Machine learning in supply chain management: A scoping review

Brylowski, Martin; Schröder, Meike; Lodemann, Sebastian; Kersten, Wolfgang Conference Paper Machine learning in supply chain management: A scoping review Provided in Cooperation with: Hamburg University of Technology (TUHH), Institute of Business Logistics and General Management Suggested Citation: Brylowski, Martin; Schröder, Meike; Lodemann, Sebastian; Kersten, Wolfgang (2021) : Machine learning in supply chain management: A scoping review, In: Kersten, Wolfgang Ringle, Christian M. Blecker, Thorsten (Ed.): Adapting to the Future: How Digitalization Shapes Sustainable Logistics and Resilient Supply Chain Management. Proceedings of the Hamburg International Conference of Logistics (HICL), Vol. 31, ISBN 978-3-7549-2770-0, epubli GmbH, Berlin, pp. 377-406, https://doi.org/10.15480/882.3961 This Version is available at: https://hdl.handle.net/10419/249623 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by-sa/4.0/ Published in: Adapting to the Future: Wolfgang Kersten, Christian M. Ringle and Thorsten Blecker (Eds.) ISBN 978-3-754927-70-0, September 2021, epubli Martin Brylowski, Meike Schroeder, Sebastian Lodemann, and Wolfgang Kersten Machine Learning in Supply Chain Management: A Scoping Review CC-BY-SA4.0 Proceedings of the Hamburg International Conference of Logistics (HICL) –31 Machine Learning in Supply Chain Management: A Scoping Review Martin Brylowski1, Meike Schroeder1, Sebastian Lodemann1 and Wolfgang Kersten1 1 – Hamburg University of Technology Purpose: Especially in supply chain management (SCM), data has become essential to the success of companies. Traditional analytical methods are being augmented by machine learning (ML), which is considered the foremost relevant branch of artificial intelligence. This article maps various ML use-cases and assigns them to the appropriate SCM tasks. Methodology: We applied a scoping review and checked scientific databases for relevant literature. Subsequently, the articles were assigned to different categories to map the research area. In the categorization, we considered, amongst others, the ML tasks and algorithms, data source and type, and the field of application. Findings: The results show that there are numerous ML use cases in SCM. These range from predictive demand forecasting and intelligent partner selection to the use of assistance systems for resource management. Various data sources, such as internal company data and publicly available data, are used for these applications. Originality: By mapping ML use cases in SCM, this complex and multifaceted field of research is presented in a transparent and structured way. Science and practice can deploy the results to improve existing ML use cases in SCM on the one hand and to identify promising areas of application on the other. First received: 21. Apr 2021 Revised: 29. Aug 2021 Accepted: 31. Aug 2021 Machine Learning in Supply Chain Management 1. Introduction A particularly important and ongoing topic in supply chain management (SCM) is the need to make complex decisions. Furthermore, supply chains are plagued by uncertainty and information asymmetry. In other words, on the one hand, decision-making processes in SCM depend on diverse, non-influenceable factors that are currently often influenced by information barriers (Ni, Xiao, and Lim, 2020, p. 1463). On the other hand, the volume and variety of data are constantly increasing, and there is more data available than ever before. Analyzing this data with the use of traditional methods is inefficient, or even impossible, in most cases. For this reason, new methods and applications are emerging from the field of advanced data analytics (ADA). One of these methods is machine learning (ML), which deals with the development and application of selflearning algorithms (Marsland, 2014, pp. 6-9; Wenzel, Smit, and Sardesai, 2019). ML has proven to be particularly capable of coping with large amounts of data and detecting patterns in the data. This property enables ML to make reliable decisions that humans are not capable of. Moreover, ML algorithms enable users to deal with rapidly changing conditions and discontinuous information. Thus, ML is a significant asset for SCM (Ni, Xiao, and Lim, 2020, p. 1463). Kersten et al. (2020) confirm this assumption and show the potential of ADA, and, thus, also ML, in SCM. Companies can particularly benefit from ML in the areas of planning, procurement, and delivery (Kersten et al., 2020). However, how is ML applied in SCM to make complex decisions related to uncertainties? The research goal of this article is to answer the following question: “What is the current state of research on the use of machine learning in supply chain management?” The remainder of this article is organized as follows. Chapter 2 describes the methodological approach and distinguishes this article from the current state of research. Chapter 3 includes the classification of the identified literature in terms of publication type, SCM task, industry focus, related data sources and types, and applied ML algorithms. Chapter 4 answers the underlying research question and identifies further research needs. Finally, Chapter 5 summarizes the results. Brylowski et al. (2021) 379 2. Methodological Approach Systematic literature reviews (SLR) are used in a wide range of research areas (Arksey and O'Malley, 2005). SLRs originated in the field of medicine. Later, Tranfield, Denyer, and Smart (2003) adapted SLRs to the needs and characteristics of management research and Durach, Kembro, and Wieland (2017) to those of SCM research. These adaptations are necessary because each research area has different requirements and characteristics (Durach, Kembro, and Wieland, 2017, p. 68). Currently, scoping reviews, which also follow a systematic approach, are becoming increasingly important for the aim of a synthesis (Munn et al., 2018). However, SLRs and scoping reviews have different objectives. Therefore, researchers should carefully determine which of the two procedures is appropriate to achieve their individual research objectives (Arksey and O'Malley, 2005). If, on the one hand, the research objective is to analyze the feasibility, appropriateness, usefulness, or effectiveness of specific practices, then an SLR is particularly well-suited to achieve the research objective. If, on the other hand, the research goal is to identify specific concepts and to present, report, or discuss those concepts, a scoping approach should be used (Munn et al., 2018, p. 144). Since the research objective of this article is to address the current state of research on the application of ML in SCM, the scoping review approaches of Arksey and O'Malley (2005) and Levac, Colquhoun, and O'Brien (2010) were chosen as the methodological foundation of this article. The following section describes the search and selection process for the scoping review to ensure that the research process is transparent. Furthermore, a differentiation of this research from the current state of research is presented to emphasize the unique research insights of this review. 2.1 Search and Selection Process According to Arksey and O'Malley (2005), scoping reviews assist in rapidly capturing the key concepts of a research area, especially when the area is complex or has not yet been analyzed. In addition, research gaps and designs can be identified and analyzed, or the macroscopic view can lay the groundwork for a further SLR (Munn et al., 2018, p. 144). Arksey and O'Malley (2005) and Levac, Colquhoun, and O'Brien (2010) described five Machine Learning in Supply Chain Management steps for conducting a scoping review. Figure 1 shows the step-by-step process of the conducted scoping review. 1) Identify the research question – The underlying research question for this paper is "What is the current state of research on the use of machine learning in supply chain management?" It was formulated by the authors to achieve the research objective of this paper. 2) Identify relevant studies – To identify relevant articles, the online databases Scopus, Web of Science, ScienceDirect and IEEE Xplore were selected. Specific search terms were determined, divided into two distinct word groups, and meaningfully related using Boolean operators for the targeted search for the relevant articles. The first word group included "machine learning" and the second word group "supply chain" or appropriate synonyms. The literature search was performed on January 1, 2021, with a search for a combination of the two word groups in the title, abstract, and keywords of the articles. Thus, an initial literature set of 1,878 articles were identified. 