econstor

A Service of

zbw
Leibniz-Informationszentrum
Wirtschaft

Make Your Publications Visible.

Leibniz Information Centre
for Economics

Wenzel, Hannah; Smit, Daniel; Sardesai, Saskia

Conference Paper
A literature review on machine learning in supply
chain management

Provided in Cooperation with:

Hamburg University of Technology (TUHH), Institute of Business Logistics and General
Management

Suggested Citation: Wenzel, Hannah; Smit, Daniel; Sardesai, Saskia (2019) : A literature
review on machine learning in supply chain management, In: Kersten, Wolfgang Blecker,
Thorsten Ringle, Christian M. (Ed.): Artificial Intelligence and Digital Transformation in Supply
Chain Management: Innovative Approaches for Supply Chains. Proceedings of the Hamburg
International Conference of Logistics (HICL), Vol. 27, ISBN 978-3-7502-4947-9, epubli GmbH,
Berlin, pp. 413-441,
[Link]

This Version is available at:

[Link]

Standard-Nutzungsbedingungen: Terms of use:

Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Documents in EconStor may be saved and copied for your
Zwecken und zum Privatgebrauch gespeichert und kopiert werden. personal and scholarly purposes.

Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle You are not to copy documents for public or commercial
Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich purposes, to exhibit the documents publicly, to make them
machen, vertreiben oder anderweitig nutzen. publicly available on the internet, or to distribute or otherwise
use the documents in public.
Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen
(insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, If the documents have been made available under an Open
gelten abweichend von diesen Nutzungsbedingungen die in der dort Content Licence (especially Creative Commons Licences), you
genannten Lizenz gewährten Nutzungsrechte. may exercise further usage rights as specified in the indicated
licence.

[Link]

[Link]
 Proceedings of the Hamburg International Conference of Logistics (HICL) – 27

Hannah Wenzel, Daniel Smit and Saskia Sardesai

A Literature Review on Machine Learning in

Supply Chain Management

Published in: Artificial Intelligence and Digital Transformation in Supply Chain Management
Wolfgang Kersten, Thorsten Blecker and Christian M. Ringle (Eds.)
September 2019, epubli
CC-BY-SA 4.0
A Literature Review on Machine Learning in
Supply Chain Management
Hannah Wenzel1, Daniel Smit1 and Saskia Sardesai1
1 – Fraunhofer Institute for Material Flow and Logistics IML

Purpose: In recent years, a number of practical logistic applications of machine

learning (ML) have emerged, especially in Supply Chain Management (SCM). By link-
ing applied ML methods to the SCM task model, the paper indicates the current ap-
plications in SCM and visualises potential research gaps.

Methodology: Relevant papers with applications of ML in SCM are extracted based

on a literature review of a period of 10 years (2009-2019). The used ML methods are
linked to the SCM model, creating a reciprocal mapping.

Findings: This paper results in an overview of ML applications and methods currently

used in the area of SCM. Successfully applied ML methods in SCM in industry and ex-
amples from theoretical approaches are displayed for each task within the SCM task
model.

Originality: Linking the SC task model with current application areas of ML yields an
overview of ML in SCM. This facilitates the identification of potential areas of appli-
cation to companies, as well as potential future research areas to science.

Keywords: Supply Chain Management, Machine Learning, Literature Review,

Predicitve Analytics
First received: [Link].2019 Revised: [Link].2019 Accepted: [Link].2019
414 Hannah Wenzel et al.

1 Motivation

In the age of information technology and increasingly complex technical

and industrial processes, agile and efficient logistics processes play a cen-
tral role. High logistic requirements such as reliability, transparency and
flexibility combined with optimal economic conditions form the foundation
for a successful supply chain (SC). Dynamically changing processes require
a technology that is able to cope with the increasing complexity of supply
chains.
The growth in data volume and diversity has led to data sets larger than
ever before. Processing with conventional, practical management tools is
often inefficient or impossible. To manage and evaluate these new and po-
tentially valuable data sets, new methods and applications have been de-
veloped in the form of predictive analytics (Waller and Fawcett, 2013, p. 77).
One of these predictive analytics methods is machine learning (ML). The
success of this methodology is achieved through the invention of sophisti-
cated ML models, the availability of large data sets ("big data") and the uti-
lization of hardware architectures such as GPUs (Abadi et al., 2016, p. 256;
Copeland, 2016; Steinbach, 2018, p. 32).
Current applications of ML focus on specific areas within SCM, while some
areas remain unexploited. Thus the paper intends to create a link between
current ML applications, actual research work and the SCM task model. This
allows for a visualization of tasks wherein ML techniques are already ap-
plied and a deduction of potential areas of future research within the SCM
task model.
The paper is based on a literature analysis by querying the "IEEE Xplore Dig-
ital Library", "Scopus" and "ScienceDirect" databases using the search
Literature Review on Machine Learning in Supply Chain Management 415

term "Supply Chain Management [AND] Machine Learning". Due to the re-
cent developments in ML, the results were restricted to publications from
2009-2019. Based on the abstracts, a total of 38 relevant papers have been
identified.
Along with the full papers, the specific SCM task and the applied ML meth-
ods have been extracted. The papers have been categorized by the source
of the datasets they used. 20 of the papers used real-world datasets ac-
quired mostly from enterprises. 9 papers used synthetic datasets (e.g. gen-
erated by a simulation). For the remaining 9 papers, the types of datasets
used were not mentioned. Furthermore, with the support of the Google
search function several practical applications of ML within SCM were iden-
tified. This was done to provide a complete picture of current practical ap-
plications.
The paper is structured as follows. Firstly, the different tasks of SCM are ex-
plained, ensued by a general introduction to ML including the different
types and methods, and explaining the process of applying ML techniques
for solving real-world issues. This is succeeded by application examples of
ML in SCM and a mapping to the SCM task model. The paper concludes with
an overview on possible areas of future research within the SCM task
model. The result of this paper addresses researchers in their future course
of research work as well as Supply Chain Managers in applying ML in their
field of work.
416 Hannah Wenzel et al.

