Supply Chain

Hindawi
Mathematical Problems in Engineering

Volume 2021, Article ID 1476043, 14 pages
https://doi.org/10.1155/2021/1476043
Review Article
Application of Machine Learning in Supply Chain Management:
A Comprehensive Overview of the Main Areas
Erfan Babaee Tirkolaee ,1 Saeid Sadeghi,2 Farzaneh Mansoori Mooseloo,3

Hadi Rezaei Vandchali,4 and Samira Aeini5
1
Department of Industrial Engineering, Istinye University, Istanbul, Turkey
2
Department of Industrial Management, University of Tehran, Tehran, Iran
3
Department of Industrial Management, University of Hormozgan, Bandar-Abbas, Iran
4
Australian Maritime College, University of Tasmania, Launceston, Australia
5
Department of Project and Construction Management, Noore Touba University, Tehran, Iran
Correspondence should be addressed to Erfan Babaee Tirkolaee; [email protected]
Received 13 April 2021; Revised 12 May 2021; Accepted 5 June 2021; Published 22 June 2021
Academic Editor: Ali Ahmadian
Copyright © 2021 Erfan Babaee Tirkolaee et al. (is is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is
properly cited.
In today’s complex and ever-changing world, concerns about the lack of enough data have been replaced by concerns about too
much data for supply chain management (SCM). (e volume of data generated from all parts of the supply chain has changed the
nature of SCM analysis. By increasing the volume of data, the efficiency and effectiveness of the traditional methods have
decreased. Limitations of these methods in analyzing and interpreting a large amount of data have led scholars to generate some
methods that have high capability to analyze and interpret big data. (erefore, the main purpose of this paper is to identify the
applications of machine learning (ML) in SCM as one of the most well-known artificial intelligence (AI) techniques. By developing
a conceptual framework, this paper identifies the contributions of ML techniques in selecting and segmenting suppliers, predicting
supply chain risks, and estimating demand and sales, production, inventory management, transportation and distribution,
sustainable development (SD), and circular economy (CE). Finally, the implications of the study on the main limitations and
challenges are discussed, and then managerial insights and future research directions are given.
1. Introduction Intelligent use of these data and the extraction of infor-

mation and knowledge of them have a great potential to
In today’s competitive environment, the advent of the in- benefit. Explosive growth in the volume of data from various
formation technology revolution, economic globalization, parts of SCM has forced companies to develop and im-
and high customer expectations has caused various changes plement new technologies which are able to rapidly and
in companies’ supply chain management (SCM), high- intelligently interpret a large amount of data [5, 6] as the
lighting the importance of competition among supply chains traditional decision support system cannot face big data
rather than companies [1–3]. SCM is the active integration of satisfactorily. Hence, in the big data era, supply chain
supply chain activities from original suppliers through end- professionals are seeking to handle big data to reach smart
users that provides services, products, and information that supply chains.
maximize customer value and achieve sustainable compet- Artificial intelligence (AI) methods are the most ap-
itive advantage [4]. In the era of big data, a large amount of propriate methods to cope with this big data-related chal-
interactive data are routinely created, collected, and archived lenge. Machine learning (ML) techniques can be considered
in different process industries, and these data are really an as the popular subdisciplines in AI which identify and ex-
important asset in process operation, control, and design. tract automatically the patterns among variables using large
2629, 2021, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1155/2021/1476043 by Universitäts- und Landesbibliothek Münster, Wiley Online Library on [07/10/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2 Mathematical Problems in Engineering
datasets [7]. ML algorithms can detect unknown patterns develops a framework in which the most commonly used
among data, generate new insights, and direct researchers ML algorithms in managing different areas of the supply
towards suitable destinations. ML techniques can be used in chain will be discussed. (e main contributions of the paper
a different context, including manufacturing, operations, are outlined as follows:
healthcare, and housing [8]. Moreover, ML is widely used in
(i) By comparing the efficiency of traditional and AI
managing different areas and aspects of the supply chain.
methods in facing big data
Recently, ML algorithms and their applications in managing
the supply chain have gained interest among researchers. (ii) By reviewing, summarizing, and classifying the
Weaknesses of traditional methods in analyzing big data most frequently used AI methods in SCM
have led scholars to utilize ML methods that have high (iii) By providing a detailed framework to explain the
capabilities to analyze and interpret large volumes of data. outputs of the application of ML techniques in
(ere are many reasons to replace traditional methods with supplier selection and segmentation, predicting
ML techniques. First, traditional methods cannot handle supply chain risks, and estimating demand and
nonlinear problems that are widespread in real-world supply sales, production, inventory management, trans-
chains. ML techniques can solve this nonlinear problem portation and distribution, sustainable development
satisfactorily. Second, traditional methods are incapable of (SD), and circular economy (CE)
dealing with large and unstructured data that come from
In this regard, Section 2illustrates a review of ML
different areas of the supply chain, where ML techniques
techniques. (e applications of ML techniques in different
precisely have been developed to deal with big and un-
parts of supply chain are presented in Section 3. Finally,
structured data. (ird, ML techniques are far stronger than
conclusions, managerial insights, and future directions are
traditional statistical methods in recognizing and predicting
presented in Section 4.
the most effective factors on supply chain performance.
(us, ML can be considered as an important tool to help
companies to analyze the big data in their SCM. 2. Machine Learning
Seeking new information and knowledge using big data
analytics can help to predict future trends, which can lead to ML techniques are used to teach machines how to auto-
informed decisions in various parts of SCM, making a matically handle a large volume of data more efficiently.
worthy paradigm for companies. Emerging technologies Sometimes the extraction of patterns or information from
such as the Internet of things (IoT), blockchain, and ad- the large volume of data and interpreting them are im-
vanced deep learning (ADL) techniques can equip supply possible by traditional techniques [15]. (e abundance of
chains with a self-aware mechanism in which different datasets available has caused a rising demand for ML
operating objects are connected simultaneously. SCM is techniques. ML techniques are widely applied in different
really in need of such a self-adaptive smart system to manage industries from medicine to the military to discover and
the fluctuations in customer demand. Increasing the extract knowledge and information from the data. Many
transparency and connectivity of the supply chain via IoT studies have been done by mathematicians and program-
and blockchain technologies can be a good sign to face the mers that resulted in the development of various ML al-
customers’ ever-changing demands. (is can provide a gorithms [16]. In this paper, a review of the application of
stream of real-time data and information between various machine learning in SCM was considered. Several studies
parts of supply chains anywhere in the world. have mentioned advantages of application of ML techniques
(ere is a strong need in the literature to investigate in demand and sales estimation [17–19], transportation and
various applications of ML techniques in different parts of distribution [20–22], production [23–25], inventory control
supply chains as the majority of the works have considered [22, 26], supplier selection and segmentation [27–30], and so
one, two, or limited areas of the supply chain. For instance, on. Some of the most famous learning algorithms with a
Piramuthu [9] applied ML techniques to design an auto- brief description are shown in Table 1.
mated SCM framework. Cavalcante et al. [10] proposed a
supervised ML technique for resilient supplier selection. Zhu 3. Machine Learning in Supply
et al. [11] used China’s SME to implement ML techniques in Chain Management
predicting financial risks in supply chains. Priore et al. [12]
applied ML techniques to find appropriate replenishment Despite a large number of publications in the field of ML and
policies in SCM. Baryannis et al. [13] employed ML to SCM separately, the number of publications that have paid
predict supply chain risks. Darvazeh et al. [5] conducted a to the applications of ML algorithms in managing a supply
review on ML in SCM. Bai et al. [14] used a multicriteria chain is not adequate [38]. On the other hand, there is no
decision-making (MCDM) and ML hybrid approach to enough connection between researchers and practitioners in
segment suppliers based on environmental factors. How- this field. It might be caused by inadequate knowledge of
ever, there is less attention to accomplish a comprehensive practitioners about the power and advantages of ML algo-
study to consider ML applications in different related aspects rithms in solving SCM problems. In this section, a review of
of the supply chain, which can affect the understanding of the application of the most famous ML algorithms in
how these valuable techniques can be effectively used in managing supply chain-related issues including supplier
managing different aspects of SCM. (erefore, this paper selection, supplier segmentation, supply chain risk
Mathematical Problems in Engineering 3
Table 1: Frequently used ML algorithms.