3) Study selection – The selection of relevant studies from the initial literature set included duplicate screening, abstract screening, and full text screening. Since the literature was retrieved from multiple databases, some articles were found to be duplicates, and these were removed during the duplicate screening process. Thus, the initial literature set was reduced from 1,878 to 1,145 articles. The abstracts of the articles were evaluated according to previously determined inclusion criteria, according to ten Hompel and Hellingrath (2007) and ten Hompel, et al. (2013). In addition, the ML method used in the article must involve one or more ML algorithms, according to Marsland (2014). Thus, the relevant articles could be further narrowed down, from 1,145 to 497. In the next step, full text screening was done. The previously mentioned inclusion criteria were used, but the full text of the articles was screened for relevance. After the full text screening, 183 articles were classified as relevant. In a further step, a search for additional literature was undertaken with Wohlin's (2014) snowballing approach. For this purpose, 14 literature reviews previously identified during the study selection process were examined to identify additional relevant literature. With consideration of the same inclusion criteria, 79 articles were added to the literature set. Finally, 262 articles that address ML algorithms in the SCM context were identified. Brylowski et al. (2021) 381 Figure 1: Methodological approach for the scoping review, following Arksey and O'Malley (2005) and Levac, Colquhoun, and O'Brien (2010). The next steps, according to Levac, Colquhoun, and O'Brien (2010), are 4) charting the data and 5) collating, summarizing, and reporting results. charting the data means classifying the data according to previously determined classes. The resulting classification is then analyzed, and the results presented descriptively. Collating, summarizing, and reporting results are done to provide an overview of the scope of the literature, but not in the form of a synthesis. Instead, a numerical analysis is done, presenting, for example, the scope and nature of the studies with the use of tables and charts. The scoping review is concluded with a thematic analysis. Steps 4 and 5, following Arksey and O'Malley (2005) and Levac, Colquhoun, and O'Brien (2010), are conducted in detail later in the article. 2.2 Related Studies As mentioned above, 14 literature reviews that had an objective comparable to that of this article were identified during the search and selection process. All 14 identified Four Databases: Scopus, Web of Science, ScienceDirect, IEEE Xplore Keyword: TITLE-ABS-KEY ( "machine learning" ) AND ( "supply chain" OR "suppl* network*" OR "value chain" OR "value network*" OR "operations management" OR "supplier*" ) 1878articles Duplicate Screening: The identified literature has been checked for duplicates 1145articles remaining Abstract Screening: Application of the exclusion criteria during abstract screening 497articles remaining Full text Screening: Application of the exclusion criteria during full text screening 183articles remaining Snowballing: 14 identified literature reviews has been checked for further relevant papers by applying the exclusion criteria via snowballing 262relevant articles Machine Learning in Supply Chain Management articles focus on the applications of ML or artificial intelligence in the logistics or SCM context. The main differences are in terms of the time periods considered, the research methods applied, the consulted databases, and the search terms used. The primary difference of this article, aside from its topicality, is the research method. The identified articles were all literature reviews, but not all were SLRs. However, none of the reviews followed the approach of a scoping review and the accompanying systematic quantitative analysis and interpretation of the research findings. Table 1 shows the 14 identified literature reviews and their respective focus topics. Table 1: Identified literature reviews No. Authors Focus of literature reviews 01 Bousqaoui, Achchab, and Tikito (2017) Supply Chain Management 02 Burggraf, Wagner, and Koke (2018) Production Management 03 Cadavid, Lamouri, and Grabot (2018) Demand & Sales Forecasting 04 Hachimi, Oubrich, and Souissi (2018) Reverse Logistics 05 Nguyen et al. (2018) Supply Chain Management 06 Baryannis et al. (2019) Supply Chain Risk Management 07 Calatayud, Mangan, and Christopher (2019) Supply Chain Management 08 Diez-Olivan et al. (2019) Industrial Applications 09 Ni, Xiao, and Lim (2020) Supply Chain Management 10 Wenzel, Smit, and Sardesai (2019) Supply Chain Management 11 Hosseini and Ivanov (2020) Supply Chain Risk Management Brylowski et al. (2021) 383 No. Authors Focus of literature reviews 12 Kumar et al. (2020) Distribution Planning 13 Seyedan and Mafakheri (2020) Demand Forecasting 14 Aamer, Yani, and Priyatna (2020) Demand Forecasting 3. Classification of Articles In the following section, the 262 identified articles are classified in terms of year of publication, publication type, SCM task, industry focus, related data sources and types, and ML algorithm applied to provide a comprehensive picture of the