2 Supply Chain Management and its Task Model

According to Arnold et al. (2008) the primary goal of SCM is to fulfill cus-
tomer needs while optimizing costs in terms of inventories, resources and
processes in the network at the same time. To ensure that this goal can be
achieved, it requires a number of subgoals e.g. improving customer orien-
tation and satisfaction, increase of delivery capability and reduction of lead
times (Arnold et al., 2008, 460 f.). The achievement of the preceding goals
must be carried out by internal and cross-company tasks of SCM. A repre-
sentation of these tasks is shown in Figure 1.
The three main tasks of the model are Supply Chain Design, -Planning and
-Execution (Kuhn and Hellingrath, 2002, 142 f.). The inner circle of tasks is
discussed briefly hereafter as it is of relevance for mapping ML applications.
Supply Chain Design
Supply Chain Design (SCD) deals with long-term planning in terms of loca-
tion decisions, make-or-buy decisions, supply relationships, capacity di-
mensioning, logistics strategy and general tasks (Hompel and Wolf, 2013, p.
148). Another important task is cost-optimized structuring of the logistics
and production processes. Consequently, this area involves a financial
evaluation of changes in the logistics network (Kuhn and Hellingrath, 2002,
143 f.).
Literature Review on Machine Learning in Supply Chain Management 417

Figure 1: Supply Chain Management task model (Hompel and Wolf, 2013,
pp. 146–147)

Supply Chain Planning

The goal of Supply Chain Planning (SCP) is medium- to long-term program
planning across the entire SC. The area of SCP reflects the planning of pro-
duction and logistics resources to fulfill customer orders. Hence, the basis
of resource planning are forecasted customer requirements. The tasks for
SCP include the areas of demand and network planning, which set the basis
for procurement, production and distribution planning and enabling avail-
able-to-promise/ capable-to-promise checks. In the short-term, procure-
ment planning, scheduling and sequencing and short-term distribution
planning support SCP (Kuhn and Hellingrath, 2002, p. 144). Thus, demand
information creates an important basis in order to coordinate decisions
and planning along the logistics chain.
418 Hannah Wenzel et al.

Tasks of demand planning are forecasting and visualization of short-term

and long-term requirements. These are particularly important when it
comes to fulfilling customer requirements, planning the utilization of the
value chain, or carrying out inventory optimization (Kuhn and Hellingrath,
2002, 144 f.).
The next task, network planning, deals with the coordination between part-
ners within the value chain. Through network planning, requirements as
well as material and capacity resources are coordinated. It is possible to
trace the fulfillment of the production program through network planning,
due to its orientation towards production programs and sales plans (Kuhn
and Hellingrath, 2002, 145 f.).
Procurement planning takes into account the preceding areas of demand
and network planning and determines the planning of the parts supply.
This is followed by production planning, which has the task of creating a
production plan for each production unit in the SC. The aim is to achieve a
high level of readiness for delivery while at the same time keeping inventory
costs as low as possible (Kuhn and Hellingrath, 2002, 146 f.).
Distribution planning deals with the fulfilment of requirements, which are
achieved through optimized planning and coordination of deliveries. It is
based on the results of production planning and the specifications of net-
work planning (Kuhn and Hellingrath, 2002, p. 147).
In order to satisfy customer requirements or customer orders, their feasi-
bility is examined via available-to-promise/ capable-to-promise. Due to the
gap in the task model between the anonymous planning tasks and the plan-
ning tasks with customer reference, this area represents an important task
of SCM (Kuhn and Hellingrath, 2002, 147 f.).
Literature Review on Machine Learning in Supply Chain Management 419

Short-term procurement planning, scheduling and sequencing, as well as

short-term distribution planning deal with more detailed tasks of procure-
ment, production and distribution planning. For more detailed infor-
mation, please refer to Kuhn and Hellingrath (2002).
Supply Chain Execution
The area of Supply Chain Execution (SCE) considers cross-company pro-
cesses such as SC management and control. The aim of this area is to sup-
port decision-making at the operational level (Kuhn and Hellingrath, 2002,
152 f.). Its executive area includes the processing and monitoring of orders.
Inventory management is also part of SC Execution, taking care of move-
ments of stocks and materials including their documentation. Two further
components are production management, which manages all information
on production processes, and transport management, which handles
transport orders on the procurement and distribution side.
An additional major task is Supply Chain Event Management. It controls the
activities within the SC, announces plan changes and initiates corrective
measures. The task aims to create transparency about parameters such as
disruptions, inventories and requirements. Therefore, it includes all reac-
tive risk managing activities (Kuhn and Hellingrath, 2002, p. 154).

3 Machine Learning

The following section interprets the term "Machine Learning" and covers
types of ML methods. This supports a common understanding of the meth-
ods used in the different application areas of ML in SCM. Challenges are also
addressed.
420 Hannah Wenzel et al.