Type of learning Algorithm Description
Using related values, decision trees (DTs) will categorize attributes in different groups
Decision tree
which can be applied for classification purposes [31].
Naı̈ve Bayes Naı̈ve Bayes can be best applied to cluster and classify objects [32].
Working on the margin calculations, support vector machine (SVM) can be best
Support vector machine
applied for classification purposes [33].
In K-nearest neighbor (KNN), the learner usually uses the training data. When the test
Supervised learning
data are introduced to the learner, it compares both the data. Here, K most correlated
K-nearest neighbor
data are taken from a training set. (e majority of K is taken which serves as the new
class for the test data [34].
Using supervised neural network (SNN), the predicted output and actual output will
Supervised neural network be compared, and according to the identified error, the parameters will be modified
and considered as the input into the neural network again [15].
Using similarity of the clusters of data,K-means (KM) clustering algorithm definesK
K-means clustering
clusters in which the center of the clusters is the mean of the values [35].
Principal component Principal component analysis (PCA) can provide faster and easier computations as it
Unsupervised
analysis reduces the dimension of the data [34].
learning
Unsupervised neural network (UNN) categorizes data based on their similarities.
Unsupervised neural
Since the output is unknown, UNN considers the correlations between different
network
inputs and categorizes them into different groups [15].
Self-training first classifies using labeled data, and then unlabeled data are used as
Self-training
inputs [15].
Semisupervised
Being an extension of SVM, transductive support vector machine (TSVM) considers
learning Transductive support vector
both labeled and unlabeled data to make sure the margin is maximized between them
machine
[15].
Boosting uses two sorts of variables, namely, weak learners and strong learners. By
Boosting grouping weak learners and converting them to strong learners, it aims to decrease
Ensemble learning bias and variances [36].
Bagging is another tool which can be applied to decrease variances and increase the
Bagging
accuracy and stability of ML [37].
prediction, demand and sales estimation, production, in- reinforcement learning (RL) technique have been applied
ventory management, and transportation is provided. many times by researchers to solve the supplier selection
problem. In the following, how to apply some of the ML
techniques to solve the supplier selection problem is
3.1. Application of Machine Learning Algorithms in Supplier discussed.
Selection. Supplier selection can be considered as the main
activity in the purchasing function [39]. Due to the im-
portant role of suppliers on time, cost, and quality, supply 3.1.1. Supplier Selection Using DT and P-SVM Combined
chain managers have put much effort into the supplier Technique. DT is one of the ML techniques that is used in
selection process. (e selection process can be covered by classification problems where the dataset is not large and can
MCDM techniques which include different conflicting be modeled by a similar model. Moreover, this classifier can
factors. (erefore, finding the right balance between those be used when only a few features are available or a model is
factors is an important task for purchasing managers. needed that can be visualized and explained in simpler. DT
MCDM techniques support the decision-makers in provides an effective and rapid method of classifying
evaluating a set of alternatives [27]. (e MCDM tech- datasets. Recently, DT has been applied in various domains
niques support decision-makers in evaluating and [40–42]. In addition, DTs can be used as alternatives to find
selecting a set of alternatives. In some cases, the number of solutions for a multiclass problem that has binary predictors
potential suppliers and the number of criteria are sig- as it has a hierarchical structure [27]. (e hierarchical
nificantly larger than those that MCDM techniques can structure includes ramifications and nodes.
cope with them satisfactorily. On the other hand, MCDM Baryannis et al. [13] compared the outputs of SVM and
methods are categorized into descriptive and static DT techniques in the supply chain area. (e results of their
methods like most of the other traditional methods, while study showed that SVM prediction models analyze data for
in today’s competitive marketplace, predictive analytics classification and regression analysis. Considering the lower
methods are definitely more useful than descriptive result for different metrics, SVM models achieve good re-
methods. In such an era, ML algorithms have much better sults across various metrics. (e results also indicate how
performance than the mentioned methods. Among the accuracy can be a misleading metric in different scenarios. In
ML techniques, DT and SVM technique as a supervised the case of DT results, it can be seen that the outcomes were
learning technique and the Q-learning technique as a quite similar when it was solved via SVM.
In this subsection, the potential support vector machine

(P-SVM) is introduced as a novel approach to the original
SVM proposed by Hochreiter and Obermayer [43], which is A
A {2}, {1,3,4}
combined with DT to solve the supplier selection problem.
P-SVM can be used to construct binary classifiers and select
2 B
various features at the same time. (e hybrid technique B {3}, {1,4}
developed by Guo et al. [27] (based on P-SVM and DT)
evaluates and selects a supplier based on three steps: C
3 C {1, 4}
Step 1. Using training samples, it generates DT via the
adapted Kruskal algorithm (for more information
1 4
about the Kruskal algorithm, refer to Guo, Yuan, Tian
[27], page 6984) {j, k} Set of classes
Step 2. P-SVM is used to construct the binary classifiers Selected features at a node
and select the features of related nodes m Class m represented as leave
Step 3. Using the values of new samples as input to Figure 1: Structure of P-SVM and DT combined technique [27].
binary classifiers, P-SVM evaluates the suppliers and
selects the best one based on DT
3.1.2. Supplier Selection Using Reinforcement Learning.
In the procedure of DT and P-SVM combined tech-
RL, in contrast to other ML methods, does not need a model
niques as shown in Figure 1, at the nodes of A, B, and C, the
of the environment and is capable of online learning. In this
binary classifiers are defined using the selected features. For
subsection, Q-learning technique as one of the most widely
instance, class 2 samples are separated from classes 1, 3, and
pursued RL techniques is introduced to solve the supplier
4 samples which are classified at nodes B and C. (erefore,
selection problem. Valluri and Croson [28] have proposed a
new samples only should be defined at each of the nodes as
modified Q-learning technique to evaluate and select sup-
a subset of the initial features. Moreover, to form a hier-
pliers. Q-learning is a model-free RL algorithm that rewards
archical system, some subsets are achieved using P-SVM.
actions with positive outcomes and penalizes actions with
For instance, class 2 samples should be defined with a
negative outcomes. In this regard, Qt (ai )indicates the
subset signed with A when they are separated from classes
1, 3, and 4 samples. Furthermore, the samples from class 3 supplier’s value at a quality level ai , at time t. Supplier’s value
should be described with a subset of features marked with B at a quality level ai , at the time t + 1, is updated with the
when they are separated from samples from classes 1 and 4. following equation:
(us, at each of the related nodes, those subsets signed with
A, B, and C shape a hierarchical system that can be used to Qt+1 ai 􏼁 � Qt ai 􏼁 + α􏼂Pt+1 ai 􏼁 + Qt ai 􏼁􏼃. (1)
classify samples [27].
As mentioned, the supplier selection problem is an (e Q-learning equation is illustrated by equation (1)
MCDM problem in which suppliers must be evaluated with where the assumption of discount factor is 1 and between
respect to many criteria. (e number of criteria makes it periods is assumed to be 1. Here, Pt+1 (ai )is the suppliers’
hard for decision-makers to accomplish consistent com- profits at the quality level of ai . Various states of Markov
pression in order to determine the weights of criteria and decision process include different levels of quality which
then makes it too complex to evaluate the performance of suppliers can achieve. Considering the increment of 0.1,
each potential supplier with respect to each weighted suppliers can produce the level of quality level ranging from
criteria by the traditional method. (e P-SVM-DT com- 0 to 1. (e modified Q-learning algorithms have been used
bined technique leads to much better results than tradi- to obtain faster convergence. In the initial algorithm, if
tional MCDM and original SVM methods in solving the supplier agents meet the requirement to reach “Grade I
supplier selection problem because of the three following goods,” the level of quality ai will be selected to be rewarded;
reasons [27]: however, in the modified algorithm, all levels of quality
higher than the chosen level will be rewarded and the ones
(I) Introducing the hierarchical structure of the
lower than the chosen level of quality will be penalized (as in
P-SVM-DT combined technique reduces the
Figure 2(a), the 0.5 is the specific level of quality and all other
number of binary classifiers
levels higher than 0.5 will be rewarded). (is is illustrated in
(II) Combined with the Kruskal algorithm, the hier- Figure 2(b), where 0.5 is the chosen level of quality, and it
archical structure of P-SVM-DT is designed to can be seen that all other levels of quality lower than 0.5 are
increase the accuracy and decrease the computation penalized.
time in different classifications (e logic behind using penalty or reward actions is
(III) To promote the generalization performance, related to the supplier’s ability to meet the requirements
P-SVM selects the most ‘‘informative” features to defined at the chosen level of quality. (is means that if
develop classifiers suppliers reach the chosen level of quality, it is highly likely
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

