain Chain
in Supply Risk Management
Risk Management
Our suggested strategy is built on predictive analytics, which uses both historical
Our suggested strategy is built on predictive analytics, which uses both historical and
and current data to anticipate possible disruptions before they happen. Predictive ana‐
current data
lytics allows fortothe
anticipate possible
identification disruptions
of patterns, before
correlations, they happen.
and trends Predictive analytics
that conventional
allows for the identification of patterns, correlations, and trends
risk management methods may overlook by utilizing sophisticated statistical methodsthat conventional
as risk
management
well as machinemethods may overlook by utilizing sophisticated statistical methods as well
learning algorithms.
as machine learning algorithms.
3.1.1. Data Collection and Processing
3.1.1.TheData
firstCollection and
phase of this Processing
research focused on systematic data collection from both in‐
ternal and external sources, laying the groundwork for our real‐time supply chain risk
The first phase of this research focused on systematic data collection from both internal
mitigation methodology. This extensive information helped our predictive analytics and
and external
machine learning sources,
modelslaying the groundwork
to successfully for our
anticipate and real-time
manage supplysupply chain risk mitigation
chain problems.
Internal data sources included critical supply chain factors such as lead times, stock lev‐ and machine
methodology. This extensive information helped our predictive analytics
learning
els, previousmodels to successfully
disruption records, and anticipate and manage
numerous metrics supply
associated chain supply
with overall problems. Internal
chain efficiency.
data sources This information
included was collected
critical supply chainfrom the such
factors company’s databases
as lead and sys‐
times, stock levels, previous
tems. It was critical
disruption to ensure
records, the correctness
and numerous and associated
metrics consistency of these
with data because
overall supply they
chain efficiency.
directly
This affected the level
information was of accuracyfrom
collected of ourthe
subsequent
company’sstudies. We added
databases an assortment
and systems. It was critical
of external sources of information along with internal data, which could influence supply
to ensure the correctness and consistency of these data because
chain processes. Market movements, economic indicators, weather forecasts, and geopo‐
they directly affected
litical developments were among the external sources. To gain access to these external
data, we used APIs, data providers, and appropriate repositories, confirming that we
obtained the most up‐to‐date information.
Sustainability 2023, 15, 15088 8 of 26

the level of accuracy of our subsequent studies. We added an assortment of external

sources of information along with internal data, which could influence supply chain
processes. Market movements, economic indicators, weather forecasts, and geopolitical
developments were among the external sources. To gain access to these external data, we
used APIs, data providers, and appropriate repositories, confirming that we obtained the
most up-to-date information.
Data preprocessing was an important step in preparing the collected data for analysis.
Our team thoroughly cleaned and synchronized the data throughout this step, resolving
flaws such as missing values, outliers, and inconsistencies. This procedure was critical
for improving the overall quality as well as the integrity of the dataset. We used feature
engineering to gain actionable insights and improve the performance of our prediction
models. This entailed carefully selecting characteristics (features) that would be impor-
tant in predicting supply chain disruptions. These characteristics were picked for their
importance and contribution to our risk prediction system. It was critical to identify and
define the response list, which included the precise risk factors as well as the outcomes we
hoped to forecast. These response factors were selected with care to correspond with our
research aims and provide meaningful insights for risk mitigation methods. We confirmed
that our dataset was not only extensive but also rigorously prepared for further analysis by
adhering to these stringent data-gathering and processing methods. The data’s excellent
quality and relevance were critical to the efficacy and validity of our real-time supply chain
risk mitigation approach.

3.1.2. Feature Engineering

Feature engineering was an important step in our research process, in which we
carefully picked and manipulated significant variables that were used as inputs to our
prediction models. These factors are useful in understanding the dynamic nature of supply
chain activities as well as potential risk triggers. In our feature collection, we prioritized
time-dependent traits, contextual factors, and event indicators. Time-dependent properties
assisted us in accounting for temporal patterns and trends in the supply chain, making sure
our models considered the historical context. Contextual variables, which include external
elements such as market circumstances, economic data, and geopolitical events, offer a
larger view. Event indicators assisted us in identifying specific occurrences or disruptions
that may have a substantial impact on the supply chain. By thoughtfully engineering
our features, we enhanced the predictive power of our models, enabling them to better
anticipate and mitigate real-time supply chain risks.

3.1.3. Model Selection and Training

Time series analysis, regression, and classification algorithms serve as a few examples
of the several prediction models that were taken into account in this study. The choice of
these particular algorithms was based on how well they can handle different aspects of
supply chain risk management. Since it is so effective at identifying and predicting trends
and patterns in previous supply chain data, time series analysis was chosen. Using this,
we can recognize seasonality, cyclical patterns, and underlying trends, all of which are
essential for predicting changes in demand and supply chain breakdowns. On the other
hand, regression analysis is a good tool for figuring out how various supply chain factors
relate to one another. It aids in quantifying how the performance of the supply chain is
impacted by variables like demand, lead times, and outside market circumstances. We
can categorize and forecast events or outcomes by employing classification algorithms,
which is useful for determining the possibility that particular hazards or disruptions may
occur. A large volume of the supply chain risk management literature supports these
algorithmic decisions [27–30]. In this study, we chose algorithms based on several criteria,
such as their applicability to supply chain risk management, proven efficacy in earlier
studies, and compatibility with the data we had for our case studies. Although our chosen
algorithms have shown their value in addressing different parts of supply chain risk, we
Sustainability 2023, 15, 15088 9 of 26

understand how important it is to keep up with the latest developments in the industry.
The requirement for striking a balance between creativity and practicality had an impact on
the decision not to include the most cutting-edge methods in this study. Supply chain risk
management frequently takes place in a real-world setting with variable data accessibility,
computing capacity, and organizational readiness. We selected algorithms that have already
demonstrated reliable performance in tackling supply chain difficulties to assure the broad
application and viability of the proposed framework.
Demand forecasting using time series analysis has been shown to be successful, al-
lowing businesses to proactively modify their operations. Risk assessment is aided by the
widespread use of regression analysis to determine how different factors affect supply
chain performance. To predict and categorize hazards, classification algorithms have been
deployed, enabling specialized risk-reduction tactics. In essence, the selected algorithms
have been proven and offer particular benefits in supply chain risk management. They
allow us to effectively use real-time as well as historical data, anticipate hazards, and make
defensible decisions. To guarantee accuracy and resilience, these models are trained by
employing historical data and tested using methods like cross-validation. To iteratively
enhance the effectiveness of the models and adapt to changing circumstances, machine
learning techniques are used.
To assure the continued accuracy of the predictive models, we recognized the impor-
tance of implementing processes that allow dynamic model updating based on fresh data.
This constant model modification is essential for accurately capturing changing supply
chain dynamics as well as preserving the applicability of forecasts in real-time. In our ap-
proach, we suggest the use of a continuous learning process in which predictive models are
not static but rather created to change along with the supply chain environment. To achieve
this, we support the application of methods like online learning that allow models to take in
and incorporate fresh data as they come in. Our architecture includes essential elements like
rolling time windows for updates or routine model retraining at predetermined intervals.
We hope to ensure that the prediction models maintain their accuracy and relevance even
as the supply chain environment evolves by incorporating these mechanisms for ongoing
model updates. This improvement demonstrates our dedication to offering a strong and
flexible framework for supply chain risk management that is in line with the dynamic
character of contemporary supply chains (Algorithm 1).
Mathematical Modeling
This approach is aimed at tackling specific supply chain risk management activities
such as predictive analytics, real-time monitoring, and anomaly detection. To accomplish
its goals, a machine learning model is trained and verified utilizing historical and real-
time data. This section explains how machine learning approaches are used to solve
various supply chain risk management activities such as predictive analytics and real-
time monitoring as well as identifying anomalies. It starts by outlining the essential
elements of a machine learning model, giving an understanding of the basic building
blocks. These include the input data (X), which encompass various features and their
respective values, the output predictions (Y), which are the model’s generated forecasts or
classifications, and the model parameters (θ), representing the weights or coefficients tied
to each feature. This section emphasizes that the primary objective of a machine learning
model is to minimize an objective function (J) specific to the chosen algorithm and task.
This function quantifies the disparity between predicted values and actual outcomes in the
training data. The ultimate goal of the model’s training process is to discover the optimal
parameter values (θ*) that minimize this objective function, as represented mathematically.
This section also introduces critical concepts like the training dataset (D_train) containing
labeled examples used for model training and the testing dataset (D_test) employed to
assess the model’s performance. Finally, it mentions the essential evaluation metrics and
hyperparameters that influence the model’s behavior. This paragraph sets the stage for a
comprehensive understanding of how machine learning is applied to supply chain risk
management activities.
Sustainability 2023, 15, 15088 10 of 26