3.1 Interpretation of Machine Learning

ML is a sub-area of artificial intelligence and represents another way of pro-
gramming. Example data replaces rigid calculation rules of a program.
From the given example data, learning methods or algorithms extract sta-
tistical regularities, and represent those in the form of models. The models
can react to new, unknown data and classify them into categories or make
predictions (Hecker et al., 2017, p. 8).
ML deals with the computer-aided modelling and realization of learning
phenomena (Görz, Schneeberger and Schmid, 2013, pp. 200–201). It is de-
fined as a process that uses experience to improve performance or make
concrete predictions. The experience refers to past information, which is
provided to the procedure from an electronic data collection. ML involves
the design of effective and precise algorithms (Mohri, Rostamizadeh and
Talwalkar, 2012).

3.2 Types of machine learning

Several types of ML can be distinguished. Well-known ones are supervised,
unsupervised and reinforcement learning. Further types can be found in
e.g. Géron (2017).
Supervised Learning
The most widely used type of ML is Supervised Learning (Marsland, 2015, p.
6). Supervised Learning is a process in which a computer program is trained
by using known example data. As the output is also known, this learning
process aims at finding a connection in the form of rules, which relate input
data to output data and finally apply the learned rules to new data. At this
point the computer program is getting trained. With this newly gained
Literature Review on Machine Learning in Supply Chain Management 421

knowledge it is now able to predict future input and output data. Two im-
portant tasks are classification and regression (Kirste and Schürholz, 2019,
25 f.).
Unsupervised Learning
Unsupervised learning describes a system that is able to discover
knowledge. Correct answers are not provided in this type of learning, thus,
there are no pre-labelled target values. This approach is also called "learn-
ing without a teacher". A well-known task of unsupervised learning is clus-
tering. The method identifies similarities between the inputs to categorize
inputs by common patterns. Similar tasks are e.g. Association Rules, Self-
Organizing Maps, Multidimensional Scaling and Nonlinear Dimension Re-
duction (Marsland, 2015, p. 281; Russell and Norvig, 2016, p. 694; Hastie,
Friedman and Tibshirani, 2017, 485 ff.).
Reinforcement Learning
With reinforcement learning, the optimal solution is unknown to the system
at the beginning of the learning phase and therefore must be determined
iteratively. In this process, sensible approaches are rewarded, and wrong
steps tend to be punished. With this approach, it is possible that the system
takes into account complex environmental influences and reacts accord-
ingly. Hence, the system finds its own solutions autonomously through di-
rectional rewards and punishments (Gentsch, 2018, 38 f.).
422 Hannah Wenzel et al.

3.3 Machine Learning Methods

Machine Learning methods can be categorized by the task they are solving.
As described in 3.2 classification and regression are two tasks of supervised
learning. Examples of tasks in unsupervised learning are clustering and as-
sociation rule mining. Another common type of learning is reinforcement
learning. Figure 2 shows selected methods of ML, grouped by the task they
commonly solve. It is important to mention that this selection of methods
is far from complete and some methods can be used to solve more than one
task. Also, there exist several variants of algorithms for each method.

Figure 2: Overview of Machine Learning Methods based on (Witten, Eibe and

Hall, 2011, pp. 116–216; The MathWorks, 2016, p. 7; Seif, 2018)

Part of the ongoing research in ML is the development of further methods,

models and algorithms. Selecting a model and an algorithm depends on
many factors and there is no one-method-fits-all-solution - also known as
"There is no free lunch in statistics" (James et al., 2013, p. 29). The selection
of an appropriate method for a given problem and dataset is one of the
most challenging tasks in data analysis (James et al., 2013, p. 29).
Literature Review on Machine Learning in Supply Chain Management 423

When using ML methods and algorithms, some challenges may occur. Two
different reasons can be mentioned when it comes to the development of
error sources. On the one hand, the problem can be found in the data. The
availability of a large amount of data which is necessary for the training of
a model is still a frequent challenge, as well as insufficiently representative
training data. Another challenge is poor data quality. In order to make pat-
terns visible in data sets, they first must be detected. If the data contains
errors or outliers, it is difficult to identify such patterns. Another issue is re-
dundant features, which do not provide added value to the model. An im-
portant step for the success of a ML method is therefore to select suitable
features for training (Géron, 2017, pp. 23–26). On the other hand, a poorly
chosen algorithm can create difficulties in the form of overfitting and un-
derfitting (Géron, 2017, p. 22). Overfitting means that ML algorithms can
sometimes generalize incorrectly although the model works on a training
data set. A model can be too complex, which means that the model follows
noise or errors in the data too closely (James et al., 2013, p. 22). Underfitting
is the opposite of overfitting and describes a model that is too simple for
the structure of the data to be recognized and learned (Géron, 2017, p. 29).

3.4 Machine Learning Process

Real world machine learning applications are usually developed as part of
a larger project. Several process model frameworks have been developed
for such projects. One popular one is the Cross Industry Standard Process
for Data Mining (Shearer, 2000; Piatetsky, 2014).
424 Hannah Wenzel et al.

Figure 3: Cross Industry Standard Process for Data Mining (Shearer, 2000,
p. 14)

The illustrated process in Figure 3 is also used for ML projects and acts as a
common standard for ML (Cohen, 2017). The process defines six phases
with specific tasks. Starting from "Business Understanding" to "Data Un-
derstanding", "Data Preparation", "Modeling", its "Evaluation" and ending
with "Deployment". The goal of the framework is to improve results while
reducing costs and time for the project. Shearer (2000) describes the spe-
cific tasks in more detail.
The paper is focused on presenting ML methods that are usually used dur-
ing the modeling phase, as this was also the focus of the examined papers.
Literature Review on Machine Learning in Supply Chain Management 425

4 Machine Learning Application Areas in Supply

Chain Management

This chapter portrays practical applications of ML by using three examples.