current state of research into ML in SCM. 3.1 Year and Type of Publication If we examine the number of publications over time, as depicted in Figure 2, it is notable that there is a sharp increase. The first noticeable increase was in 2005, when six articles on the topic of ML in SCM were published. By 2015, with the exception of some fluctuation, the number of publications per year had more than doubled, from six to 17. The most interesting development in terms of the absolute number of publications is evident during the period from 2015 to 2020. After a decrease from 17 publications in 2016 to 11 in 2017, there was rapid increase during the following two years. During 2019, 60 papers were published. ML in SCM seems to have gained importance, which can be related to the increasing amount and importance of data in research and in practice (Kersten et al., 2020). The decrease in the number of publications in 2020 and 2021 can be explained by the fact that, as described in Subchapter 2.1, the literature search was conducted on January 1, 2021, and not all submitted and published articles had yet been entered into the databases. Machine Learning in Supply Chain Management article, Zhang, Li, and Peng (2020) demonstrate how risks in the transportation of fresh products can be assessed and monitored. The remaining 23 articles were assigned to the class “No reference to industry” because, as the designation indicates, they include no references to an industry, according to ISIC. The analysis of the identified articles related to ISIC industries showed that manufacturing as well as wholesale and retail particularly benefit from ML in the context of the various SCM tasks. 3.4 Data Sources and Types ML applications or EDA require data that is available in structured or unstructured forms (Marsland, 2014, p. 6). Ideally, however, a high quantity of data should be available, it should be of high quality, and it should be both diverse and granular (Kersten et al., 2020). Therefore, in the following section, we first classify the identified articles in terms of the data sources used, following Ni, Xiao, and Lim (2020), which is followed by a classification of the articles in terms of the data types considered, following Seyedan and Mafakheri (2020). As previously mentioned, ML applications require data for analysis (Marsland, 2014, p. 6). This data can be obtained from a wide variety of data sources as Ni, Xiao, and Lim (2020) describe in their literature review. Figure 5 shows the data sources that were used to examine the problems in the different SCM tasks in the 262 identified articles. One hundred and fifty-three of the identified articles used historical data provided by companies for scientific research. Baryannis, Dani, and Antoniou (2019) used historical data provided by several companies in a manufacturing aerospace supply chain over a six-year period to predict supply chain risks. Furthermore, 47 articles used publicly available data for their research. One example is Wichmann et al. (2020), who generated supply chain maps using news articles. Twenty-seven of the identified articles analyzed data from laboratories with ML algorithms. Alfian et al. (2020) deserve a mention as a striking example with their perishable food traceability system, based on internet of things sensors and ML algorithms. A total of 25 articles included interviews with experts to generate or verify data. Badurdeen et al. (2014) developed a supply chain risk Brylowski et al. (2021) 391 taxonomy, which was subsequently incorporated into the ML analysis, through expert interviews. Finally, 11 of the identified articles used simulation data and 10 used data from other articles. Figure 5: Classification in relation to data source, following Ni, Xiao, and Lim (2020) Companies seem to have recognized the potential of ML in SCM and, therefore, make their data available for scientific analysis. In addition, publicly available data is often used in analyses with ML to gain new insights in the research area of SCM. In addition to the data source, the data type also plays a significant role in the application of ML in SCM. Following Seyedan, and Mafakheri (2020), Figure 6 shows the data types that were used in the 262 identified articles for the ML analyses. It is noticeable that most of the data types include information regarding external partners or were sourced from operational SCM task areas. Seventy-seven articles used supplier data and interpreted it with ML. Brintrup et al. (2019) analyzed supplier-related data to predict supplier disruptions in the manufacture of complex equipment. Gružauskas, Gimžauskienė, and Navickas (2019) provide another example of using data that includes information about external partners. They evaluated sales-related data to evaluate the predictive accuracy of logistics cluster activities. A total of 51 articles used sales-related data for ML analyses. Demand data was analyzed in 49 articles in total. Bhosekar and Ierapetritou (2021) developed a framework for supply chain optimization in modular production, using demand data. A final class that should be mentioned in the context of data types that include information regarding external partners is customer data. Twenty-eight articles were assigned to this class, including Simeone, Zeng, and Caggiano (2020), who 153 47 27 25 11 10 Identified data sources Historical data (Company) Historical data (Public) Historical date (Laboratory) Expert judgements Simulation data Data based on other studies Machine Learning in Supply Chain Management evaluated customer data for decision making to recommend manufacturing solutions in a cloud environment. In addition, data types from operational SCM tasks play a significant role in SCM. There were 41 articles that used manufacturing data for ML analysis. González Rodríguez, Gonzalez-Cava, and Méndez Pérez (2019) used productionrelated data for ML-based production planning, and 39 articles evaluated inventory data to make predictions regarding backorders (Hajek and Abedin, 2020). Furthermore, 26 articles explored the use of sensor data in the SCM context. For example, Ma, Wang, and Wang (2018) evaluated RFID sensor data with ML to detect false-positive decisions. Finally, product-related data was evaluated in 25 articles. In this context, Zgodavova et al. (2020) interpreted product-related data to improve small batch production systems. Figure 6: Classification in relation to data types following Seyedan and Mafakheri (2020). In summary, data containing information regarding external partners, such as suppliers and customers, is particularly likely to be used for ML-based analysis to make strategic decisions in the SCM context. In addition, the application of ML in operational SCM task areas lead to improved processes. Particularly production and logistics-related data types are frequently examined for this purpose. 3.5 Machine Learning Algorithms In addition to the data source and type, the choice of the optimal algorithm plays an essential role in ML applications. Marsland (2014) distinguishes between supervised, unsupervised, and reinforcement learning. Supervised learning refers to the analysis of structured input data and can be further differentiated to classification and regression algorithms (Marsland, 2014, p. 6). On the one hand, regression algorithms are primarily 77 51 49 41 39 28 26 25 Identified data types Supplier data Sales data Demand data Manufacturing data Product data Sensor data Customer data Inventory data Brylowski et al. (2021) 393 used for forecasts and to predict the course of data points (Marsland, 2014, pp. 6-8). Classification algorithms, on the other hand, assign defined classes to data and bundle them based on previously learned rules (Marsland, 2014, pp. 8-9). Unsupervised learning algorithms work with unstructured data and independently attempt to detect unknown patterns in the data to, for example, subsequently cluster them or to detect anomalies (Marsland, 2014, p. 6). Reinforcement learning approaches are a hybrid of supervised and unsupervised learning. When the algorithm makes incorrect decisions, it is corrected, but it does not learn the solution. This process continues until the algorithm has independently derived the optimal solution through trial and error (Marsland, 2014, p. 6). Figure 7 illustrates the classification of the identified articles and use cases in terms of the ML algorithms used, according to Marsland (2014), on the vertical axis and the underlying SCM tasks, according to ten Hompel et al. (2013) and ten Hompel and Hellingrath (2007), on the horizontal axis. In the remainder of this article, we first describe articles that use supervised learning algorithms for ML-based analysis. This is followed by examples of unsupervised learning and, thereafter, articles from the reinforcement learning class. Figure 7: Classification in terms of algorithms following Marsland (2014); Ten Hompel et al. (2013) and Ten Hompel and Hellingrath (2007). Supply Chain Design Supply Chain Planning Supply Chain Execution Supply Chain Collaboration Supervised Learning Unsupervised Learning Reinforcement Learning 1 3 1 4 20 60 14 54 19 51 11 22 36 29 178 Classification Regression Others Machine Learning in Supply Chain Management A total of 54 of the identified articles