X
(a)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
X
(b)
Figure 2: RL under the modified learning algorithm [28].
to reach a higher level in the next periods, which can be the application of ML in supplier segmentation has been
rewarded by more profits. However, if suppliers fail to satisfy considered. Bai et al. [14] proposed an MCDM-ML method
the requirement at the chosen level of quality, it is highly to evaluate and segment the suppliers. In the following, the
possible that the suppliers fail in the lower level of quality. steps of MCDM-ML proposed to evaluate and segment the
suppliers are presented:
3.2. Application of Machine Learning Algorithms in Supplier Step 1. Identify a set of criteria to evaluate suppliers
Segmentation. Supplier segmentation is one of the strategic Step 2. Evaluate and score the suppliers with respect to
activities of any organization through which suppliers are each of the weighted criteria using an outranking
categorized into groups based on their similarities. Con- MCDM method (e.g., Entropy, PROMETHEE, and
tracts, managing, and handling suppliers’ affairs are similar TOPSIS (Technique for order of preference by simi-
within each group and different from the suppliers in other larity to ideal solution (TOPSIS)))
groups. Supplier segmentation promotes the efficiency and Step 3. Determine the importance/weight of each cri-
effectiveness of relationships with suppliers and leads to the terion using a weighted MCDM method (e.g., Entropy,
development and improvement of the organization’s per- AHP, and BWM)
formance [29, 44]. A review of the literature on segmen-
Step 4. Segment the suppliers based on their score using
tation shows that supplier segmentation has received little
a supervised/unsupervised ML method (e.g., K-means,
attention in comparison to customer segmentation and is
fuzzy C-means, SVM, and DT)
still in its initial stages. While customer segmentation is one
of the targets of the market demand side, supplier seg-
mentation focuses on the objectives of the market supply
3.3. Application of Machine Learning Algorithms in Managing
side [30].
Supply Chain Risks. Supply chain risk management (SCRM)
Generally, market segmentation can be divided into
refers to the coordinated and collaborative efforts of all
three categories as follows:
parties involved in a supply chain to assess, identify,
(a) Segmentation of customers or end consumers monitor, and mitigate risks aiming at increasing robustness
(b) Segmentation of industrial customers or demand and resilience, reducing the vulnerability of the supply chain,
side of B2B2 market and guaranteeing continuity and profitability [54, 55].
SCRM encompasses a wide range of actions and decisions
(c) Segmentation of suppliers or supplied side of B2B
that have resulted in an equally wide spectrum of solutions
market
pursued by many scholars.
While the research literature is relatively rich for the first
(1) Multiplecriteria Decision Analysis (MCDA). (is
two cases, the third case—supplier segmentation—has rather
category encompasses well-established approaches
been neglected, and few studies have been conducted in
to assess various risk-related criteria that affect
order to provide a practical tool that can be applied and
supply chain performance, as well as the efficacy of
implemented in real-world conditions [45]. A review on
potential solutions, such as a hybrid AHP, PROM-
supplier segmentation indicates that in most of the research
ETHEE approach [56], and an integrated fuzzy
studies in this area, MCDM methods under fuzzy conditions
TOPSIS-CRITIC approach [57, 58].
have been used to evaluate and segment the suppliers. Some
of these studies are presented in Table 2. (2) Mathematical Modeling and Optimization. (is
Most of the MCDM methods that are mentioned in category is by far the most common one and en-
Table 2are expert-based and/or pairwise comparison-based compasses approaches based on fuzzy programming
methods. An important disadvantage of these methods is or stochastic and robust optimization [51, 54, 59, 60].
that the weaknesses in dealing with big data. (erefore, (3) AI techniques have received relatively little attention
researchers have drawn attention to the use of ML methods in the field of SCRM or typically supply chain re-
in evaluating and managing suppliers [51–53] which are search [61]. Recently, because of the availability of
even more efficient than traditional methods. In this section, large amounts of data and increased computing
Table 2: (e widely used methods in supplier segmentation.