Each of these examples represents one of the three main tasks of the SCM
Model and is intended to show the reader how to use ML in practice. The
three applications are selected as each of them represents a concise struc-
ture, a comprehensible explanation and gives a first impression of ML ap-
plication.

4.1 Selecting Supply Chain Partners

In the field of SCD, one of the major challenges remains to find suitable
business partners such as customers or suppliers to exploit new opportuni-
ties. This is especially valid regarding the globalization and the fast devel-
opment of technology. To simplify finding new plausible business partners,
Mori et al. (2010) use ML techniques based on company profiles and trans-
action relationships.
For data collection, the authors obtained company data of suppliers and
customers via the "Teikoku Data Bank". A total of 30,660 companies from
the manufacturing industry could be retrieved from the databank, all of
them located in the Kanto and Koshinetsu areas of Japan. The training data
consisted of 34,441 supplier and customer pairs from the existing inter-firm
network. Those pairs are used as positive examples. Negative examples are
defined as randomly chosen supplier and customer pairs, which are not re-
lated.
426 Hannah Wenzel et al.

A Support Vector Machine (SVM) was applied to model the existing relation-
ships by using features inherent to the company, such as number of em-
ployees and capital, as well as features defined by the interlinkage of firms,
like customers of a supplier and common Industrial categories. The accu-
racy of the SVM-model reached up to 85%.
The most important determinants of customer-supplier relationships were
identified as number of employees or industry category. The authors ex-
plain the relevance of these categories with country specific characteristics
such as hierarchical structures. This shows that different determinants
might be relevant when applying the method elsewhere. Along with the
SVM, the ML model provides a tool to predict plausible candidates for future
customer-supplier relationships (Mori et al., 2010).

4.2 Demand forecast for DM

One of Germany's largest drugstore companies, dm, uses machine learning
algorithms to predict future demands. With 3,350 stores worldwide, the dm
drugstore chain runs six distribution centers to ensure the availability of
goods. The distribution centers have to ensure that the incoming goods of
the industry partners are arranged for the individual stores in such a way
that there is no product deficit and customer satisfaction remains high. Fre-
quently, stores require a delivery of products within a short time frame, but
the manufacturers have relatively long delivery times. This remains the big-
gest challenge for the distribution centers. Planning well in advance is ac-
companied by long-term orders, resulting in high storage costs and tied up
capital.
To overcome this issue, artificial intelligence algorithms are used to create
weekly demand forecasts based on SKU level over a time horizon of half a
Literature Review on Machine Learning in Supply Chain Management 427

year. For training of the algorithm, the data from the last 2.5 years of the
particular distribution center has been used and the seasonality factor has
been taken into account.
As a result of the artificial intelligence algorithms, the demand forecasts be-
came so precise over a period of six months that a significant improvement
in forecast quality could be achieved and industrial partners are now able
to plan much earlier. Delivery reliability and product availability can thus
be significantly optimized, which is supported by an automated infor-
mation flow of future requirements to industrial partners. The advantages
of using the algorithms are obvious for the industrial partners as well as for
the dm drugstore market. The industrial partners can count on increased
planning and ordering security through valid forecasts and the drugstore
chain has secure product availability and less excess stocks. The results are
satisfied customers and lower costs (JDA Software, 2019, pp. 8–9).

4.3 Detecting False-Positive RFID Tag Reads in Transport

Management
One of SC's key processes is the shipment of goods from distribution cen-
ters. An error-free shipment depends on several factors, be it the picking
process or the assignment of pallets to the allotted truck. To integrate a
control mechanism, the METRO Group Cash & Carry implemented RFID por-
tals to check on outgoing pallets. In its distribution center in Unna, Ger-
many, loading ramps are equipped with RFID portals to automatically de-
tect goods leaving the warehouse. For the detection of a pallet, it carries a
transponder. As soon as the pallets pass the portals, the reading device rec-
ords goods leaving the warehouse and automatically adjusts the available
428 Hannah Wenzel et al.

inventory via direct communication with the warehouse management sys-

tem.
The RFID identification supports the shipment of goods and the automatic
inventory adjustment but carries one major issue. The antennas of the por-
tals have ranges of several meters and therefore identify all pallets in range,
pallets to be shipped and pallets placed for intermediate storage. This
means that during a collection cycle it is not possible to distinguish be-
tween a moving pallet loaded into a truck and a static pallet that is not rel-
evant for the current shipment and hence, has been accidentally read. The
non-relevant pallets represent a false-positive reading of the RFID portals.
In order to differentiate between the moving and static pallets, METRO
Group Cash & Carry applied ML techniques based on the recorded attrib-
utes by each pallet. The attributes include the reception of the signal
strength indicator (RSSI), the time stamp of each tag event and the number
of reads per antenna based on the low-level reader data. It became appar-
ent, that the tags of static and moving pallets submit different RSSI values,
based on frequencies submitted to the portals. This is where ML comes in,
as classifying two characteristics according to their threshold value is a
well-known ML task. If the RSSI average value is below -58.1dBm, it is a
static tag (false positive); otherwise it is a moving tag.
To automatically evaluate the attributes by ML, a very large data set is re-
quired for the training phase. The distribution center installed RFID portals
on all 70 loading ramps and monitored them over a period of 7 months. A
total of 53,998 pallets were monitored, of which 40,743 were static and
13,245 were moved through the outgoing goods portal.
Literature Review on Machine Learning in Supply Chain Management 429

After completion of the data acquisition, the used algorithm could correctly
classify more than 95.5% of the data. Thus the pallets can be quickly iden-
tified and false-positive pallets can be captured directly. This minimized
faulty inventory adjustments and deliveries (Keller et al., 2010).