could be assigned in the classification algorithms class in the SCD. Harikrishnakumar et al. (2019) used classification algorithms to classify suppliers. ML applications that can be assigned to the regression class in the SCD are represented by 19 articles. For example, Brintrup et al. (2019) used regression algorithms to predict supplier risks. In the SCP task area of SCM, 20 articles used classification algorithms and 60 articles used regression algorithms. González Rodríguez, GonzalezCava, and Méndez Pérez (2019) used classification algorithms for intelligent production planning in the SCP task area, and Priyadarshi et al. (2019) addressed the prediction of vegetable demand using regression analysis. In the SCE task area of SCM, 51 articles that applied classification algorithms for ML based analysis were identified. However, 11 articles applied regression algorithms for operational SCE tasks. In this context, Ahmadi, Javidi, and Shahbazmohamadi (2018) are worth mentioning because they used classification algorithms to identify counterfeits in the electrical industry. Amirkolaii et al. (2017) showed the potential of regression algorithms for predicting the demand of spare parts in the aerospace industry. Finally, in terms of supervised learning algorithms, in the SCC task domain, 36 articles could be assigned to the classification algorithms and 29 to the regression algorithms. In the collaborative SCM task domains, 36 articles describe the use of ML-based classification algorithms and 29 describe the use of regression algorithms. Illustrative examples include Abdollahnejadbarough et al. (2020), with their classification approach for supplier base rationalization, and Li et al. (2020), with their regression approach for sustainable production capacity assessment. In addition to the supervised learning algorithms, articles that followed an unsupervised learning approach were also identified in all SCM task areas. In SCD, a total of eight articles followed an unsupervised learning approach. For example, Khalid and HerbertHansen (2018) showed how international location decisions can be made with clustering algorithms. In the tactical SCP, 14 articles could be identified. In this context, Meiners et al. (2019) investigated how production ramp-up can be accelerated and relationships in the process chain can be derived. There were 22 articles in the operational SCE domain. Becker and Intoyoad (2017) demonstrated how information on logistics processes can be extracted using ML in a data-based manner. Finally, 17 articles were assigned to Brylowski et al. (2021) 395 collaborative SC tasks (SCC). Sabbagh and Ameri (2018) developed an ML approach to cluster suppliers based on unstructured production data. As previously mentioned, it was also possible to identify articles that used reinforcement algorithms for ML-based analysis. One article that could be assigned to the SCD is by Zhang and Bhattacharyya (2007), who determined the design of a supply network. From the SCP domain, three articles could be identified. Vanvuchelen, Gijsbrechts, and Boute (2020) developed an approach to solve joint replenishment problems using deep reinforcement. In the SCM task domain SCE, similar to SCD, one article could be identified. Döring, Dangelmaier, and Danne (2007) applied reinforcement learning to production systems in the automotive industry. Finally, four articles were identified in the SCC context. Peng et al. (2019) are investigated the prediction of supply chain wide capacity under uncertainty. In summary, with the supervised learning algorithms in the SCD and SCE task domains, significantly more classification algorithms are used than regression algorithms. In the SCP domain, the opposite effect can be seen. In the SCC, classification and regression analyses are almost equally distributed. Unsupervised learning algorithms, however, are primarily used in the SCE domain. The SCM tasks SCP and SCC form the middle field in unsupervised learning context. In SCD, unsupervised learning is uncommon. Reinforcement learning does not seem to be used frequently in the SCM context in general and is rather underrepresented in all SCM task domains, compared to supervised and unsupervised learning approaches. 