Authors Method used for supplier segmentation
Rezaei and Ortt [46] Supplier segmentation using two general criteria of suppliers’ capability and willingness using fuzzy logic
Rezaei and Ortt [45] Supplier segmentation using two general criteria of suppliers’ capability and willingness using fuzzy AHP3
Supplier segmentation using two general criteria of suppliers’ capability and willingness, using BWM(Best-
Rezaei et al. [47]
worst method (BWM)) technique
Supplier segmentation using two strategic and vital dimensions using MAUT (Multi-attribute utility theory
Segura and Maroto [48] (MAUT)) and PROMETHEE (Preference ranking organization method for enrichment of evaluations
(PROMETHEE)) techniques
Supplier segmentation based on environmental criteria using ELECTRE (Elimination and choice translating
Rezaei et al. [44]
reality (ELECTRE)) TRI-rC techniques
Supplier segmentation based on the hybrid approach of PPM (Purchasing portfolio matrix (PPM)) and SPP
Rezaei and Fallah lajimi [49]
(Supplier potential matrix (SPM)) matrices using BWM technique
Selection and segmentation of suppliers based on their resilience using DEMATEL (Decision-making trial
Parkouhi et al. [50]
and evaluation laboratory (DEMATEL)) and gray SAW (Simple additive weighting (SAW)) techniques
power, as well as the success of ML techniques, AI demand estimation. By introducing nonlinear analyses, ML
has been a resurgence. It has also led to apply the techniques have been used to accurately predict and forecast
potential of AI techniques in SCRM by researchers in demand, sales, and inventory, leading to optimization of
processes such as prediction, risk identification, supply chain performance. An intelligent forecasting system
assessment, and response [62–69]. leads to optimize performance, reduce costs, and increase
sales and profit. In this case, it is required to relate the
A data-driven framework for managing supply chain
parameters associated with multiple explanatory variables to
risks includes two general phases:
their dependent variables in a highly nonlinear manner.
(i) Risk Identification. Risk identification is the first Being different from those traditional methods like expo-
activity in SCRM. Different researchers have used nential smoothing, moving average, time-series method, and
different algorithms and methods to identify risks. Box–Jenkins method, ML techniques usually do not heavily
Classification and clustering techniques such as rely on the accuracy of historical data so that ML techniques
SVM and DT are widely employed techniques to have been promoted as great alternatives for demand
identify risks. After identifying SCR, it is the time to forecasting and planning in SCM [72]. For instance, (o-
assess the risks and propose some appropriate ac- massey [73] proposed a forecasting model by advanced
tivities to mitigate the risks. methods such as data mining and fuzzy logic for sale. (eir
(ii) Risk Assessment and Response. ML and big data model performance is more reliable than traditional models.
analytics have also been utilized to deal with risk (is model worked well under situations such as strong
assessment. Artificial neural networks (ANNs) and seasonality of sales, volatile demand, and a wide number of
Bayesian networks are the two most used ML items with a short life cycle or the lack of historical data
techniques that have shown acceptable potential in background. Figure 4indicates the principles of a fuzzy
modeling for risk assessment [70, 71]. forecasting system. (is system makes it possible to generate
a model which automatically recreates the expert approach
In the following, a framework to integrate AI techniques when they carry out intuitive predictions with respect to
within the SCRM process has been proposed that can en- explanatory variables. (ree stages are needed to implement
hance supply chain risk prediction. (e aim of this two-way this fuzzy forecasting system as follows:
framework is the creation of synergy and interactivity be-
tween AI and supply chain experts: any decision that should (1) Removing the impact of explanatory variables from
be made by the AI experts always depends on the specific historical sales
inputs by the supply chain, while the developed models and (2) Using the baseline sales as a resulting data to forecast
produced results have to be interpretable so that decision- the baseline of the next year, for example, applying a
making is based on them or they can affect SCRM decision- basic forecasting technique according to the sea-
making. Figure 3represents the flow of the framework’s sonality average
process. On the left-hand side of the figure, the focal point is (3) Adding the impact of explanatory variables re-
on the major tasks involved in a data-driven AI method- garding the next year to this baseline forecast
ology. (e right-hand side contains the traditional tasks that
are involved in a standard SCRM process. As can be per-
ceived, this framework is built upon effective synergies 3.5. Application of Machine Learning Algorithms in Inventory
between a team of experts of data-driven AI and a team of Management. Storage incurs important costs in SCM. For
experts of risk management within a supply chain. instance, the annual costs of storage in supply chains are
about 15%–35% of their total business value [74]. Supply
chain inventory management seeks to decrease costs,
3.4. Application of Machine Learning Algorithms in Demand/ increase product variety, and improve customer service as
Sales Estimation. In SCM, planning is based on sales/ well. However, precisely estimating, predicting, and
Data-driven artificial intelligence Supply chain
Internal and external Supply chain risk

data collection management planning
Data preprocessing
Feature selection Risk prediction goals
Metric (s) selection
Prediction priorities
Algorithm (s) selection
Algorithm 1 Algorithm N
Prediction models
Supply chain risks

management decision-making
Prediction results
Figure 3: Data-driven risk prediction framework [54].
accessing the concerned information of all these goals by lead-time forecasting. (ey employed neuro-fuzzy de-
traditional decision rules are hard because this infor- mand in a multi-echelon supply chain. (e results showed
mation usually is high according to the experience and that their proposed model efficiently improved the in-
judgment of inventory managers themselves. Hence, the ventory management performance. Furthermore, ML has
inventory input is an uncertain parameter. In recent years, demonstrated an efficient performance for in-house de-
the inefficiency of traditional methods in facing uncer- livery or automating inspecting the damage inside lo-
tainty has led researchers to apply AI. ML tools can ex- gistics hubs. As a result, the ML tools are able to find the
plore quick input comparable patterns on warehouse hidden inventory patterns that have never been unfolded
datasets. Gumus et al. [26] applied neural networks to in decreasing and saving costs.
Baseline sales Baseline forecast

Forecast based Final
Historical sales Remove influence Add influence
on seasonality forecast
Corrective coefficient
Corrective coefficient
Fuzzy interference Fuzzy interference
Explanatory variable Explanatory variable
system system
Figure 4: Fuzzy forecasting system [73].
3.6. Application of Machine Learning Algorithms in Trans- 3.8. Application of Machine Learning Algorithms in Sus-
portation and Distribution. Vehicle routing problems tainable Development. In different industries, the emergence
(VRPs) are categorized as one of the most applicable issues of decentralization in the level of organizational structure
in SCM [75–80]. Solving VRPs is one of the high and well- restricts the SCM scope whereas concentrating on sustainable
known applications of ML in SCM. (e optimal route for performance [83–87]. In terms of quantity and time, ML has
transporting vehicles is an important issue in SCM to deliver real-time capability inputs related to activities and resources
products and services on time to customers; in most cases, [88]. Accordingly, ML gathers and employs the data, conducts
routing is beyond the ability of a human brain. ML tools analysis, and renders the required inputs to decide on coping
have excellent performance for analyzing large and diverse with the dealt affairs. ML permits industries to minimize
datasets and accurate forecasting of parameters. Ćirović et al. human intervention and maximize the effectiveness and ef-
[81] developed a model for the routing of light delivery ficiency of work processes simultaneously.
vehicles through logistics operators in which an adaptive Focusing on customer demand and production planning
neural network was trained by a simulated annealing al- in traditional SCM is not an effective way, and that is why the
gorithm. (is model assesses the performance of the dis- concept of sustainable SCM is brought up [89]. Furthermore,
tribution network routes. ML algorithms are able to generate the lack of objectivity in human experts limits the research
better delivery routes by objectively and timely exploring the works on the data analysis on demand forecast in the tra-
pattern of transportation, consumers’ behavior, infrastruc- ditional SCM. In fact, the SCM literature shows that analysis
tures, and vehicles. in the traditional supply chain is highly affected by the biases
of decision-makers and unreasonable market conjectures.
Due to this, the inclusion of ML is the best choice. For ex-
3.7. Application of Machine Learning Algorithms in ample, ML provides better production planning since it helps
Production. ML tools will improve the production planning managers identify consumers’ requirements and their pur-
and factory scheduling accuracy by taking into account chase patterns under different scenarios. On the other hand,
multiple constraints. ML tools will also make it possible to for a sustainable SCM, ML can play a vital role due to its high
balance the constraints more effectively than those that were potential to tackle the uncertainty and conserve industries for
manually done in the past, particularly for manufacturers efficiently encountering the approaching challenges. As a real
who rely on build-to-order and make-to-stock production case, Sharma et al. [89] performed a comprehensive review on
workflows. Manufacturers can apply ML tools to reduce the application of ML techniques in sustainable agriculture
supply chain latency for components and parts used in their SCM, where 93 papers were surveyed.
most heavily customized products. For example, faced with
the different production regulations and customization re-
quirements of each country, Chen et al. [82] applied NNs to 3.9. Application of Machine Learning Algorithms in Circular
group similar customization needs. (en, they used the Economy. A CE is a systemic procedure that contributes to
existing inventory information to select the parts for pro- maintain the activities related to reverse flows in supply chains
duction managers, which hugely reduced the cost during the including resource usage, recovery, recycling, reuse, repair,
supply chain compared with human decision. As a whole, remanufacturing, and refurbishment [90]. It motivates
the ML tools can yield the lead-time prediction in pro- the perpetual reuse of materials to reduce waste, as well as
duction with a shorter response time. directing demand toward natural resource consumption
Supplier selection and

segmentation
1. Extracting informative features

for supplier evaluation
2. Reducing computational
consumption because of reducing the
number of features
Production Supply chain risk management
1. Improving the accuracy of 1. Identifying, prediciting, assessing,

scheduling and production planning monitoring, and mitigating risks
2. Balancing the constraints in 2. Increasing robustness and
bulid-to-order and make-to-stock resilience of the supply chain
production workflows effectively Supervised learning 3. Ensuring profitability and
continuity
Semisupervised learning
Ensemble learning
ML
Transportation and distribution Demand and sales estimation