4.4 Mapping of Use Cases to the Supply Chain Task

Model
The analysis of the papers resulted in an assignment of the ML use cases to
all three main task areas of the SCM model, which are outlined below. Out
of illustration purposes in Figure 4, a number is assigned to each paper.
In the area of SCD, all retrieved papers deal with supplier selection. The au-
thors apply different ML methods. While Kong and Xue (2013) utilize Radial
Basis Function (RBF) neural networks [1], Zhang et al. (2017) use ensemble
learning, decision trees and logistic regression [2]. Similar to the case of
Mori et al. (2010) [3], the authors apply SVM, too. Another paper by Cheng
et al. (2017) integrates Adaboost on expert knowledge basis to deal with
supplier selection [4].
In the area of SCP, ML topics address the task of demand planning and pro-
curement planning. The number of papers in relation to other tasks is very
comprehensive. Based on the literature analysis carried through, fourteen
use cases could be assigned to this area.
The authors Xue et al. (2018) use machine learning to solve the dynamic
forecasting problem of the supply of goods in the event of a catastrophe [5].
Gamasaee, Zarandi and Turksen (2015) [6] and Sarhani and El Afia (2014)
deal with approaches to demand forecasts of SC. It should be noted that all
three papers use Support Vector Regression (SVR) as the ML method [7]. Ap-
430 Hannah Wenzel et al.

plication examples for artificial neural networks are provided by the au-
thors Gaur, Goel and Jain (2015), who apply a data set from Walmart to fore-
cast demand for the SC [8], and Yang and Sutrisno (2018), who distribute
perishable goods among franchise companies [9]. Adhikari et al. (2017) con-
tribute with an ensemble technique to combine different ML methods in or-
der to obtain demand forecasts [10]. The described practical application of
the drugstore chain dm can be assigned to SCP, too, as the methodology
creates a planning basis for a more effective SC to the stores. JDA Software
(2019) mentions that the goal is to achieve high store availability [11].
Souichirou (2015) deals with commodity demand forecasting and system-
atization of SCM by using heterogenous mixture learning, and uses this to
simulate sales measures [14]. As one of the rare studies so far, Cui et al.
(2018) [12] and Lau, Zhang and Xu (2018) [13] integrate social media sources
to forecast sales. The last three authors base their analysis on real data.
In the area of procurement planning, Hogenboom et al. (2009) apply RBF
Networks [15] to deal with product pricing as well as Kiekintveld et al.
(2009) who use a Bayesian Network and artificial neuronal network [16]. In
addition to the two authors mentioned above, Lee and Sikora (2016) also
deal with product pricing by using Q-learning softmax function [17]. In the
same context, Ketter et al. (2009) use gaussian mixture models [18]. It is in-
teresting to mention that all three authors deal with the same topic based
on the same synthetic data, but use different methods.
Within distribution planning Mokhtarinejad et al. (2015) outline vehicle
routing and scheduling problems in cross-docking systems by using Bi-
Clustering, K-means and artificial Neural Networks [19].
Literature Review on Machine Learning in Supply Chain Management 431

In the third, operationally structured SCE, the literature addresses the areas
of Order Management, Supply Chain Event Management, Inventory, Pro-
duction and Transport Management.
Order management contains a total of four use cases. Based on synthetic
data, the authors Sun and Zhao (2012) present a multi-agent coordination
mechanism using reinforcement learning to derive an optimal ordering
strategy for the entire SC with several levels [20]. Wang, Ng and Ng (2018)
group SKUs according to their demand and performance attributes by un-
supervised learning [21]. Real data is used in Wang and Liu (2009) to train a
model using neural networks, which represents an index system for evalu-
ating order priority [22]. Zhu, Ma and Zhang (2014) determine the manufac-
turing priority of an order with the help of RBF neural networks, Kriging and
SVR [23].
There are four use cases in the area of Supply Chain Event Management.
Among them, Arumugam, Umashankar and Narendra (2018) create an in-
telligent logistics solution that negotiates contracts, and includes logistics
planning and condition monitoring of the facilities [24]. The early detection
of supplier risks, Frerichs (2018) [25], as well as the identification of fraud
and deceptive practices Zage, Glass and Colbaugh (2013) [26] are topics ad-
dressed in the context of risk management. Both applications rely on real
data. Hiromoto, Haney and Vakanski (2017) use neural networks to identify
vulnerabilities within SC's and mitigate the consequences of unforeseen
risks [27].
A total of four use cases could be assigned to inventory management. The
authors Barbosa de Santis, Pestana de Aguiar and Goliatt (2017) use ML
classifiers to identify material residues within an inventory management
432 Hannah Wenzel et al.

system that have negative effects on inventory management [28]. Further-

more, Inprasit and Tanachutiwat (2018) use neural networks to optimize
safety stocks and reorder points for products, taking into account various
influencing factors such as performance and lead time [29]. The applica-
tions outlined above rely on real data. Another application is the identifica-
tion of technically obsolete spare parts in the warehouse. Supervised learn-
ing methods were used to solve this problem (Cherukuri and Ghosh;
Inprasit and Tanachutiwat, 2018) [30]. The authors Priore et al. (2018) use
decision trees on the basis of synthetic data to determine the optimal re-
plenishment rule [31].
Production management contains four use cases, while all of the papers
had access to real data. Ali Ahmadi et al. (2016) use machine learning ap-
proaches to differentiate between the manufacturing locations of prod-
ucts. ML distinguishes products that were manufactured in a ratified fac-
tory from the ones which originate from unknown sources [32]. In the paper
of Tirkel (2011) predictive models of ML and data mining are used to deter-
mine the cycle time [33]. By using linear regression and tree based meth-
ods, Dávid Gyulai et al. (2018) and Lingitz et al. (2018) estimate production
lead times [34], [35].
The described use case of Keller et al. (2010), whose authors use ML to dis-
tinguish between static and moving pallets, is assigned to transport man-
agement [36]. In addition, the authors Gulisano et al. (2018) formulate a
challenge to predict the arrival time and destinations of vessels [37].
An interesting approach is taken by Kao, Niraula and Whyatt (2018) who an-
alyze part identification to enable data curation. Hence, the paper can
cover several areas within the supply chain task model and is a rather gen-
eral topic.
Literature Review on Machine Learning in Supply Chain Management 433

Figure 4: Illustration of the ML use cases in the SC model with numbers of

the use cases based on (Hompel and Wolf, 2013, pp. 146–147)
434 Hannah Wenzel et al.