4. Discussion and Further Research Data analytics and ML are becoming increasingly relevant, especially in the SCM context (Kersten et al., 2020). Therefore, the research objective of this article was to analyze for which specific SCM tasks ML is applied. To achieve this research goal, the following research question was derived and answered progressively throughout this article through the analysis of the identified articles in a scoping review, based on Arksey and O'Malley (2005) and Levac, Colquhoun, and O'Brien (2010): Machine Learning in Supply Chain Management “What is the current state of research on the use of machine learning in supply chain management?” In the remainder of this article, the core findings of this work are discussed, and two recommendations for further research that could examine the topic of ML in SCM from a new direction in terms of methodology and content are provided. First, it is worth mentioning that only articles using EDA or SLR as research methods were identified. These articles either describe the detailed implementation of ML algorithms in the SCM context (EDA) or provide a systematic overview of the current state of research (SLR). Qualitative interview studies, quantitative surveys, or conceptual papers on ML in SCM describing, for example, procedural models for the step-by-step and situationappropriate implementation of ML in the different SCM task areas, could not be identified for this scoping review. Three main application areas for ML in SCM emerged in this scoping review. These areas are supplier management, which can be assigned to the SCM task areas SCD and SCC, demand management, which is predominantly applied in SCP and SCC, and quality management, that can be assigned to SCE. Articles that involved ML analyses in the main application areas of supplier and demand management often used structured data that contain cross-company information regarding external SC partners, such as supplier-, sales-, and demand-related data. It is worth mentioning that supplier management is almost exclusively considered in the context of the manufacturing industry, and ML applications in demand management are more prevalent in the wholesale and retail sector. Furthermore, supplier management and demand management differ in terms of the ML algorithms used. Articles on the topic of supplier management generally use classification algorithms, and ML applications on the topic of demand management use regression algorithms. In quality management, production, product, and sensor data are analyzed, often with structured data in combination with unstructured data. Both supervised and unsupervised algorithms are used. The three main application areas identified, supplier management, demand management, and quality management, differ fundamentally in terms of the industry addressed, data source, data types, and ML algorithms. Future work should explore the three identified main application areas for ML in SCM in detail, in terms of industry, data source, data type, and ML algorithms. Brylowski et al. (2021) 397 5. Conclusion From this review, it can be concluded that ML in SCM has gained relevance over recent years, as the number of publications has increased. ML is frequently used for crosscompany SCM tasks to improve collaboration with suppliers and customers or to monitor internal processes, identify weak points, and initiate mitigation activities in a timely manner. Manufacturing companies and retailers are particularly likely to make their data available to academia, which can be evidenced by the high number of articles relating to these industries, thus exploiting the potential of ML to improve their strategic, tactical, operational, and collaborative SCM processes. Publicly available data is brought in by researchers to generate further value-adding insights and, thus, enhance the performance of ML algorithms in the SCM context. The data types most frequently used include information about suppliers or customers as well as information from the production environment or warehousing. This again underpins the relevance of ML in terms of strategic cross-company as well as tactical and intra-company applications. If we consider the distribution of classification and regression algorithms in the individual SCM task areas, it is notable that they are distributed differently. This is because, in SCD and SCE, categorical decisions are made, such as "is this supplier eligible - yes or no" or "is the quality acceptable - yes or no." In comparison, SCP deals with continuous questions, such as "how is demand evolving?" The equal distribution in SCC can be explained by the fact that both categorical and continuous issues are discussed in this SCM task domain. Unsupervised learning algorithms, however, are primarily used in the SCE domain. This is because image or video data, which is unstructured data, are often interpreted in the context of quality and production monitoring. The main application areas of ML in SCM are in supplier management, demand management and quality management, and