Unsupervised learning
1. Generating better deliver routes
1. Improving the accuracy of
2. Timely exploring the pattern of
predicting in sales and demand
consumers’ behavior
2. Reducing inventory costs by
more accurate calculation of
required inventory
Inventory management
1. Identifying the hidden inventory

patterns and reducing costs
2. Automating inspecting the
damage inside logistics hubs or in-
house delivery
Monitoring
Control
Defining criteria to evaluate
effectiveness, efficiency, and
adequacy of the outputs of ML
application in SCM
Figure 5: Conceptual framework for ML application in SCM.
[86]. AI and ML algorithms can serve as efficient plat- (i) Design Circular Materials, Components, and Prod-
forms for a quick transformation to the CE. (ere are ucts. It is obvious that ML tools can facilitate the
three key factors of CE where ML techniques can be development of new materials and products, such as
applied as follows: rapid ML-driven prototyping and testing
(ii) Operate Circular Business Models. ML can improve in managing the supply chain [13, 14, 52, 53] which are even
product circulation through intelligent demand more efficient than traditional techniques in dealing with a
prediction, pricing, inventory management, pre- large number of data. On the other hand, there are two
dictive maintenance, and so on criticisms about ML methods as follows:
(iii) Optimize Circular Infrastructure. ML can enhance (I) (e nature of most of the AI-based techniques is
components remanufacturing, products sorting and learning historical data, and then they make a
disassembling, and materials recycling which can decision about new samples. (ese techniques are
provide the reverse logistics infrastructure not capable of dealing with big changes and
As a recent study, Wilts et al. [91] investigated the ap- revolutions in problem situations. For example,
plication of AI in sorting of municipal solid waste (MSW) the COVID-19 outbreak has made large and
according to the CE principles. ML was discussed as one of unpredictable changes in the business environ-
the most efficient technologies for automatic waste sorting. ments that ML techniques could not have ac-
ceptable performance in dealing with. In such a
condition that the problem situation is under-
4. A Framework for ML Application in SCM going big and unpredictable changes, AI is faced
with a condition that has no resemblance to what
In the previous sections, the application of ML techniques in
it has learned and faced in the past, and the ef-
different areas of SCM was discussed. In this section, a
fectiveness and efficiency of AI methods could be
conceptual framework was presented to give a broad
dramatically reduced.
overview of the outputs of the application of ML in each area
of SCM, which cannot be achieved by applying traditional (II) (ere is a doubt about the validity and fairness of
methods. Moreover, some useful information about the ML techniques in some cases. At the heart of the
necessary and sufficient conditions to use ML in SCM ef- problem is the fact that ML techniques calculate
fectively was provided. As it is clear in Figure 5, one of the optimal models from the data they are given.
most important outputs of applying ML techniques such as Meaning they can end up replicating the problems
SVM in supplier evaluation is to extract some seminal they are meant to correct [92].
features which could have a significant effect on supplier (erefore, as mentioned at the bottom of Figure 5, the
selection and segmentation processes. Moreover, using ML outputs of ML techniques in SCM must be monitored,
techniques can reduce the time of evaluating suppliers based evaluated, and controlled based on their logical desirability
on criteria by reducing the number of features applied to and cultural feasibility. In other words, the outputs of AI-
classification. (e outputs of applying ML in SCRM are based methods must be evaluated by the human agent (SCM
predicting, assessing, and mitigating risks intelligently. expert) to ensure that the outputs and gained strategies from
Furthermore, applying ML in SCRM can make it possible them are logically desirable and culturally feasible or not, or
to cultivate a robust and resilient supply chain and ensure the outputs of ML application in SCM are effective, efficient,
the continuity and profitability of the organization. One of and adequate.
the outputs of applying ML in demand and sales estimation
is to predict demand and sales more accurately than tra-
ditional methods, and this leads to lower inventory costs by 5. Conclusion and Managerial Implication and
an accurate required inventory estimation. ML techniques Future Research Direction
can recognize the hidden inventory patterns that have a
significant impact on reducing costs. Using AI makes it (is paper aimed to provide an overview of the application
possible to establish an automated inspecting system to of ML techniques in different areas of the supply chain. In
mitigate the damage inside logistics hubs or in-house this regard, after a brief review of widely used supervised,
delivery. Transportation and distribution is an important unsupervised, and semisupervised ML techniques, the ap-
activity in SCM that loads heavy costs on organizations. plication of each one in managing different areas of the
(e results of applying ML in transportation and distri- supply chain was presented. Accordingly, supplier selection,
bution systems can lead to timely delivery of a product to supplier segmentation, supply chain risk prediction, demand
the corresponding customers by generating better delivery estimation, inventory management, transportation, sus-
routes and exploring the consumer behavior. (e other tainable development, and circular economy in SCM were
activity in SCM that has faced multiple constraints is discussed as the most significant areas. In the section on
production planning. An appropriate production planning supplier selection, two different machine learning methods
system should be dealt with these constraints effectively. were explained. First, how the use of DT and P-SVM
ML techniques can balance the constraints in build-to- combined method in evaluating and selecting suppliers was
order and make-to-stock production workflows and ef- explained. Second, the Q-learning technique as an RL
fectively improve the accuracy of scheduling and pro- technique was introduced in order to solve the supplier
duction planning. selection problem. In the next section, the procedure of
As mentioned earlier, traditional methods are weak in supplier segmentation using an MCDM-FCM combined
dealing with big data problems. (erefore, recently, re- method and based on environmental criteria was described.
searchers have drawn attention to the use of ML techniques (ird, the stages of designing a data-driven framework for
predicting supply chain risks were introduced. Fourth, the chains,” Expert Systems with Applications, vol. 115, pp. 618–
application of a fuzzy forecasting system as one of the widely 634, 2019.
used machine learning techniques in demand/sales esti- [2] S. C. Koh, S. Saad, and S. Arunachalam, “Competing in the
mation was explained. Finally, a summary of ML algorithms’ 21st century supply chain through supply chain management
applications in inventory management, transportation, and and enterprise resource planning integration,” andInterna-
production was introduced. Although the main goal of this tional Journal of Physical Distribution and Logistics Man-
agement, vol. 36, no. 6, pp. 455–465, 2006.
paper was addressed as expected and the advantages of
[3] S. Sadeghi, N. Rasouli, and G. Jandaghi, “Identifying and
applying ML in managing supply chains were concluded, prioritizing contributing factors in supply chain competi-
there are some limitations. For instance, considering the tiveness by using PLS-BWM techniques (case study: payam
target audience of this paper (beginners and industry shoes company),” World Scientific News, vol. 49, no. 2,
managers), providing the details of ML algorithms was not pp. 117–143, 2016.
possible and just general explanation and sometimes [4] D. M. Lambert, M. C. Cooper, and J. D. Pagh, “Supply chain
drawing some flowcharts to acquaint the beginners and management: implementation issues and research opportu-
managers with the process of implementing and utilizing the nities,” @e International Journal of Logistics Management,
power of ML algorithms in the SCM area were provided. In vol. 9, no. 2, pp. 1–20, 1998.
this paper, without any concentration on a specific industry, [5] S. S. Darvazeh, I. R. Vanani, and F. M. Musolu, “Big data
the application of ML generally in different SCs was con- analytics and its applications in supply chain management,” in
sidered; it is obvious that the nature of the industry and the New Trends in the Use of Artificial Intelligence for the Industry
type and the volume of data have a significant effect on 4.0, IntechOpen, London, UK, 2020.
selecting a suitable algorithm. (erefore, it is suggested that [6] S. Tiwari, H. M. Wee, and Y. Daryanto, “Big data analytics in
managers should be careful in using different algorithms and supply chain management between 2010 and 2016: insights to
industries,” Computers and Industrial Engineering, vol. 115,
consider the suitability of the selected algorithm with the
pp. 319–330, 2018.
nature of the data and its interpretability for the industry. [7] B. Biggio and F. Roli, “Wild patterns: ten years after the rise of
(ere are some industries that are in the early stages of using adversarial machine learning,” Pattern Recognition, vol. 84,
ML techniques to improve their various supply chain pp. 317–331, 2018.
processes. For example, there is a great space in the re- [8] F. M. Musolu, S. Sadeghi Darvazeh, and I. Raeesi Vanani,
newable energy supply chain to cover and fruitful research “Deep learning and its applications in medical imaging,” in
contents to see in this domain in the future. (ere is also a Internet of @ings for Healthcare Technologies, pp. 137–153,
great gap in leveraging the power of the mathematical op- Springer, Singapore, 2021.
timizing model and machine learning in order to design and [9] S. Piramuthu, “Machine learning for dynamic multi-product
optimize SC, which can be considered in future research supply chain formation,” Expert Systems with Applications,
studies. Furthermore, the presented framework can be tested vol. 29, no. 4, pp. 985–990, 2005.
using structural equation modeling in some industries in [10] I. M. Cavalcante, E. M. Frazzon, F. A. Forcellini, and
future research studies. Despite the presence of AI for the D. Ivanov, “A supervised machine learning approach to data-
last half-century and its recent emergence in the SCM area, driven simulation of resilient supplier selection in digital
there is still a lack of research regarding the specific topics of manufacturing,” International Journal of Information Man-
agement, vol. 49, pp. 86–97, 2019.
artificial intelligence for different areas of the supply chain.
[11] Y. Zhu, C. Xie, G.-J. Wang, and X.-G. Yan, “’Predicting
Based on projected AI research trends, we suggest further China’s SME credit risk in supply chain finance based on
exploration in the application of RL techniques in real-time machine learning methods,” Entropy, vol. 18, no. 5, p. 195,
pricing. Most AI techniques applications in the SCM area 2016.
remain limited to relatively well-structured (well-defined), [12] P. Priore, B. Ponte, R. Rosillo, and D. de la Fuente, “Applying
operational, and tactical SCM problems. Future research can machine learning to the dynamic selection of replenishment
address the application of AI techniques (especially agent- policies in fast-changing supply chain environments,” In-
based systems) to solve a variety of soft but strategic SCM ternational Journal of Production Research, vol. 57, no. 11,
issues. (e main reason is that providing efficient solutions pp. 3663–3677, 2019.
for SCM problem is either too expensive or difficult due to [13] G. Baryannis, S. Dani, and G. Antoniou, “Predicting supply
the inherent complexity and ill-structured nature of the chain risks using machine learning: the trade-off between
problem, such as outsourcing relationships, supplier rela- performance and interpretability,” Future Generation Com-
tionship management, supply chain coordination, and puter Systems, vol. 101, pp. 993–1004, 2019.
strategic alliances among supply chain partners. [14] C. Bai, J. Rezaei, and J. Sarkis, “Multicriteria green supplier
segmentation,” IEEE Transactions on Engineering Manage-
ment, vol. 64, no. 4, pp. 515–528, 2017.
Conflicts of Interest [15] A. Dey, “Machine learning algorithms: a review,” Interna-
tional Journal of Computer Science and Information Tech-
(e authors declare that they have no conflicts of interest. nologies, vol. 7, no. 3, pp. 1174–1179, 2016.
[16] M. Bowles, Machine Learning in Python: Essential Techniques
References for Predictive Analysis, John Wiley and Sons, Hoboken, NJ,
USA, 2015.
[1] Z. Dai, F. Aqlan, K. Gao, and Y. Zhou, “A two-phase method [17] R. Carbonneau, K. Laframboise, and R. Vahidov, “Application
for multi-echelon location-routing problems in supply of machine learning techniques for supply chain demand
forecasting,” European Journal of Operational Research, [34] P. Harrington, Machine Learning in Action, Manning Pub-
vol. 184, no. 3, pp. 1140–1154, 2008. lications Co, Shelter Island, NY, USA, 2012.