Figure 4 depicts the ML application cases described above sorted into the
respective supply chain tasks. The papers have been integrated into the
task model along with their assigned numbering. All tasks with ML use cases
are highlighted.
It shows that with four papers, SCD contains the lowest number of ML ap-
plications, all of them in the area of supplier selection. SCP has a total num-
ber of 15 use cases in the area of demand, procurement and distribution
planning, most of them allocated to demand planning. Within SCE, each
component of the SCM task model has at least one use case assigned. The
highest number of applications, 18, could be found in SCE. Many of the ap-
plications are based on real data, in some cases synthetic data serves to
verify the applicability of ML methods.

5 Conclusion

This paper assigns use cases of machine learning to the task model of Sup-
ply Chain Management, resulting in an overview of ML applications within
the different supply chain tasks. It was demonstrated that in the SCM task
model a single area could have different ML methods applied for a common
goal. A large portion of research focused on demand planning, with 10 out
of 38 papers handling this task. An investigation of ML methods for inter-
company areas such as SRM and CRM could be promising for SCM. Future
research should review relevant literature to suggest which methods apply
best to certain SCM tasks.
With regard to the standard process model, the research focus lies in the
modeling phase. Further research emphasizing on and explaining the con-
cepts used in Business, Data Understanding and Evaluation is required. The
Literature Review on Machine Learning in Supply Chain Management 435

examined papers cover these phases, but hardly present useful insights
and concepts for SCM. The Deployment phase is not treated at all.
436 Hannah Wenzel et al.

References

Abadi, M., Barham, P., Chen, J. and Chen, Z., 2016. Proceedings of OSDI ’16: 12th
USENIX Symposium on Operating Systems Design and Implementation. [e-
book]. Berkeley, CA. <[Link]
ference/osdi16/[Link]>.
Adhikari, N. C., Garg, R., Datt, S., Lalit Das, Deshpande, Srinivas and Misra, A., 2017.
Proceedings of the International Conference on Intelligent Sustainable Sys-
tems (ICISS 2017). Piscataway, NJ.
Ali Ahmadi, Mohammad-Mahdi Bidmeshki, Amit Nahar, Bob Orr, Michael Pas and
Yiorgos Makris, 2016. A machine learning approach to fab-of-origin attestation.
Arnold, D., Isermann, H., Kuhn, A., Tempelmeier, H. and Furmans, K., 2008. Hand-
buch Logistik. [e-book]. Berlin, Heidelberg. <[Link]
tent/pdf/10.1007%[Link]>.
Arumugam, S. S., Umashankar, V. and Narendra, C. N., 2018. IOT Enabled Smart Lo-
gistics Using Smart Contracts. [online].
Barbosa de Santis, R., Pestana de Aguiar, E. and Goliatt, L., 2017. Reordering Point
Determination using Machine Learning Technique for Inventory Management.
[online]. Piscataway, NJ [Accessed 3 May 2019].
Cheng, Y., Zhou, Z., Gu, X. and Liu, W., 2017. A Hybrid DEA-Adaboost Model in Sup-
plier Selection for Fuzzy Variable and Multiple Objectives. [online].
Cherukuri, M. B. and Ghosh, T., 2016. Control Spare Parts Inventory Obsolescence
by Predictive Modelling. In: X. Liu, ed. 2016. iThings-GreenCom-CPSCom-Smart
Data 2016. Piscataway, NJ, pp. 865–869.
Cohen, S., 2017. What is the most important step in a machine learning project?
[online]. Available at: <[Link]
[Link]>.
Copeland, M., 2016. What’s the Difference Between Artificial Intelligence, Machine
Learning, and Deep Learning? [online]. Available at:
<[Link]
gence-machine-learning-deep-learning-ai/>.
Literature Review on Machine Learning in Supply Chain Management 437