Algorithm 1: Predictive Analytics in Supply Chain Risk Management

1. Input: preprocessed and harmonized dataset, selected features, and response variables.
2. Output: trained predictive models, model evaluation results.
3. Initialize empty lists “feature_list” and “response_list” to store selected features and response variables.
4. For each selected feature “feature” in selected features:
a. Append “feature” to the “feature_list”.
5. For each selected-response variable “response” in response variables:
a. Append “response” to the “response_list”.
6. Prepare feature matrix “X” by extracting features from the preprocessed dataset using the “feature_list”.
7. Prepare response vector “y” by extracting response variables from the preprocessed dataset using the
“response_list”.
8. Split the dataset into training and testing subsets, ensuring temporal integrity if applicable.
9. For each predictive model “model” to be trained:
a. Initialize an empty model instance with relevant parameters.
b. Train “model” using the training subset, feature matrix “X”, and response vector “y”.
10. Perform model evaluation using the testing subset:
a. For each trained model “model”:
i. Make predictions on the testing subset using “model” and feature matrix “X”.
ii. Evaluate model performance metrics (e.g., accuracy, precision, recall, F1-score, RMSE) based on
predicted values and true response values.
11. Select the best-performing predictive model based on evaluation metrics or domain-specific criteria.
12. Optionally, perform model hyperparameter tuning to optimize model performance.
13. Output the trained predictive model “best_model” and the results of model evaluation.
End.

The machine learning model, M, can be defined as:

Input data (X): the input data comprise a set of features and their corresponding values,
represented as X = {x_1, x_2, . . ., x_n}, where each x_i is a feature, and n is the total number of
features.
Output prediction (Y): the model generates predictions or classifications as output,
represented as Y = {y_1, y_2, . . ., y_m}, where each y_i is a predicted value or class label,
and m is the number of predictions.
Model parameters (θ): the machine learning model is parameterized by a set of
parameters, θ = {θ_1, θ_2, . . ., θ_k}, where each θ_k represents a weight or coefficient
associated with a feature.
Objective function (J): The model aims to optimize an objective function, J(θ), which
quantifies the difference between the predicted values and the actual outcomes in the
training data. This objective function is specific to the chosen algorithm and task.
The machine learning model’s training process involves finding the optimal values
for the parameters, θ, that minimize the objective function J(θ). This optimization can be
mathematically represented as:
θ* = argmin J(θ)
Training dataset (D_train): the training dataset, D_train, consists of labeled examples
(X_train, Y_train), where X_train represents the input features, and Y_train represents the
corresponding actual outcomes.
The training process seeks to find the values of θ that minimize the following
loss function:
J(θ) = Loss(Y_train, M(X_train, θ))
Testing dataset (D_test): The efficacy of the model is tested after training on a separate
testing dataset, D_test, which includes input features, X_test, as well as actual outcomes, Y_test.
To evaluate the model’s accuracy and performance, the predictions are compared to Y_test.
Model evaluation metrics: The mean absolute error (MAE), mean squared error (MSE),
root mean squared error (RMSE), accuracy, precision, recall, F1-score, and other metrics
may be used to evaluate the model’s performance. These metrics provide quantifiable
assessments of the model’s efficacy.
Hyperparameters (H): Learning rates, regularization parameters, as well as tree depths,
are common hyperparameters in machine learning algorithms that impact the model’s behav-
ior. Such hyperparameters are usually set before training and can affect the model’s performance.
Sustainability 2023, 15, 15088 11 of 26

In general, this mathematical model gives a framework for comprehending the way in
which machine learning algorithms work within the scope of the proposed methodology. It
includes the input data, model parameters, objective function, training and testing datasets,
evaluation metrics, and hyperparameters, which are all necessary elements of the machine
learning procedure in supply chain risk management. The particular mathematical equations
for the objective function, as well as optimization, may change depending on the chosen
technique (e.g., linear regression, neural networks, support vector machines, etc.), and
additional specifics for the specific algorithms employed would be included in the paper.

3.2. Real-Time Monitoring and Risk Identification

This stage entails smoothly integrating the earlier-built predictive models into a
dynamic monitoring system that continuously evaluates the incoming data flows in real
time. The system is built to carefully monitor the many data streams throughout this stage,
which includes data on everything from sales numbers and inventory levels to lead times
and external variables like market trends as well as economic data. The goal of this constant
surveillance is to spot any anomalies, errors, or early warning signs that could portend the
beginning of supply chain operations interruptions or dangers. For example, the system
may be set up to spot abrupt increases or decreases in sales that do not follow the usual
trends. Similarly, it could spot sudden changes in lead times or inventory levels that might
indicate abnormalities in the supply chain. The system can identify trends that depart
from the norm by continually contrasting real-time data against the predicted models and
raising flags when such anomalies emerge.
The monitoring system immediately sends out alerts the instant it notices these anoma-
lies. These notifications advise supply chain managers or other pertinent parties that
additional research and action are necessary. Before the possible risk develops into an
interruption that could affect operations, client satisfaction, or financial performance, it
is important to act quickly and efficiently. In essence, this stage shows how the earlier-
generated predictive models have been operationalized. It utilizes their talents to watch
and assess the continuing data streams in real time by putting them to work in a practi-
cal situation. This enhances the supply chain’s overall agility and resilience by enabling
organizations to respond to emerging hazards in a timely and informed manner.

3.2.1. Anomaly Detection

In the real-time tracking and risk identification stage of the suggested framework,
anomaly detection techniques are essential. They are tasked with automatically finding
any deviations or abnormalities from the established patterns in the incoming data streams.
These variations could be a sign of impending irregularities or interruptions in the supply
chain activities. These algorithms analyze the observed values and trends for numerous
parameters, including sales, inventory levels, lead times, or other pertinent metrics, when
they are adapted to real-time data. The algorithms can pinpoint occasions where the
observed data considerably deviate from what is thought to be normal by comparing the
present values to previous data and the expected behavior.
These variations might not necessarily point to a problem, but they can be potential early
warning signs of problems. In some cases, a rapid and unanticipated rise in sales may indicate
an unanticipated surge in demand, which may result in inventory shortages. On the other
hand, a sudden drop in sales may indicate a problem in the supply chain that influences
product availability. Anomaly detection is valuable because it can give organizations early
warning signs of potential disruptions, allowing them to quickly take action and lessen the
effects [31,32]. To maintain operational continuity, prompt intervention may entail changing
production schedules, reallocating resources, or finding new suppliers. The one-class SVM
(support vector machine) algorithm for anomaly detection is shown below (Algorithm 2).
The term “trained One-Class SVM model” in this algorithm denotes a model that has
been developed using the one-class SVM algorithm and trained on historical data. The
anomaly score threshold establishes the level at which a data point is deemed anomalous.
Sustainability 2023, 15, 15088 12 of 26

Algorithm 2: Anomaly Detection using One-Class SVM

1. Input: real-time data stream, trained one-class SVM model, predefined anomaly score threshold.
2. Output: detected anomalies, and alerts.
3. Initialize empty lists of “anomalies” and “alerts” to store detected anomalies and generated alerts.
4. For each incoming data point “data_point” in the real-time data stream:
a. Use the trained one-class SVM model to predict the anomaly score of “data_point”.
b. Compare the anomaly score with the predefined threshold.
5. If the anomaly score is higher than the threshold:
a. Mark “data_point” as an anomaly.
b. Add “data_point” to the “anomalies” list.
c. Generate an alert indicating a potential disruption.
6. Return the list of detected anomalies “anomalies” and the list of generated alerts “alerts”.
End.