[18] A. Ning, H. Lau, Y. Zhao, and T. T. Wong, “Fulfillment of [35] S. Shalev-Shwartz, Y. Singer, N. Srebro, and A. Cotter,
retailer demand by using the MDL-optimal neural network “Pegasos: primal estimated sub-gradient solver for SVM,”
prediction and decision policy,” IEEE Transactions on In- Mathematical Programming, vol. 127, no. 1, pp. 3–30, 2011.
dustrial Informatics, vol. 5, no. 4, pp. 495–506, 2009. [36] Z.-H. Zhou, “When semi-supervised learning meets ensemble
[19] Y. Pan, R. Pavur, and T. Pohlen, “Revisiting the effects of learning,” in Proceedings of the International Workshop on
forecasting method selection and information sharing under Multiple Classifier Systems, pp. 529–538, Springer, Reykjavik,
volatile demand in SCM applications,” IEEE Transactions on Iceland, June 2009.
Engineering Management, vol. 63, no. 4, pp. 377–389, 2016. [37] A. Lemmens and C. Croux, “Bagging and boosting classifi-
[20] M. Maghrebi, C. Sammut, and S. T. Waller, “Feasibility study cation trees to predict churn,” Journal of Marketing Research,
of automatically performing the concrete delivery dispatching vol. 43, no. 2, pp. 276–286, 2006.
through machine learning techniques,” Engineering, Con- [38] M. Bertolini, D. Mezzogori, M. Neroni, and F. Zammori,
struction and Architectural Management, vol. 22, no. 5, “Machine learning for industrial applications: a compre-
pp. 573–590, 2015. hensive literature review,” Expert Systems with Applications,
[21] S. Mercier and I. Uysal, “Neural network models for pre- vol. 175, Article ID 114820, 2021.
dicting perishable food temperatures along the supply chain,” [39] Q. Pang, T. Yang, M. Li, and Y. Shen, “A fuzzy-grey multi-
Biosystems Engineering, vol. 171, pp. 91–100, 2018. criteria decision making approach for green supplier selection
[22] S. Shervais, T. T. Shannon, and G. G. Lendaris, “Intelligent in low-carbon supply chain,” Mathematical Problems in
supply chain management using adaptive critic learning,” Engineering, vol. 2017, Article ID 9653261, 9 pages, 2017.
IEEE Transactions on Systems, Man, and Cybernetics - Part A: [40] G. A. Levow, “Characterizing and recognizing spoken cor-
Systems and Humans, vol. 33, no. 2, pp. 235–244, 2003. rections in human-computer dialogue,” in Proceedings of the
[23] W. W. C. Chung, K. C. M. Wong, and P. T. K. Soon, “An 36th Annual Meeting of the Association for Computational
ANN-based DSS system for quality assurance in production Linguistics and 17th International Conference on Computa-
network,” Journal of Manufacturing Technology Management, tional Linguistics, vol. 1, pp. 736–742, Quebec, Canada,
vol. 18, no. 7, pp. 836–857, 2007. August 1998.
[24] C. Liu, T. Shu, S. Chen, S. Wang, K. K. Lai, and L. Gan, “An [41] M. K. Markey, G. D. Tourassi, and C. E. Floyd, “Decision tree
classification of proteins identified by mass spectrometry of
improved grey neural network model for predicting trans-
blood serum samples from people with and without lung
portation disruptions,” Expert Systems with Applications,
cancer,” Proteomics, vol. 3, no. 9, pp. 1678-1679, 2003.
vol. 45, pp. 331–340, 2016.
[42] B. Zmazek, L. Todorovski, S. Džeroski, J. Vaupotič, and
[25] H. Wu, G. Evans, and K.-H. Bae, “Production control in a
I. Kobal, “Application of decision trees to the analysis of soil
complex production system using approximate dynamic
radon data for earthquake prediction,” Applied Radiation and
programming,” International Journal of Production Research,
Isotopes, vol. 58, no. 6, pp. 697–706, 2003.
vol. 54, no. 8, pp. 2419–2432, 2016.
[43] S. Hochreiter and K. Obermayer, “Support vector machines
[26] A. T. Gumus, A. F. Guneri, and F. Ulengin, “A new meth-
for dyadic data,” Neural Computation, vol. 18, no. 6,
odology for multi-echelon inventory management in sto-
pp. 1472–1510, 2006.
chastic and neuro-fuzzy environments,” International Journal [44] J. Rezaei, M. Kadziński, C. Vana, and L. Tavasszy, “Embed-
of Production Economics, vol. 128, no. 1, pp. 248–260, 2010. ding carbon impact assessment in multi-criteria supplier
[27] X. Guo, Z. Yuan, and B. Tian, “Supplier selection based on segmentation using ELECTRE TRI-rC,” Annals of Operations
hierarchical potential support vector machine,” Expert Sys- Research, pp. 1–23, 2017.
tems with Applications, vol. 36, no. 3, pp. 6978–6985, 2009. [45] J. Rezaei and R. Ortt, “Multi-criteria supplier segmentation
[28] A. Valluri and D. Croson, “Agent learning in supplier se- using a fuzzy preference relations based AHP,” European
lection models,” Decision Support Systems, vol. 39, no. 2, Journal of Operational Research, vol. 225, no. 1, pp. 75–84,
pp. 219–240, 2005. 2013.
[29] W. Jiang and J. Liu, “Inventory financing with overconfident [46] J. Rezaei and R. Ortt, “A multi-variable approach to supplier
supplier based on supply chain contract,” Mathematical segmentation,” International Journal of Production Research,
Problems in Engineering, vol. 2018, Article ID 5054387, vol. 50, no. 16, pp. 4593–4611, 2012.
12 pages, 2018. [47] J. Rezaei, J. Wang, and L. Tavasszy, “Linking supplier de-
[30] S. Erevelles and T. H. Stevenson, “Enhancing the business-to- velopment to supplier segmentation using Best Worst
business supply chain: insights from partitioning the supply- Method,” Expert Systems with Applications, vol. 42, no. 23,
side,” Industrial Marketing Management, vol. 35, no. 4, pp. 9152–9164, 2015.
pp. 481–492, 2006. [48] M. Segura and C. Maroto, “A multiple criteria supplier
[31] S. B. Kotsiantis, I. D. Zaharakis, and P. E. Pintelas, “Machine segmentation using outranking and value function methods,”
learning: a review of classification and combining tech- Expert Systems with Applications, vol. 69, pp. 87–100, 2017.
niques,” Artificial Intelligence Review, vol. 26, no. 3, [49] J. Rezaei and H. Fallah Lajimi, “Segmenting supplies and
pp. 159–190, 2006. suppliers: bringing together the purchasing portfolio matrix
[32] D. Lowd and P. Domingos, “Naive Bayes models for prob- and the supplier potential matrix,” International Journal of
ability estimation,” in Proceedings of the 22nd International Logistics Research and Applications, vol. 22, pp. 1–18, 2018.
Conference on Machine Learning, pp. 529–536, Germany, [50] S. V. Parkouhi, A. S. Ghadikolaei, and H. F. Lajimi, “Resilient
August 2005. supplier selection and segmentation in grey environment,”
[33] A. Karatzoglou, D. Meyer, and K. Hornik, “Support vector Journal of Cleaner Production, vol. 207, pp. 1123–1137, 2019.
machines in R,” Journal of Statistical Software, vol. 15, no. 9, [51] G. Baryannis, S. Dani, S. Validi, and G. Antoniou, “Decision
pp. 1–28, 2006. support systems and artificial intelligence in supply chain risk
management,” in Revisiting Supply Chain Risk, pp. 53–71, [67] W. Yan, J. He, and A. J. C. Trappey, “Risk-aware supply chain
Springer, Cham, Switzerland, 2019. intelligence: AI-enabled supply chain and logistics manage-
[52] A. Bruzzone and A. Orsoni, “AI and simulation-based ment considering risk mitigation,” Advanced Engineering
techniques for the assessment of supply chain logistic per- Informatics, vol. 42, Article ID 100976, 2019.
formance,” in Proceedings of the 36th Annual Simulation [68] L. Malviya, P. Chittora, P. Chakrabarti, R. S. Vyas, and
Symposium, pp. 154–164, IEEE, Florida, FL, USA, March S. Poddar, “Backorder prediction in the supply chain using
2003. machine learning,” Materials Today: Proceedings, 2021.
[53] K. Zhao and X. Yu, “A case based reasoning approach on [69] S. S. Kamble, A. Gunasekaran, V. Kumar, A. Belhadi, and
supplier selection in petroleum enterprises,” Expert Systems C. Foropon, “A machine learning based approach for pre-
with Applications, vol. 38, no. 6, pp. 6839–6847, 2011. dicting blockchain adoption in supply chain,” Technological
[54] G. Baryannis, S. Validi, S. Dani, and G. Antoniou, “Supply Forecasting and Social Change, vol. 163, Article ID 120465,
chain risk management and artificial intelligence: state of the 2021.
art and future research directions,” International Journal of [70] S.-S. Leu and C.-M. Chang, “Bayesian-network-based safety
Production Research, vol. 57, no. 7, pp. 2179–2202, 2019. risk assessment for steel construction projects,” Accident
[55] J. Luo, X. Zhang, and X. Jiang, “Multisources risk manage- Analysis and Prevention, vol. 54, pp. 122–133, 2013.
ment in a supply chain under option contracts,” Mathematical [71] L. Zhang, X. Wu, M. J. Skibniewski, J. Zhong, and Y. Lu,
Problems in Engineering, vol. 2019, Article ID 7482584, “Bayesian-network-based safety risk analysis in construction
12 pages, 2019. projects,” Reliability Engineering and System Safety, vol. 131,
[56] S. Prasanna Venkatesan and S. Kumanan, “Supply chain risk pp. 29–39, 2014.
prioritisation using a hybrid AHP and PROMETHEE ap- [72] D. Ni, Z. Xiao, and M. K. Lim, “A systematic review of the
proach,” International Journal of Services and Operations research trends of machine learning in supply chain man-
Management, vol. 13, no. 1, pp. 19–41, 2012. agement,” International Journal of Machine Learning and
[57] R. Rostamzadeh, M. K. Ghorabaee, K. Govindan, A. Esmaeili Cybernetics, pp. 1–20, 2019.
et al., “Evaluation of sustainable supply chain risk manage- [73] S. (omassey, “Sales forecasts in clothing industry: the key
ment using an integrated fuzzy TOPSIS- CRITIC approach,” success factor of the supply chain management,” International
Journal of Cleaner Production, vol. 175, pp. 651–669, 2018. Journal of Production Economics, vol. 128, no. 2, pp. 470–483,
[58] M. Abdel-Basset and R. Mohamed, “A novel plithogenic 2010.
TOPSIS- CRITIC model for sustainable supply chain risk [74] S. G. Timme and C. Williams-Timme, “(e real cost of
holding inventory,” Supply Chain Management Review, vol. 7,
management,” Journal of Cleaner Production, vol. 247, Article
no. 4, pp. 30–37, 2003.
ID 119586, 2020.
[75] E. B. Tirkolaee, A. Goli, A. Faridnia, M. Soltani, and
[59] Y. H. Lee, P. Golinska-Dawson, and J. Z. Wu, “Mathematical
G.-W. Weber, “Multi-objective optimization for the reliable
models for supply chain management,” Mathematical Prob-
pollution-routing problem with cross-dock selection using
lems in Engineering, vol. 2016, Article ID 6167290, 4 pages,
Pareto-based algorithms,” Journal of Cleaner Production,
2016.
vol. 276, Article ID 122927, 2020.
[60] S. S. Darvazeh, M. Amiri, and F. M. Mooseloo, “Artificial
[76] M. Alinaghian, E. B. Tirkolaee, Z. K. Dezaki, S. R. Hejazi, and
intelligence and its application in data-driven optimization,”
W. Ding, “An augmented Tabu search algorithm for the green
in @e Open Access Book, Data Mining, IntechOpen, London,
inventory-routing problem with time windows,” Swarm and
UK, 2021. Evolutionary Computation, vol. 60, Article ID 100802, 2021.
[61] R. Toorajipour, V. Sohrabpour, A. Nazarpour, P. Oghazi, and [77] E. B. Tirkolaee and N. S. Aydın, “A sustainable medical waste
M. Fischl, “Artificial intelligence in supply chain manage- collection and transportation model for pandemics,” Waste
ment: a systematic literature review,” Journal of Business Management and Research, Article ID 0734242X211000437,
Research, vol. 122, pp. 502–517, 2021. 2021.
[62] H. Chen, Y. Zhu, K. Hu, and X. Li, “Virtual enterprise risk [78] E. B. Tirkolaee, P. Abbasian, and G.-W. Weber, “Sustainable
management using artificial intelligence,” Mathematical fuzzy multi-trip location-routing problem for medical waste
Problems in Engineering, vol. 2010, Article ID 572404, management during the COVID-19 outbreak,” Science of the
20 pages, 2010. Total Environment, vol. 756, Article ID 143607, 2021.
[63] Y. Fan, L. Heilig, and S. Voß, “Supply chain risk management [79] M. Yousefi Nejad Attari, A. Ebadi Torkayesh, B. Malmir, and
in the era of big data,” in Design, User Experience, and Us- E. Neyshabouri Jami, “Robust possibilistic programming for
ability: Design Discourse, pp. 283–294, Springer, Cham, joint order batching and picker routing problem in warehouse
Switzerland, 2015. management,” International Journal of Production Research,
[64] M. He, H. Ji, Q. Wang, C. Ren, and R. Lougee, “Big data fueled pp. 1–19, 2021.
process management of supply risks: sensing, prediction, [80] N. Furian, M. O’Sullivan, C. Walker, and E. Çela, “A machine
evaluation and mitigation,” in Proceedings of the Winter learning-based branch and price algorithm for a sampled
Simulation Conference 2014, pp. 1005–1013, IEEE, Savannah, vehicle routing problem,” OR Spectrum, pp. 1–40, 2021.
GA, USA, December 2014. [81] G. Ćirović, D. Pamučar, and D. Božanić, “Green logistic
[65] D. Zage, K. Glass, and R. Colbaugh, “Improving supply chain vehicle routing problem: routing light delivery vehicles in
security using big data,” in Proceedings of the 2013 IEEE urban areas using a neuro-fuzzy model,” Expert Systems with
International Conference on Intelligence and Security Infor- Applications, vol. 41, no. 9, pp. 4245–4258, 2014.
matics, pp. 254–259, IEEE, Seattle, WA, USA, June 2013. [82] M.-K. Chen, T.-W. Tai, and T.-Y. Hung, “Component se-
[66] S. Ye, Z. Xiao, and G. Zhu, “Identification of supply chain lection system for green supply chain,” Expert Systems with
disruptions with economic performance of firms using multi- Applications, vol. 39, no. 5, pp. 5687–5701, 2012.
category support vector machines,” International Journal of [83] H. R. Vandchali, S. Cahoon, and S.-L. Chen, “Creating a
Production Research, vol. 53, no. 10, pp. 3086–3103, 2015. sustainable supply chain network by adopting relationship
management strategies,” Journal of Business-To-Business