Cui, R., Gallino, S., Moreno, A. and Zhang, D. J., 2018. The Operational Value of So-
cial Media Information. Production and Operations Management, 27(10), pp.
1749–1769.
Dávid Gyulai, András Pfeiffer, Gábor Nick, Viola Gallina, Wilfried Sihn and László Mo-
nostori, 2018. Lead time prediction in a flow-shop environment with analytical
and machine learning approaches.
Frerichs, P., 2018. Welche Chancen künstliche Intelligenz dem Supply Chain Ma-
nage-ment bietet. <[Link]
cen-kuenstliche-intelligenz-dem-supply-chain-management-bietet/>.
Gamasaee, R., Zarandi, M. F. and Turksen, I. B., 2015. A type-2 fuzzy intelligent agent
based on sparse kernel machines for reducing bullwhip effect in supply chain.
[online] [Accessed 3 May 2019].
Gaur, M., Goel, S. and Jain, E., 2015. Comparison between Nearest Neighbours and
Bayesian Network for Demand Forecasting in Supply Chain Management. New
Delhi, India.
Gentsch, P., 2018. Künstliche Intelligenz für Sales, Marketing und Service. Mit AI und
Bots zu einem Algorithmic Business - Konzepte, Technologien und Best Prac-
tices. [e-book]. Wiesbaden: Springer Gabler. <[Link]
tent/pdf/10.1007%[Link]>.
Géron, A., 2017. Hands-on machine learning with Scikit-Learn and TensorFlow. Con-
cepts, tools, and techniques to build intelligent systems. [e-book]: O'Reilly.
<[Link]
Görz, G., Schneeberger, J. and Schmid, U., 2013. Handbuch der Künstlichen Intelli-
genz. [e-book]. München: Oldenbourg.
<[Link]
Gulisano, V., Jerzak, Z., Smirnov, P., Strohbach, M., Ziekow, H. and Zissis, D., 2018.
The DEBS 2018 Grand Challenge. In: Unknown, ed. 2018. DEBS'18. Proceedings
of the 12th ACM International Conference on Distributed and Event-Based Sys-
tems : June 25-29, 2018, Hamilton, New Zealand. New York, pp. 191–194.
Hastie, T., Friedman, J. and Tibshirani, R., 2017. The Elements of Statistical Learn-
ing. New York, NY: Springer New York.
438 Hannah Wenzel et al.

Hecker, D., Döbel, I., Petersen, U., Rauschert, A., Schmitz, V. and Voss, A., 2017. Zu-
kunftsmarkt künstliche Intelligenz. Potenziale und Anwendungen. [online].
Available at: <[Link]
[Link]> [Accessed 16 April 2019].
Hiromoto, R. E., Haney, M. and Vakanski, A., 2017. A Secure Architecture for IoT with
Supply Chain Risk Management. Romania.
Hogenboom, A., Ketter, W., van Dalen, J., Kaymak, U., Collins, J. and Gupta, A., 2009.
Product Pricing using Adaptive Real-Time Probability of Acceptance Estima-
tions based on Economic Regimes. New York, NY: ACM.
Hompel, M. ten and Wolf, O., eds., 2013. IT in der Logistik 2013/2014. Marktübersicht
& Funktionsumfang: [Link], Warehouse-Management,
Transport-Management & Supply-Chain-Management-Systeme. Stuttgart.
Inprasit, T. and Tanachutiwat, S., 2018. Reordering Point Determination Using Ma-
chine Learning Technique for Inventory Management. In: . 2018. "Exploring in-
novative solutions for smart society". Piscataway, NJ: IEEE, pp. 1–4.
James, G., Witten, D., Hastie, T. and Tibshirani, R., 2013. An Introduction to Statisti-
cal Learning. New York, NY: Springer New York.
JDA Software, 2019. Erfolgreicher Einsatz von Bedarfsprognosen bei dm. Wie künst-
liche Intelligenz bei der Drogeriemarktkette Mehrwert schafft. [online] [Acces-
sed 26 April 2019].
Kao, A., Niraula, N. B. and Whyatt, D., 2018. PANDA - Discovering Part Name in Noisy
Text Data. In: . 2018. 2018 IEEE International Conference on Prognostics and
Health Management (ICPHM). Piscataway, NJ, pp. 1–6.
Keller, T., Thiesse, F., Kungl, J. and Fleisch, E., 2010. Using Low-Level Reader Data to
Detect False-Positive RFID Tag Reads. In: . 2010. Internet of things (IOT), 2010.
Papers presented at the Second International Internet of Things Conference ;
Tokyo, Japan, 29 November - 1 December 2010. Piscataway, NJ: IEEE, pp. 1–8.
Ketter, W., Collins, J., Gini, M., Gupta, A. and Schrater, P., 2009. Detecting and fore-
casting economic regimes in multi-agent automated exchanges. Decision Sup-
port Systems, 47(4), pp. 307–318.
Literature Review on Machine Learning in Supply Chain Management 439

Kiekintveld, C., Miller, J., Jordan, P. R., Callender, L. F. and Wellman, M. P., 2009.
Forecasting market prices in a supply chain game. Electronic Commerce Re-
search and Applications, 8(2), pp. 63–77.
Kirste, M. and Schürholz, M., 2019. Entwicklungswege zur KI. In: V. Wittpahl, ed.
2019. Künstliche Intelligenz. Entwicklungswege zur KI. Berlin, Heidelberg: Sprin-
ger Berlin Heidelberg, pp. 25–26.
Kong, Z.-j. and Xue, J.-B., 2013. A comparative study of supplier selection based on
support vector machine and RBF neural networks.
Kuhn, A. and Hellingrath, B., 2002. Supply Chain Management. Optimierte Zusam-
menarbeit in der Wertschöpfungskette. [e-book]. Berlin, Heidelberg, s.l.: Sprin-
ger Berlin Heidelberg. <[Link]
Lau, R. Y. K., Zhang, W. and Xu, W., 2018. Parallel Aspect-Oriented Sentiment Analy-
sis for Sales Forecasting with Big Data. Production and Operations Manage-
ment, [e-journal] 27(10), pp. 1775–1794. [Link]
Lee, Y. and Sikora, R., 2016. Design of Intelligent Agents for Supply Chain Manage-
ment. In: V. Sugumaran, V. Yoon, and M. J. Shaw, eds. 2016. E-Life: web-enabled
convergence of commerce, work, and social life. 15th Workshop on e-Business,
WEB 2015 ; Fort Worth, Texas, USA, December 12, 2015 ; Revised Selected Pa-
pers: Springer International Publishing, pp. 27–39.
Lingitz, L., Gallina, V., Ansari, F., Gyulai, D., Pfeiffer, A., Sihn, W. and Monostori, L.,
2018. Lead time prediction using machine learning algorithms: A case study by
a semiconductor manufacturer. Procedia CIRP, [e-journal] 72, pp. 1051–1056.
[Link]
Marsland, S., 2015. Machine learning. An algorithmic perspective. Boca Raton, FL.
Mohri, M., Rostamizadeh, A. and Talwalkar, A., 2012. Foundations of machine learn-
ing. [e-book]. Cambridge MA. <[Link]
[Link]/e923/[Link]>.
Mokhtarinejad, M., Ahmadi, A., Karimi, B. and Rahmati, S. H. A., 2015. A novel learn-
ing based approach for a new integrated location-routing and scheduling prob-
lem within cross-docking considering direct shipment. Applied Soft Computing,
pp. 274–285.
440 Hannah Wenzel et al.