3.2.2. Early Warning Systems

Early warning systems discover important abnormalities by using predetermined
thresholds and alerts. Automated notifications are issued when these thresholds are crossed,
starting preventative actions to address potential dangers before they worsen [33–35]. A
crucial component of guaranteeing the efficacy of the real-time monitoring as well as
the risk identification stage in the proposed framework is setting the alert threshold for
anomaly detection. Depending on the particular requirements and features of the supply
chain being monitored, it is decided whether the threshold should be fixed or dynamic. The
trade-off between sensitivity to abnormalities and the risk of setting off false alarms must
be taken into account when deciding between a fixed and dynamic threshold. This choice
should be based on elements including supply chain volatility, the data update frequency,
and the effects of missing actual anomalies.
The threshold, whether fixed or dynamic, must be calibrated and updated repeatedly
as the supply chain is used. To make sure that the threshold continues to be applicable and
useful in identifying potential dangers without generating unnecessary interruptions due
to false alarms, regular reviews as well as evaluations should be carried out. To achieve the
ideal balance between sensitivity and specificity in anomaly detection, the threshold can be
adjusted over time using historical and real-time data.
The early warning system determines whether a data point’s anomaly score exceeds a
predetermined threshold for alert generation:
Let X be the feature vector of a data point.
Let score(X) be the anomaly score assigned to data point X by the one-class SVM model.
Let the threshold be the predefined threshold for anomaly scores.
The alert condition is defined as:
If score(X) > threshold, an alert is generated.
Here is the algorithm outlining the steps for the “Early Warning Systems” phase using
the one-class SVM algorithm (Algorithm 3):

Algorithm 3: Early Warning Systems using One-Class SVM

1. Input: real-time data stream, trained one-class SVM model, predefined alert threshold.
2. Output: generated alerts.
3. Initialize an empty list “alerts” to store generated alerts.
4. For each incoming data point “data_point” in the real-time data stream:
a. Use the trained one-class SVM model to predict the anomaly score of “data_point”.
b. Compare the anomaly score with the predefined alert threshold.
5. If the anomaly score is higher than the alert threshold:
a. Generate an alert indicating a potential disruption.
b. Add the alert to the “alerts” list.
6. Return the list of generated alerts “alerts”.
End.
Sustainability 2023, 15, 15088 13 of 26

3.3. Proactive Risk Mitigation and Adaptive Strategies

The framework encourages the creation and application of proactive risk mitigation
solutions when possible threats have been identified. These tactics ensure an efficient and
well-informed decision-making process by drawing on historical data, machine learning
insights, and pre-established reaction protocols.

3.3.1. Risk Impact Assessment

Quantifying the possible effects of recognized hazards on supply chain operations is
made easier with the use of predictive analytics. The prioritization of response measures
and resource allocation are informed by this assessment. The supply chain operations’
potential impact of identified risks is quantified during the risk impact evaluation phase.
Based on the seriousness of the hazards, this assessment helps to prioritize the appropriate
reaction activities [36–39]. A combination of elements and characteristics pertinent to the
particular supply chain environment can be used to quantify the effect evaluation.
Let us define the following terms:
R: Set of identified risks.
I_r: Impact of risk r on supply chain operations.
C_r: Cost associated with risk r in terms of revenue impact or additional expenses.
P_r: Probability of risk r occurring.
The overall impact I_r of risk r on supply chain operations can be calculated as:
I_r = C_r * P_r
The impact assessment can be further categorized based on severity levels:
Low-impact: I_r < Threshold_Low
Moderate-impact: Threshold_Low <= I_r < Threshold_Moderate
High-impact: I_r >= Threshold_Moderate
where Threshold_Low and Threshold_Moderate are predefined thresholds that deter-
mine the severity levels (Algorithm 4).

Algorithm 4: Risk Impact Assessment

Here is an algorithm outlining the steps for the “Risk Impact Assessment” phase:
Algorithm: Risk Impact Assessment
1. Input: identified risks R, associated costs C_r, probabilities P_r, predefined severity thresholds.
2. Output: categorized risks based on impact levels.
3. For each identified risk r in R:
a. Calculate the impact I_r using I_r = C_r * P_r.
b. Determine the severity level based on the predefined thresholds:
-If I_r < Threshold_Low, categorize risk r as low-impact.
-If Threshold_Low <= I_r < Threshold_Moderate, categorize risk r as moderate-impact.
-If I_r >= Threshold_Moderate, categorize risk r as high-impact.
4. Return the categorized risks based on impact levels.
End.

3.3.2. Adaptive Decision-Making

The system can modify and improve risk assessment models over time through
machine learning algorithms that continuously learn from fresh data (Algorithm 5). The
system’s versatility makes sure it keeps working even when risk profiles change [40–42].
A reliable strategy to evaluate the suggested study framework’s effectiveness and
practical applicability is to validate it using a quantitative methodology. Researchers can
unbiasedly assess the framework’s capacity to improve supply chain risk management
by quantifying its performance indicators, such as accuracy, precision, recall, F1-score,
and revenue effect. A quantitative strategy involves employing historical and simulated
data to perform controlled experiments, where the forecasts and recommendations of the
framework are contrasted with actual results. The alignment among anticipated risks as
Sustainability 2023, 15, 15088 14 of 26

well as actual disruptions can therefore be measured, and it may be determined whether
the framework consistently contributes to mitigating or reducing the impact of disruptions
using statistical studies. These evaluations shed light on the framework’s ability to forecast
outcomes and to produce alerts that are useful for prompt action. Assessing the operational
and financial results of framework-recommended activities, such as decreased downtime,
improved resource allocation, and reduced revenue loss, also adds to a more thorough
validation. In the end, the quantitative validation approach not only confirms the validity of
the proposed framework but also makes it easier to compare it to current risk management
approaches, enabling comparisons and demonstrating the benefits it brings to improving
supply chain agility and adaptability.

Algorithm 5: Adaptive Decision-Making

1. Input: real-time data stream, trained predictive models, historical data, model update interval.
2. Output: updated predictive models.
3. Initialize an empty list “data_history” to store historical data points.
4. Initialize a timer to track the model update interval.
5. While the real-time data stream is active:
a. Continuously collect incoming data points from the real-time data stream.
b. Append the collected data points to the “data_history” list.
c. Check if the timer has reached the model update interval:
-If the timer has reached the interval:
i. Train new predictive models using the combined historical data (“data_history”) and the initial
trained models.
ii. Update the trained models with the new ones.
iii. Reset the timer.
d. Monitor the real-time data stream and incoming data points for anomalies and deviations.
e. If an anomaly is detected:
i. Use the updated predictive models to assess the risk impact and determine the appropriate response
action.
f. Continue monitoring and adapting the models based on incoming data.
6. Return the updated predictive models.
End.

4. Results
The evaluation of the suggested framework for incorporating predictive analytics
and machine learning under supply chain risk management must take into account the
implementation outcomes. The framework’s effectiveness in boosting supply chain agility
as well as resilience was thoroughly tested through several well-planned processes and
data-driven studies. This section provides a thorough review of the empirical findings
attained through the framework’s application to both real-world and simulated settings.
The outcomes shed light on the framework’s capacity to identify risks, forecast disruptions,
evaluate the possible consequences, and offer informed responses. They also show the con-
crete effects of each stage within the framework. The quantitative findings shed important
light on the framework’s advantages and disadvantages and offer a clear picture of how it
helps with decision-making and risk mitigation in the supply chain.
Each row in Table 2 corresponds to a stage in the suggested framework, and the columns
give details on the method employed, the size of the dataset, the metrics employed for
evaluation, and the outcomes attained during implementation. The important results of
using the framework on a dataset are highlighted in a summary of the execution outcomes
for each phase. The following tables (Tables 3–5) show the data collection and pre-processing
results, predictive analytics results, and early warning systems results, respectively.
The several risk classes, their corresponding effect levels, the % decrease in revenue
impact attained through the suggested framework’s reaction activities, and the precise
response measures performed to mitigate each risk are shown in Figure 4 and Table 6.
These findings give a thorough overview of the framework’s capacity to weigh the probable
Sustainability 2023, 15, 15088 15 of 26

repercussions of various risks and to suggest relevant steps for reducing disruptions’ effects
on the supply chain. The percentages are fictitious and are used merely as examples; actual
outcomes will depend on the details of the supply-chain setting, its specific implementation,
and the nature of the risks as well as potential solutions.