Marketing, vol. 27, no. 2, pp. 125–149, 2020.
[84] H. R. Vandchali, S. Cahoon, and S.-L. Chen, “(e impact of
supply chain network structure on relationship management
strategies: an empirical investigation of sustainability prac-
tices in retailers,” Sustainable Production and Consumption,
vol. 28, pp. 281–299, 2021.
[85] H. R. Vandchali, S. Cahoon, and S. L. Chen, “(e impact of
power on the depth of sustainability collaboration in the
supply chain network for Australian food retailers,” Inter-
national Journal of Procurement Management, vol. 14, no. 2,
pp. 165–184, 2021.
[86] D. Wang and Y. Zhang, “Implications for sustainability in
supply chain management and the circular economy using
machine learning model,” Information Systems and E-Busi-
ness Management, pp. 1–13, 2020.
[87] P. Wang, “On defining artificial intelligence,” Journal of
Artificial General Intelligence, vol. 10, no. 2, pp. 1–37, 2019.
[88] D. Nagar, S. Raghav, A. Bhardwaj, R. Kumar, P. L. Singh, and
R. Sindhwani, “Machine learning: best way to sustain the
supply chain in the era of industry 4.0,” Materials Today:
Proceedings, 2021.
[89] R. Sharma, S. S. Kamble, A. Gunasekaran, V. Kumar, and
A. Kumar, “A systematic literature review on machine
learning applications for sustainable agriculture supply chain
performance,” Computers and Operations Research, vol. 119,
Article ID 104926, 2020.
[90] B. T. Hazen, I. Russo, I. Confente, and D. Pellathy, “Supply
chain management for circular economy: conceptual
framework and research agenda,” @e International Journal of
Logistics Management, vol. 32, no. 2, pp. 510–537, 2021.
[91] H. Wilts, B. R. Garcia, R. G. Garlito, L. S. Gómez, and
E. G. Prieto, “Artificial intelligence in the sorting of municipal
waste as an enabler of the circular economy,” Resources,
vol. 10, no. 4, p. 28, 2021.
[92] D. D. Cremer, What Does Building a Fair AI Really Entail?,
Harvard Business Review, Maasachussets, MA, USA, 2020,
https://hbr.org/2020/09/what-does-building-a-fair-ai-really-
entail.