Mori, J., Kajikawa, Y., Sakata, I. and Kashima, H., 2010. Predicting customer-supplier
relationships using network-based features. In: . 2010. IEEE International Con-
ference on Industrial Engineering and Engineering Management (IEEM), 2010.
Piscataway, NJ, pp. 1916–1920.
Piatetsky, G., 2014. CRISP-DM, still the top methodology for analytics, data mining,
or data science projects. <[Link]
[Link]>.
Priore, P., Ponte, B., Rosillo, R. and La Fuente, D. de, 2018. Applying machine learn-
ing to the dynamic selection of replenishment policies in fast-changing supply
chain environments. International Journal of Production Research, 27(11).
Russell, S. J. and Norvig, P., 2016. Artificial intelligence. A modern approach. Bos-
ton, Columbus, Indianapolis.
Sarhani, M. and El Afia, A., 2014. Intelligent system based support vector regression
for supply chain demand forecasting. In: . 2014. 2014 Second World Conference
on Complex Systems (WCCS): IEEE, pp. 79–83.
Seif, G., 2018. The 5 Clustering Algorithms Data Scientists Need to Know. [online].
Available at: <[Link]
[Link]>.
Shearer, C., 2000. The CRISP-DM model: the new blueprint for data mining. J Data
Warehouse, 5.
Souichirou, N., 2015. Application of Big Data Technology in Support of Food Manu-
facturers’ Commodity Demand Forecasting. NEC Technical Journal, (10).
Steinbach, P., 2018. Wo noch nie eine CPU gewesen ist. In: . 2018. iX Developer 2018
- Machine Learning: Verstehen, verwenden, verifizieren, pp. 32–37.
Sun, R. and Zhao, G., 2012. Analyses about efficiency of reinforcement learning to
supply chain ordering management. [online].
The MathWorks, I., 2016. Introduction to Machine Learning. [online] [Accessed 30
April 2019].
Tirkel, I., 2011. Cycle time prediction in wafer fabrication line by applying data min-
ing methods. In: S. Lantz, ed. 2011. 2011 22nd Annual IEEE/SEMI Advanced Sem-
iconductor Manufacturing Conference (ASMC). Piscataway, NJ, pp. 1–5.
Literature Review on Machine Learning in Supply Chain Management 441

Waller, M. A. and Fawcett, S. E., 2013. Data Science, Predictive Analytics, and Big
Data: A Revolution That Will Transform Supply Chain Design and Management.
Journal of Business Logistics, (34(2)), pp. 77–84. <[Link]
[Link]/9c1b/[Link]>.
Wang, F., Ng, H. Y. and Ng, E.T., 2018. Novel SKU Classification Approach for Autono-
mous Inventory Planning. [online] [Accessed 6 May 2019].
Wang, X.-F. and Liu, B., 2009. Study on the method of order optimization and ar-
rangement based on neural network technology. In: . 2009. International Con-
ference on Machine Learning and Cybernetics, 2009. Piscataway, NJ, pp. 2697–
2701.
Witten, H. I., Eibe, F. and Hall, A. M., 2011. Data Mining: Practical Machine Learning
Tools and Techniques: Elsevier.
Xue, H., Jiang, C., Cai, B. and Yuan, Y., 2018. Research on demand forecasting of re-
tail supply chain emergency logistics based on NRS-GA-SVM. In: . 2018. Proceed-
ings of the 30th Chinese Control and Decision Conference (2018 CCDC). Pisca-
taway, NJ, pp. 3647–3652.
Yang, C.-L. and Sutrisno, H., 2018. Short-Term Sales Forecast of Perishable Goods
for Franchise Business. In: . 2018. "Cybernetics in the next decades". Pisca-
taway, NJ, pp. 101–105.
Zage, D., Glass, K. and Colbaugh, R., 2013. Improving supply chain security using big
data. In: K. Glass, ed. 2013. IEEE International Conference on Intelligence and
Security Informatics (ISI), 2013. Piscataway, NJ, pp. 254–259.
Zhang, H., Cheng, C. B., Wang, H., Yang, D., Liu, J., Nakayama, A., Cen, F. J., Liu, P.,
Jia, X., Zhang, H., Li, W., Pang, Y., HUANG, D., Xu, H. and uU, H., eds., 2017. Sup-
plier credit evaluation model based on ensemble learning. Riverwood, NSW:
Aussino Academic Publishing House.
Zhu, L.-y., Ma, Y.-z. and Zhang, L.-y., 2014. Ensemble model for order priority in
make-to-order systems under supply chain environment. In: H. Lan, ed. 2014.
International Conference on Management Science & Engineering (ICMSE), 2014.
Piscataway, NJ, pp. 321–328
.