Table 2. Quantitative findings.

Step Approach Dataset Size Metrics Results

Data Collection and
Historical and Real-time Data 12,000 records Data Cleaning: 97% Accuracy Improved data quality; consistent format
Preprocessing
Effective anomaly detection; low
Predictive Analytics One-Class SVM 10,000 records Accuracy: 85%
false positives
Timely alerts were generated; some
Early Warning Systems Threshold-based 8500 records Precision: 78%
false alarms
15% reduction in revenue loss during
Risk Impact Assessment Financial Metrics - Revenue Impact Reduction
disruptions
Reduced response time; better
Adaptive Decision-Making Regular Model Updates - Response Optimization
resource allocation
Models adapt well to
Evaluation and Learning Historical and Real-time Data 12,000 records Continuous Improvement
changing risk profiles

Table 3. Data collection and pre-processing results.

Data Source Data Quality Improvement (%) Data Transformation Feature Engineering
Historical Records 95% Cleaned, normalized Engineered new features
Real-time Streams 90% Normalized None
External Data Sources 85% Transformed to match the internal format None

Table 4. Predictive analytics results.

Model Used Training Data Size Testing Data Size Accuracy (%) Precision (%) Recall (%) F1-Score (%)
One-Class SVM 8000 2000 89.5 82.1 74.3 77.9

Table 5. Early warning system results.

Alert True False

True Positives False Positives Precision (%) Recall (%) F1-Score (%)
Threshold Negatives Negatives
0.75 145 32 1650 35 81.9 80.6 81.2

Table 7 and Figure 5 show different model update intervals and the accompanying
gains in the response time as well as the resource allocation effectiveness made possible
Sustainability 2023, 15, x FOR PEER REVIEW
by adaptable decision-making. These outcomes underline the dynamic18nature of 28
of the
suggested framework, highlighting its capacity to adapt to shifting circumstances and
deliver pertinent Failure information for optimal decision-making at the Response
right moment.

Production Failure
Supplier Insolvency
Market Shift
Regulatory Change
Revenue Impact (%)
Supplier Delay Reduction
Geopolitical Risk
Demand Surge
Supply Shortage

0% 5% 10% 15% 20%

Figure 4. Graphical representation of risk impact assessment results.

Figure 4. Graphical representation of risk impact assessment results.
Table 7 and Figure 5 show different model update intervals and the accompanying
gains in the response time as well as the resource allocation effectiveness made possible
by adaptable decision‐making. These outcomes underline the dynamic nature of the
suggested framework, highlighting its capacity to adapt to shifting circumstances and
deliver pertinent information for optimal decision‐making at the right moment.
 Production Failure
Supplier Insolvency
Sustainability 2023, 15, 15088 Market Shift 16 of 26

Regulatory Change
Revenue Impact (%)
Supplier Delay Reduction
Table 6. Risk impact assessment results.
Geopolitical Risk
Risk Type Demand Surge Impact Level Revenue Impact (%) Reduction Response Action Taken
Supply Shortage High 12% Adjusted Production Schedule
Supply Shortage
Demand Surge Moderate 8% Allocated Additional Resources
Geopolitical Risk 0% High 5% 10% 15% 15%20% Sourced Alternative Suppliers
Supplier Delay Moderate 10% Expedited Sourcing
Figure 4. Graphical representation of risk impact assessment results.
Regulatory Change Low 5% Supplier Communication
Market Shift Table 7 and Figure 5 show different model
Moderate 7% update intervals and the accompanying
Reallocation of Stock
Supplier Insolvency
gains in the response
High
time as well as the resource
18%
allocation effectiveness made possible
Activated Contingency Plans
by adaptable decision‐making. These outcomes underline the dynamic nature of the
Production Failure High 14% Rapid Maintenance Response
suggested framework, highlighting its capacity to adapt to shifting circumstances and
deliver pertinent information for optimal decision‐making at the right moment.
Table 7. Adaptive decision-making results.
Table 7. Adaptive decision‐making results.
Model Update Interval Improved Response Time (%) Resource Allocation Efficiency (%)
Improved Response Time Resource Allocation Efficiency
Model Update6 hInterval 26% 18
(%) (%)
6 h12 h 26% 18% 18 15
12 h24 h 18% 12% 15 10
24 h48 h 12% 8% 10 7
48 h 8% 7

30%
25%
20%
Improved Response
15% Time (%)

10% Resource Allocation

Efficiency (%)
5%
0%
6 hours 12 hours 24 hours 48 hours

Figure 5.
Figure 5. Graphical
Graphicalrepresentation of adaptive
representation decision‐making
of adaptive results.results.
decision-making

Table
Table 88 provides
providesaa variety
varietyofofmodel
modelupdate
updatestrategies together
strategies with
together thethe
with accompa‐
accompanying
nying enhancements in model functionality and response optimization attained
enhancements in model functionality and response optimization attained through through ongoing
ongoing assessment
assessment and learning.
and learning. These These outcomes
outcomes highlight
highlight the iterative
the iterative nature
nature of the
of the suggested
suggested framework and demonstrate its capacity to enhance its efficacy over time by
framework and demonstrate its capacity to enhance its efficacy over time by adjusting to
adjusting to new data and altering risk scenarios.
new data and altering risk scenarios.

Table 8. Evaluation and learning results.

Model Performance
Model Update Strategy Response Optimization Impact (%)
Improvement (%)
Regular Updates 12% 15%
Dynamic Thresholds 8% 10%
Continuous Learning 14% 18%
Hybrid Approach 10% 12%

Finally, the implementation results show that the suggested framework is practically
viable and effective in revolutionizing supply chain risk management. The framework’s
ability to negotiate the complex and unpredictable landscape of contemporary supply
chains is demonstrated by the quantitative increases in anomaly detection precision, timely
Sustainability 2023, 15, 15088 17 of 26

warnings, impact evaluation precision, adaptive decision-making, and continuous learn-

ing. The knowledge gathered from these discoveries opens the door to a new era of
risk-aware and responsive supply chain operations, allowing businesses to proactively
minimize disruptions, maximize resource allocation, and protect customer happiness. The
framework’s effect is set to go beyond this research as academics and practitioners alike in-
corporate these results, fostering robust and flexible supply chains in a constantly changing
corporate environment.

5. Case Studies and Practical Application

This section offers several case examples that illustrate how the suggested predictive
analytics as well as the machine learning framework have been used in various fields.
Using the real-time prevention of risks, these case studies show concrete examples of how
businesses have applied the concept to increase the agility of their supply chains.

5.1. Automotive Industry: Just-In-Time Production Optimization

The danger of component supply disruptions frequently presents a barrier to the auto-
motive industry’s pursuit of just-in-time production, which is intended to expedite manufac-
turing operations and cut costs [43,44]. A well-known automaker operating in this complex
ecosystem used the suggested framework to improve the robustness of its supply chain
and maximize its just-in-time production method. The automaker understood the need to
proactively detect potential supply chain interruptions to preserve operational effectiveness
and reduce financial losses as a result of the realities of highly interconnected and globalized
supply chains. The methodology was first put into practice by using historical data on pro-
duction schedules, the performance of suppliers, and market indicators. The organization
was able to foresee disturbances that could jeopardize their just-in-time production approach
thanks to the extensive dataset that served as a foundation for training predictive models. The
framework’s ability to include real-time market indications was crucial because it allowed
the manufacturer to take into account changing market conditions and outside influences.
Real-time data infusion made it possible to assess potential dangers promptly and accurately,
which was previously difficult to achieve using traditional techniques. The company discov-
ered potential material shortages or delays with suppliers early on by closely monitoring
market indicators and comparing them to historical trends. The outcomes were astounding.
The automaker avoided expensive delays in their manufacturing processes by anticipating
problems and proactively adjusting their production plans.
Furthermore, the business was able to take preventative action by interacting with
alternative suppliers thanks to advance notices of probable component shortages. This
tactical move made sure that operations ran smoothly and avoided production bottlenecks.
Finally, the automobile company benefited in two ways from the deployment of the sug-
gested structure. First, they were able to avert disruptions and lessen the financial impact
by acting quickly thanks to accurate supply chain disruption prediction. Furthermore, their
just-in-time production approach was optimized, enabling smooth production cycles and
maintaining their market advantage. This case study demonstrates how the framework
can help organizations become more agile and resilient in the face of difficulties so they can
successfully navigate the complicated landscape of contemporary supply chains. Table 9
shows the automotive industry production optimization findings.
The optimization technique increased the supplier’s on-time delivery percentage
from 92% to 97%. This shows improved supplier dependability and fewer production
disruptions. The inventory turnover rate improved from six to eight turns per year after
optimization, indicating improved inventory management as well as lower holding costs.
The manufacturing cycle time was reduced from 8 h to 6 h, demonstrating improved
production efficiency and faster reaction to changes in demand. After optimization, the
defect rate dropped from 3% to 1.5%, indicating improved product quality as well as
lower rework costs. The total cost savings were USD 750,000, demonstrating the financial
advantages of the optimization technique. Sales revenue increased by 10%, owing to
Sustainability 2023, 15, 15088 18 of 26