Supply Chain

Uploaded by

Copyright:

Available Formats

Supply Chain

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Supply Chain

Uploaded by

Copyright:

Available Formats

Hindawi

Mathematical Problems in Engineering

Erfan Babaee Tirkolaee ,1 Saeid Sadeghi,2 Farzaneh Mansoori Mooseloo,3

Correspondence should be addressed to Erfan Babaee Tirkolaee; [email protected]

Academic Editor: Ali Ahmadian

1. Introduction Intelligent use of these data and the extraction of infor-

Table 1: Frequently used ML algorithms.

In this subsection, the potential support vector machine

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Figure 2: RL under the modiﬁed learning algorithm [28].

Table 2: (e widely used methods in supplier segmentation.

Data-driven artificial intelligence Supply chain

Internal and external Supply chain risk

Feature selection Risk prediction goals

Metric (s) selection

Algorithm (s) selection

Supply chain risks

Figure 3: Data-driven risk prediction framework [54].

Baseline sales Baseline forecast

Figure 4: Fuzzy forecasting system [73].

Supplier selection and

1. Extracting informative features

1. Improving the accuracy of 1. Identifying, prediciting, assessing,

Transportation and distribution Demand and sales estimation

1. Identifying the hidden inventory

Figure 5: Conceptual framework for ML application in SCM.

management strategies,” Journal of Business-To-Business

You might also like