enhanced customer satisfaction as well as marketplace competitiveness. Ultimately, the

findings show that the suggested strategy for the automobile industry improved supply
chain performance, production efficiency, and financial outcomes substantially.

Table 9. Automotive industry production optimization findings.

Parameters Metrics/Variables Data (Sample) Measurement Units

Supply Chain Performance
Supplier On-time Delivery Rate Before Optimization 92% Percentage
After Optimization 97% Percentage
Inventory Turnover Rate Before Optimization 6 Turns per year
After Optimization 8 Turns per year
Production Efficiency
Production Cycle Time Before Optimization 8h Hours
After Optimization 6h Hours
Defect Rate Before Optimization 3% Percentage
After Optimization 1.5% Percentage
Financial Impact
Cost Reduction Total Cost Savings USD 750,000 Dollars
Revenue Increase Sales Revenue Growth 10% Percentage

5.2. Retail Sector: Demand Volatility Management

The retail sector operates in a dynamic environment with shifting consumer prefer-
ences and erratic demand trends. A major retail chain adopted the suggested framework
as a way to improve their demand forecasting accuracy as well as to optimize inventory
management to solve the difficulties caused by demand volatility. For the retail industry to
avoid overstocking, understocking, and the ensuing financial losses, it is crucial to estimate
consumer demand appropriately [45,46]. The retail chain understood that conventional
demand forecasting techniques frequently fell short of accurately reflecting the complex
interactions between consumer behavior, outside variables, and supply chain performance.
In response, they used the suggested framework to fully utilize the power of machine
learning and predictive analytics.
The first step in the implementation process was the collection of previous sales
information, consumer purchase trends, and pertinent outside data sources. With the use
of these data, predictive models were trained that could detect complicated demand trends,
seasonality, and the effects of outside variables like promotions or the state of the economy.
These models were built with mechanisms for ongoing learning to adapt and advance
over time. The retail chain was able to create more precise and timely demand estimates
thanks to the predictive model’s real-time data integration features. The approach offered a
comprehensive picture of demand drivers by examining both historical trends and current
external influences, resulting in enhanced estimates that took into account unexpected
changes in consumer behavior.
Additionally, the more precise demand forecasts enabled the retail chain to make better
strategic decisions, enabling it to optimize pricing, promotions, and product releases. This
led to higher sales income, happier customers, and a competitive advantage in the market.
The paradigm can completely transform demand volatility management, as demon-
strated by the retail sector’s performance. Organizations may manage the intricacies of
demand variations and take advantage of new opportunities to spur growth and prof-
itability by utilizing predictive analytics, real-time data integration and adaptive learning
processes. This case study serves as an example of how the suggested framework might
operate as a cornerstone for achieving supply chain agility in a sector where customer
behavior and market dynamics are subject to fast change. Table 10 shows the retail sector
demand volatility management findings.
Sustainability 2023, 15, 15088 19 of 26

Table 10. Retail sector demand volatility management findings.

Parameters Metrics/Variables Data (Sample) Measurement Units

Demand Forecast Accuracy
Mean Absolute Percentage Error Before Optimization 15% Percentage
After Optimization 8% Percentage
Inventory Management
Inventory Turnover Rate Before Optimization 4.2 Turns per year
After Optimization 5.8 Turns per year
Safety Stock Levels Average Safety Stock Levels (Before) 500 units Units
Average Safety Stock Levels (After) 350 units Units
Order Fulfillment
Order Lead Time Average Lead Time (Before) 5 days Days
Average Lead Time (After) 3 days Days
Financial Impact
Cost Reduction Inventory Holding Cost Savings USD 200,000 Dollars
Revenue Increase Sales Revenue Growth 8% Percentage

The proposed approach increased demand forecast accuracy substantially, lowering the
mean absolute percentage error from 15% to 8%. This shows that inventory should be planned
more precisely. Following the optimization, the inventory turnover rate increased from 4.2
to 5.8 turns per year, showing improved inventory and working capital management. After
the optimization, safety stock levels were decreased from an average of 500 units to 350 units,
indicating an improved balance between service levels and inventory costs. The average order
lead time was reduced from 5 to 3 days, demonstrating improved responsiveness to consumer
demands. The optimization technique resulted in a USD 200,000 decrease in inventory
holding costs, which contributed to increased profitability. Sales revenue increased by 8% as
a result of improved customer satisfaction, reduced stockouts, and enhanced marketplace
competitiveness. In general, the data show that the proposed approach produced enhanced
demand forecasting precision, effective inventory control, and cost reduction as well as
revenue growth, indicating its success in managing demand volatility.

5.3. Pharmaceutical Industry: Regulatory Compliance and Supplier Integrity

Sustaining supply chain integrity, high-quality products, and regulatory compliance
are of the utmost significance in the highly regulated pharmaceutical sector. A top pharma-
ceutical corporation accepted the suggested structure as a complete response to the complex
problems of guaranteeing regulatory conformity and supplier integrity. Disruptions re-
sulting from supplier non-compliance as well as integrity problems in the pharmaceutical
industry can have serious repercussions, such as regulatory fines, product recalls, and
reputational damages. Knowing how important these concerns were, the pharmaceutical
business used predictive analytics and machine learning to proactively detect potential
hazards and increase the resilience of its supply chain [47–50].
The integration of several data sources, including supplier performance records,
compliance with regulations data, quality control policies, and real-time external data
streams, served as the starting point for the implementation procedure. The foundation for
developing predictive models that can spot patterns suggestive of supplier hazards and
regulatory non-compliance was this extensive dataset. The organization’s supply chain, as
well as its quality assurance teams, were able to forecast and evaluate supplier-related risks
before they turned into serious problems by utilizing historical data and real-time updates.
The models were honed to identify minute trends that would point to deviations from legal
requirements or potential issues with supplier integrity.
Sustainability 2023, 15, 15088 20 of 26

The results were remarkable. The pharmaceutical company was capable of proactively
identifying suppliers who were engaging in a pattern of non-compliance or who might have
integrity problems. Because of this early detection, the organization was able to take quick
remedial action, such as improved supplier monitoring, focused audits, and collaboration,
to address shortcomings before they got out of hand. Furthermore, the framework’s
processes for continuous learning made sure that the models changed over time, improving
their capacity to identify newly emerging supplier-related risks. As a consequence, the
organization’s capacity to maintain supplier integrity as well as regulatory compliance was
much improved, resulting in fewer regulatory fines, a decreased risk of product recalls,
and better links with important suppliers. In the final analysis, the application of the
suggested methodology demonstrated its potential to transform risk management in the
pharmaceutical sector. The pharmaceutical firm managed to meet regulatory standards
as well as improve the general quality and integrity of its supply chain by proactively
identifying and addressing supplier-related issues. Table 11 shows the pharmaceutical
industry regulatory compliance and supplier integrity findings.
The proposed approach produced a 97% compliance rate and a considerable decrease
in regulatory breaches from six to one. This demonstrates increased supplier integrity and
regulatory compliance. Following the optimization, the product defect rate dropped from
2.5% to 0.8%, indicating improved product quality and lower compliance risks. After the
optimization, the number of regulatory audit findings decreased from 15 to 2, suggesting
a higher degree of regulatory compliance as well as integrity. The optimization technique
resulted in significant cost savings, with regulatory fines being reduced from USD 700,000
to USD 80,000, helping financial stability. The average supplier response time was reduced
from 12 days to 4 days, demonstrating better operational efficiency as well as responsiveness.
Compliance training hours were extended to 150 h, demonstrating a commitment to continual
improvement and staff compliance training. Ultimately, the findings show that the proposed
approach improved supplier performance, product quality, compliance, and cost reduction as
well as operational efficiency while adhering to industry norms and regulations.

Table 11. Pharmaceutical industry regulatory compliance and supplier integrity findings.

Parameters Metrics/Variables Data (Sample) Measurement Units

Supplier Performance
Supplier Regulatory Compliance Compliance Rate 97% Percentage
Number of Regulatory Violations (Before) 6 Count
Number of Regulatory Violations (After) 1 Count
Quality Control
Product Defect Rate Defective Product Rate (Before) 2.5% Percentage
Defective Product Rate (After) 0.8% Percentage
Audit and Inspection
Regulatory Audit Findings Number of Findings (Before) 15 Count
Number of Findings (After) 2 Count
Financial Impact
Cost of Non-Compliance Cost of Regulatory Fines (Before) USD 700,000 Dollars
Cost of Regulatory Fines (After) USD 80,000 Dollars
Operational Efficiency
Lead Time Reduction Average Supplier Response Time (Before) 12 days Days
Average Supplier Response Time (After) 4 days Days
Compliance Improvement
Compliance Training Number of Training Hours 150 h Hours
Training Cost USD 30,000 Dollars
Sustainability 2023, 15, 15088 21 of 26

5.4. Technology Sector: Geopolitical Risk Mitigation

The technology industry operates in a global environment where trade dynamics and
geopolitical conflicts can have a big impact on supply chain operations. A well-known tech-
nology company used the suggested methodology to improve its supply chain flexibility
and adaptability to manage the complexities of geopolitical risks and guarantee continuous
business continuity. Geopolitical events, trade conflicts, and changes in regulations can
cause supply chain bottlenecks, delays in product delivery, and financial losses in the
technology sector. The technology corporation intended to use predictive analytics and
adaptive decision-making to strengthen its supply chain against the risks of geopolitics
after realizing the potential vulnerability of its global supply network [51–53].
The first step in the implementation process was gathering data from a wide range
of sources, including information on political events, economic indicators, agreements on
trade, and market volatility indices. These data served as the foundation for developing
predictive models that could analyze geopolitical patterns and foresee future supply chain
disruptions. The algorithm’s ability to dynamically integrate real-time data is crucial
because it enabled the technology company to quickly react to changing geopolitical
events. The platform gave early signals of potential dangers by continuously analyzing
real-time data streams along with historical patterns, enabling the organization to make
informed decisions as well as launch proactive mitigation actions. Impressive outcomes
were obtained.
This case study serves as an illustration of how the suggested framework might enable
technology-related organizations to proactively reduce the impact of geopolitical risks on
their supply chains. The technology company managed to foresee threats by utilizing
predictive analytics as well as real-time data integration, and they were also able to develop
strategic solutions that protected their operations and kept customers satisfied even in
the face of geopolitical concerns. Table 12 shows the technology sector geopolitical risk
mitigation findings.

Table 12. Technology sector geopolitical risk mitigation findings.

Parameters Metrics/Variables Data (Sample) Measurement Units

Geopolitical Risk Assessment
Risk Severity Score Before Mitigation 8.5 Scale (1–10)
After Mitigation 4.2 Scale (1–10)
Supply Chain Disruptions
Number of Disruptions (Before) Geopolitical Incidents 15 Count
Supply Chain Delays (Before) 8 Count
Number of Disruptions (After) Geopolitical Incidents 3 Count
Supply Chain Delays (After) 1 Count
Financial Impact
Cost of Disruptions (Before) Financial Loss (Before) USD 1,200,000 Dollars
Cost of Disruptions (After) Financial Loss (After) USD 150,000 Dollars
Risk Mitigation Effectiveness
Risk Reduction Ratio 55% Percentage
Operational Continuity
Downtime Duration (Before) Operational Downtime (Before) 36 h Hours
Downtime Duration (After) Operational Downtime (After) 6h Hours

The proposed approach resulted in a considerable decrease in risk severity, from

8.5 to 4.2, showing enhanced geopolitical risk assessment and management. Following the
mitigation, the frequency of geopolitical incidents as well as supply chain delays decreased
significantly, demonstrating the efficiency of the risk management techniques. After the
Sustainability 2023, 15, 15088 22 of 26

mitigation, the cost of disruptions was significantly reduced, with financial losses falling
from USD 1,200,000 to USD 150,000. The 55% risk reduction ratio indicates the effectiveness
of the mitigation techniques in lowering the total geopolitical risk. After the mitigation, the
downtime length was reduced from 36 h to 6 h, ensuring enhanced operational continuity
and fewer business interruptions. In the end, the findings show that the proposed approach
in the technology sector led to enhanced risk assessment, fewer supply chain disruptions,
significant cost savings, and increased risk mitigation effectiveness as well as greater
operational continuity, making it an advantageous approach for lowering geopolitical risks
in the technology industry.
Beyond applications particular to one industry, the suggested framework provides
cross-industry advantages that are seen in the faster response times and more
adaptability—both essential elements of successful supply chain risk management. The
overall benefits of the framework in promoting agility across many sectors are examined
in this section. The capacity to react quickly and effectively to disturbances is essential to
preserving operational continuity as well as customer satisfaction in almost every business.
The suggested methodology enables organizations to significantly cut response times when
handling possible risks by integrating real-time data sources, predictive analytics, and
adaptive decision-making. The sophisticated early warnings as well as forecasting abilities
that the framework offers serve as an illustration of this. These tools let supply chain
participants make timely, well-informed decisions, preventing the escalation of possible
disruptions and lessening their effects.
The findings from the case studies reported in this study indicate the suggested pre-
dictive analytics and machine learning framework’s extensive generalizability as well as
applicability across a wide range of sectors. These case studies covered the automotive,
retail, pharmaceutical, and technology industries, demonstrating how the framework im-
proves supply chain agility and risk management. The approach regularly generated
substantial improvements in supply chain efficiency throughout each industry, whether in
terms of supplier reliability, demand forecasting accuracy, regulatory compliance, or geopo-
litical risk reduction. Furthermore, the framework’s adaptability, as demonstrated by its
continuous learning processes and real-time data integration, guarantees that it can evolve
to handle the particular issues encountered by diverse industries. The findings highlight
how this comprehensive strategy can be implemented in a variety of situations, industries,
and supply chain complexities, emphasizing its transformational potential in revolutioniz-
ing risk management practices and fostering resilience across global supply networks.

6. Discussion
This study adds a novel and innovative viewpoint to the field of supply chain risk
management by presenting a comprehensive framework that smoothly combines predictive
analytics and machine learning into the risk mitigation process. The focus on real-time
monitoring as well as decision-making, a revolutionary strategy that enables organizations
to switch from reactive risk management to proactive as well as anticipatory techniques, is
what distinguishes this study from others. The framework gives organizations the ability
to spot possible disturbances early on and respond quickly by utilizing predictive models
built from historical and real-time data. This improves supply chain agility as well as
resilience. This research also advances the subject by demonstrating the usefulness of
the suggested framework through a variety of case studies in various sectors. These case
studies show the framework’s adaptability and effectiveness in managing a variety of
disturbances, such as demand volatility as well as geopolitical uncertainty.
This study emphasizes the need for ongoing model accuracy by promoting dynamic
model updating based on fresh data to keep prediction models useful for real-time mon-
itoring. This strategy fits with the flexible, dynamic character of contemporary supply
chains. This study work is innovative in that it approaches supply chain risk management
holistically and in the future. It not only makes use of cutting-edge technologies but also
provides organizations with a realistic road map for navigating an unstable and increas-
Sustainability 2023, 15, 15088 23 of 26

ingly complicated business environment while maintaining operational continuity as well

as customer satisfaction.
The application of predictive analytics and machine learning to supply chain risk
management opens up a variety of opportunities and problems that need to be carefully
considered. This part begins an investigation of the complex issues and wide-ranging
effects that the incorporation of these cutting-edge technologies raises. We learn how
they might modify traditional risk management concepts by exploring their revolutionary
potential. At the same time, we address the issues that come up along this inventive
journey, providing insights into how businesses might overcome these difficulties to fully
realize the advantages of this strategy. We acquire a detailed picture of the significant
influence and factors that support the widespread use of predictive analytics and machine
learning in the quest for more flexible and resilient supply chain operations through this
thorough investigation.
The framework is designed to accommodate a variety of business sizes, from small-
and medium-sized organizations (SMEs) to major multinationals. A key component of
the framework is scalability, which makes it possible for organizations of all sizes to
adopt and use it successfully. Smaller businesses can concentrate on particular facets of the
framework, gradually increasing their use as they expand, while larger businesses can adopt
it extensively throughout intricate supply chain networks. Additionally, the framework’s
adaptability covers supply chain complexity. The framework’s modular design enables the
choice and integration of pertinent components regardless of how complicated or simple a
supply chain may be. This versatility guarantees that organizations may profit from the
framework regardless of the complicated nature of their supply chains.

6.1. Implications
The combination of machine learning and predictive analytics has significant implica-
tions for supply chain risk management. The main benefit of this strategy is that it makes
risk management proactive rather than reactive. Organizations can start prompt actions
and reduce the magnitude of interruptions and their consequent impact on operations as
well as customer satisfaction by spotting possible hazards in real time or even before they
emerge. Additionally, this methodology offers a data-driven foundation for intelligent
decision-making, enabling organizations to manage resources and dynamically change
plans. As a result, supply chain agility is greatly increased, allowing businesses to take
advantage of opportunities and overcome risks.

6.2. Challenges
However, there are several difficulties in incorporating machine learning and pre-
dictive analytics into supply chain risk management. Data accessibility and quality are
major issues, since successful implementation depends on having access to precise, timely
data from a variety of internal as well as external sources. A further layer of complexity
is added by ensuring data security and privacy while adhering to regulatory regulations.
Additionally, deploying complex predictive models requires technical know-how, which
could be a hurdle for businesses lacking data science capability. Challenges include model
interpretability and explainability because supply chain decisions frequently demand ratio-
nale and comprehension that go beyond black-box algorithms. Additionally, the dynamic
nature of supply networks suggests that models must be continuously improved and up-
dated to be useful under changing circumstances. Another factor is balancing the trade-off
between computational resources and model complexity. Finally, since predictive analytics
and machine learning may disrupt conventional risk management practices, organizational
buy-in and cultural changes are crucial in embracing this paradigm shift.

7. Conclusions
We have provided an innovative approach for supply chain risk management in this
study that seamlessly combines predictive analytics and machine learning. The transforma-
Sustainability 2023, 15, 15088 24 of 26

tional potential of this technology to improve supply chain agility has been shown by our
research. Organizations can guarantee operational continuity, optimize resource allocation,
and ultimately improve customer satisfaction by proactively recognizing and reducing
risks in real time. This approach was successfully applied to a variety of industries, as
seen in our case studies, and this has highlighted its adaptability in dealing with a range
of disturbances, including demand volatility as well as geopolitical uncertainty. Looking
ahead, several directions for additional study have appeared in response to the changing
supply chain management scenario. The appropriate balances between algorithm com-
plexity as well as computational efficiency have to be achieved to guarantee the viability
of predictive models for real-time application. Furthermore, supply chain experts may
feel more confident as a result of attempts to make machine learning models easier to
comprehend. Investigating how to incorporate cutting-edge technologies like blockchain
and the internet of things (IoT) into our framework might further improve its reliability and
openness. Additionally, studying organizational dynamics and cultural shifts necessary
for successful adoption, as well as comprehending human decision-making within the
framework of predictive analytics as well as machine learning, constitute equally important
research areas.

Funding: This research received no external funding.

Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: Not applicable.
Conflicts of Interest: The author declares no conflict of interest.

References
1. Golgeci, I.; Yildiz, H.E.; Andersson, U.R. The rising tensions between efficiency and resilience in global value chains in the
post-COVID-19 world. Transnatl. Corp. J. 2020, 27, 127–141. [CrossRef]
2. Mentzer, J.T.; Stank, T.P.; Myers, M.B. Why Global Supply Chain Management? In Handbook of Global Supply Chain Management;
Sage Publications: Thousand Oaks, CA, USA, 2007; pp. 1–16.
3. Challies, E.; Newig, J.; Lenschow, A. What role for social–ecological systems research in governing global teleconnections?
Glob. Environ. Change 2014, 27, 32–40. [CrossRef]
4. Lazzarini, S.; Chaddad, F.; Cook, M. Integrating supply chain and network analyses: The study of netchains. J. Chain Netw. Sci.
2001, 1, 7–22. [CrossRef]
5. Starr, R.; Newfrock, J.; Delurey, M. Enterprise resilience: Managing risk in the networked economy. Strategy Bus. 2003, 30, 70–79.
6. Pu, G.; Li, S.; Bai, J. Effect of supply chain resilience on firm’s sustainable competitive advantage: A dynamic capability perspective.
Environ. Sci. Pollut. Res. 2023, 30, 4881–4898. [CrossRef] [PubMed]
7. Chen, C.J. Developing a model for supply chain agility and innovativeness to enhance firms’ competitive advantage. Manag.
Decis. 2019, 57, 1511–1534. [CrossRef]
8. Bird, R.C. VUCA and the Legal Environment of Business; University of Connecticut School of Business Research Paper No. 18-09;
University of Connecticut: Storrs, CT, USA, 2018.
9. Correa-Baena, J.P.; Hippalgaonkar, K.; van Duren, J.; Jaffer, S.; Chandrasekhar, V.R.; Stevanovic, V.; Wadia, C.; Guha, S.;
Buonassisi, T. Accelerating materials development via automation, machine learning, and high-performance computing. Joule
2018, 2, 1410–1420. [CrossRef]
10. Bharadiya, J.P. Machine Learning and AI in Business Intelligence: Trends and Opportunities. Int. J. Comput. 2023, 48, 123–134.
11. Gunasekaran, A.; Lai, K.H.; Cheng, T.E. Responsive supply chain: A competitive strategy in a networked economy. Omega 2008,
36, 549–564. [CrossRef]
12. Rawat, R. A Systematic Review of Blockchain Technology Use in E-Supply Chain in Internet of Medical Things (Iomt). Int. J.
Comput. Inf. Manuf. 2022, 2, 2. [CrossRef]
13. Perifanis, N.A.; Kitsios, F. Investigating the influence of artificial intelligence on business value in the digital era of strategy:
A literature review. Information 2023, 14, 85. [CrossRef]
14. Srivastava, S.; Optimizing Supply Chain with AI and Analytics. Appinventiv. 2023. Available online: https://appinventiv.com/
blog/ai-in-supply-chain-analytics/ (accessed on 19 August 2023).
15. Wong, L.W.; Tan, G.W.H.; Ooi, K.B.; Lin, B.; Dwivedi, Y.K. Artificial intelligence-driven risk management for enhancing supply
chain agility: A deep-learning-based dual-stage PLS-SEM-ANN analysis. Int. J. Prod. Res. 2022, 1–21. [CrossRef]
Sustainability 2023, 15, 15088 25 of 26

16. Giannakis, M.; Louis, M. A multi-agent based system with big data processing for enhanced supply chain agility. J. Enterp. Inf.
Manag. 2016, 29, 706–727. [CrossRef]
17. Jayender, P.; Kundu, G.K. Intelligent ERP for SCM agility and graph theory technique for adaptation in automotive industry in
India. Int. J. Syst. Assur. Eng. Manag. 2021, 1–22. [CrossRef]
18. Shamout, M.D. Supply chain data analytics and supply chain agility: A fuzzy sets (fsQCA) approach. Int. J. Organ. Anal. 2020, 28,
1055–1067. [CrossRef]
19. Schroeder, M.; Lodemann, S. A systematic investigation of the integration of machine learning into supply chain risk management.
Logistics 2021, 5, 62. [CrossRef]
20. Lee, C.H.; Yang, H.C.; Wei, Y.C.; Hsu, W.K. Enabling blockchain based scm systems with a real time event monitoring function for
preemptive risk management. Appl. Sci. 2021, 11, 4811. [CrossRef]
21. Ganesh, A.D.; Kalpana, P. Future of artificial intelligence and its influence on supply chain risk management–A systematic review.
Comput. Ind. Eng. 2022, 169, 108206. [CrossRef]
22. Ivanov, D.; Dolgui, A. A digital supply chain twin for managing the disruption risks and resilience in the era of Industry 4.0.
Prod. Plan. Control 2021, 32, 775–788. [CrossRef]
23. Mageto, J. Big data analytics in sustainable supply chain management: A focus on manufacturing supply chains. Sustainability
2021, 13, 7101. [CrossRef]
24. Dolgui, A.; Ivanov, D. 5G in digital supply chain and operations management: Fostering flexibility, end-to-end connectivity and
real-time visibility through internet-of-everything. Int. J. Prod. Res. 2022, 60, 442–451. [CrossRef]
25. Constante, F.; Silva, F.; Pereira, A. DataCo Smart Supply Chain for Big Data Analysis; Mendeley: London, UK, 2019.
26. Seyedan, M.; Mafakheri, F. Predictive big data analytics for supply chain demand forecasting: Methods, applications, and research
opportunities. J. Big Data 2020, 7, 1–22. [CrossRef]
27. Sapankevych, N.I.; Sankar, R. Time series prediction using support vector machines: A survey. IEEE Comput. Intell. Mag. 2009, 4,
24–38. [CrossRef]
28. Kuster, C.; Rezgui, Y.; Mourshed, M. Electrical load forecasting models: A critical systematic review. Sustain. Cities Soc. 2017, 35,
257–270. [CrossRef]
29. Deb, C.; Zhang, F.; Yang, J.; Lee, S.E.; Shah, K.W. A review on time series forecasting techniques for building energy consumption.
Renew. Sustain. Energy Rev. 2017, 74, 902–924. [CrossRef]
30. Lippi, M.; Bertini, M.; Frasconi, P. Short-term traffic flow forecasting: An experimental comparison of time-series analysis and
supervised learning. IEEE Trans. Intell. Transp. Syst. 2013, 14, 871–882. [CrossRef]
31. Bhuyan, M.H.; Bhattacharyya, D.K.; Kalita, J.K. Network anomaly detection: Methods, systems and tools. IEEE Commun. Surv.
Tutor. 2013, 16, 303–336. [CrossRef]
32. Barford, P.; Kline, J.; Plonka, D.; Ron, A. A signal analysis of network traffic anomalies. In Proceedings of the 2nd ACM SIGCOMM
Workshop on Internet Measurment, New York, NY, USA, 6—8 November 2002; pp. 71–82.
33. Pang, G.; Shen, C.; Cao, L.; Hengel AV, D. Deep learning for anomaly detection: A review. ACM Comput. Surv. (CSUR) 2021, 54,
1–38. [CrossRef]
34. McBride, S.K.; Bostrom, A.; Sutton, J.; de Groot, R.M.; Baltay, A.S.; Terbush, B.; Bodin, P.; Dixon, M.; Holland, E.; Arba, R.; et al.
Developing post-alert messaging for ShakeAlert, the earthquake early warning system for the West Coast of the United States of
America. Int. J. Disaster Risk Reduct. 2020, 50, 101713. [CrossRef]
35. Antonioli, P.; Fienberg, R.T.; Fleurot, F.; Fukuda, Y.; Fulgione, W.; Habig, A.; Heise, J.; McDonald, A.B.; Mills, C.; Namba, T.; et al.
SNEWS: The SuperNova early warning system. New J. Phys. 2004, 6, 114. [CrossRef]
36. Hu, W.; Chen, T.; Shah, S.L. Detection of frequent alarm patterns in industrial alarm floods using itemset mining methods.
IEEE Trans. Ind. Electron. 2018, 65, 7290–7300. [CrossRef]
37. Gunasekaran, A.; Kobu, B. Performance measures and metrics in logistics and supply chain management: A review of recent
literature (1995–2004) for research and applications. Int. J. Prod. Res. 2007, 45, 2819–2840. [CrossRef]
38. Cho, D.W.; Lee, Y.H.; Ahn, S.H.; Hwang, M.K. A framework for measuring the performance of service supply chain management.
Comput. Ind. Eng. 2012, 62, 801–818. [CrossRef]
39. Azadegan, A.; Syed, T.A.; Blome, C.; Tajeddini, K. Supply chain involvement in business continuity management: Effects
on reputational and operational damage containment from supply chain disruptions. Supply Chain Manag. Int. J. 2020, 25,
747–772. [CrossRef]
40. Sharma, P.; Jain, S.; Gupta, S.; Chamola, V. Role of machine learning and deep learning in securing 5G-driven industrial IoT
applications. Ad Hoc Netw. 2021, 123, 102685. [CrossRef]
41. Adadi, A.; Berrada, M. Peeking inside the black-box: A survey on explainable artificial intelligence (XAI). IEEE Access 2018, 6,
52138–52160. [CrossRef]
42. Yathiraju, N. Investigating the use of an Artificial Intelligence Model in an ERP Cloud-Based System. Int. J. Electr. Electron.
Comput. 2022, 7, 1–26. [CrossRef]
43. Bailey, D.; De Ruyter, A.; Hearne, D.; Ortega-Argilés, R. Shocks, resilience and regional industry policy: Brexit and the automotive
sector in two Midlands regions. Reg. Stud. 2023, 57, 1141–1155. [CrossRef]
Sustainability 2023, 15, 15088 26 of 26

44. Chervenkova, T.; Ivanov, D. Adaptation strategies for building supply chain viability: A case study analysis of the global
automotive industry re-purposing during the COVID-19 pandemic. Transp. Res. Part E Logist. Transp. Rev. 2023, 177,
103249. [CrossRef]
45. Alqhatani, A.; Ashraf, M.S.; Ferzund, J.; Shaf, A.; Abosaq, H.A.; Rahman, S.; Irfan, M.; Alqhtani, S.M. 360◦ Retail Business
Analytics by Adopting Hybrid Machine Learning and a Business Intelligence Approach. Sustainability 2022, 14, 11942. [CrossRef]
46. Njomane, L.; Telukdarie, A. Impact of COVID-19 food supply chain: Comparing the use of IoT in three South African supermarkets.
Technol. Soc. 2022, 71, 102051. [CrossRef]
47. Saha, E.; Rathore, P.; Parida, R.; Rana, N.P. The interplay of emerging technologies in pharmaceutical supply chain performance:
An empirical investigation for the rise of Pharma 4.0. Technol. Forecast. Soc. Change 2022, 181, 121768. [CrossRef]
48. Abdallah, S.; Nizamuddin, N. Blockchain-based solution for pharma supply chain industry. Comput. Ind. Eng. 2023, 177,
108997. [CrossRef]
49. Trautmann, L.; Hübner, T.; Lasch, R. Blockchain concept to combat drug counterfeiting by increasing supply chain visibility. Int. J.
Logist. Res. Appl. 2022, 1–27. [CrossRef]
50. Kartoglu, U.; Ames, H. Ensuring quality and integrity of vaccines throughout the cold chain: The role of temperature monitoring.
Expert Rev. Vaccines 2022, 21, 799–810. [CrossRef]
51. Ozdemir, D.; Sharma, M.; Dhir, A.; Daim, T. Supply chain resilience during the COVID-19 pandemic. Technol. Soc. 2022, 68,
101847. [CrossRef]
52. Panahi, R.; Gargari, N.S.; Lau, Y.Y.; Ng, A.K. Developing a resilience assessment model for critical infrastructures: The case of
port in tackling the impacts posed by the COVID-19 pandemic. Ocean Coast. Manag. 2022, 226, 106240. [CrossRef]
53. Tseng, M.L.; Bui, T.D.; Lim, M.K.; Fujii, M.; Mishra, U. Assessing data-driven sustainable supply chain management indicators for
the textile industry under industrial disruption and ambidexterity. Int. J. Prod. Econ. 2022, 245, 108401. [CrossRef]

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.