it differ fundamentally in terms of the industry considered, data source, data types, and ML algorithms. Machine Learning in Supply Chain Management References Aamer, A. M., Yani, L. P. E. and Priyatna, I. M. A., 2020. Data Analytics in the Supply Chain Management: Review of Machine Learning Applications in Demand Forecasting. Operations and Supply Chain Management: An International Journal, pp. 1–13. http://dx.doi.org/10.31387/oscm0440281. Abdollahnejadbarough, H., Mupparaju, K. S., Shah, S., Golding, C. P., Leites, A. C., Popp, T. D., Shroyer, E., Golany, Y. S., Robinson, A. G. and Akgun, V., 2020. Verizon Uses Advanced Analytics to Rationalize Its Tail Spend Suppliers. INFORMS Journal on Applied Analytics, [e-journal] 50(3), pp. 197–211. http://dx.doi.org/10.1287/inte.2020.1038. Ahmadi, B., Javidi, B. and Shahbazmohamadi, S., 2018. Automated detection of counterfeit ICs using machine learning. Microelectronics Reliability, [e-journal] 88-90, pp. 371–377. http://dx.doi.org/10.1016/j.microrel.2018.06.083. Alfian, G., Syafrudin, M., Farooq, U., Ma'arif, M. R., Syaekhoni, M. A., Fitriyani, N. L., Lee, J. and Rhee, J., 2020. Improving efficiency of RFID-based traceability system for perishable food by utilizing IoT sensors and machine learning model. Food Control, [e-journal] 110, p. 107016–107016. http://dx.doi.org/10.1016/j.foodcont.2019.107016. Amirkolaii, K. N., Baboli, A., Shahzad, M. K. and Tonadre, R., 2017. Demand Forecasting for Irregular Demands in Business Aircraft Spare Parts Supply Chains by using Artificial Intelligence (AI). IFAC-PapersOnLine, [e-journal] 50(1), pp. 15221– 15226. http://dx.doi.org/10.1016/j.ifacol.2017.08.2371. Arksey, H. and O'Malley, L., 2005. Scoping studies: towards a methodological framework. International Journal of Social Research Methodology, [e-journal] 8(1), pp. 19– 32. http://dx.doi.org/10.1080/1364557032000119616. Babajanivalashedi, R., Baboli, A., Shahzad, M. K. and Tonadre, R., 2018. A Predictive Approach to Define the Best Forecasting Method for Spare Parts: A Case Study in Business Aircrafts’ Industry. International Conference on Industrial Engineering and Engineering Management. Badurdeen, F., Shuaib, M., Wijekoon, K., Brown, A., Faulkner, W., Amundson, J., Jawahir, I. S., J. Goldsby, T., Iyengar, D. and Boden, B., 2014. Quantitative modeling and Brylowski et al. (2021) 399 analysis of supply chain risks using Bayesian theory. Journal of Manufacturing Technology Management, [e-journal] 25(5), pp. 631–654. http://dx.doi.org/10.1108/JMTM-10-2012-0097. Baryannis, G., Dani, S. and Antoniou, G., 2019. Predicting supply chain risks using machine learning: The trade-off between performance and interpretability. Future Generation Computer Systems, [e-journal] 101, pp. 993–1004. http://dx.doi.org/10.1016/j.future.2019.07.059. Baryannis, G., Validi, S., Dani, S. and Antoniou, G., 2019. Supply chain risk management and artificial intelligence: state of the art and future research directions. International Journal of Production Research, [e-journal] 57(7), pp. 2179–2202. http://dx.doi.org/10.1080/00207543.2018.1530476. Becker, T. and Intoyoad, W., 2017. Context Aware Process Mining in Logistics. CIRP Conference on Manufacturing Systems. Bhosekar, A. and Ierapetritou, M., 2021. A framework for supply chain optimization for modular manufacturing with production feasibility analysis. Computers & Chemical Engineering, [e-journal] 145, p. 107175–107175. http://dx.doi.org/10.1016/j.compchemeng.2020.107175. Bousqaoui, H., Achchab, S. and Tikito, K., 2017. Machine Learning applications in supply chains. Piscataway, NJ: IEEE. Available at: Essaaidi, Mohammed (HerausgeberIn) Zbakh, Mostapha (HerausgeberIn). <http://ieeexplore.ieee.org/servlet/opac?punumber=8277265>. Brintrup, A., Pak, J., Ratiney, D., Pearce, T., Wichmann, P., Woodall, P. and McFarlane, D., 2019. Supply chain data analytics for predicting supplier disruptions: a case study in complex asset manufacturing. International Journal of Production Research, pp. 1–12. http://dx.doi.org/10.1080/00207543.2019.1685705. Burggraf, P., Wagner, J. and Koke, B., 2018. Artificial intelligence in production management: A review of the current state of affairs and research trends in academia. Cadavid, J. P. U., Lamouri, S. and Grabot, B., 2018. Trends in Machine Learning Applied to Demand & Sales Forecasting: A Review. Machine Learning in Supply Chain Management Zhang, G., Li, G. and Peng, J., 2020. Risk Assessment and Monitoring of Green Logistics for Fresh Produce Based on a Support Vector Machine. Sustainability, [e-journal] 12(18), p. 7569–7569. http://dx.doi.org/10.3390/su12187569. Zhang, Y. and Bhattacharyya, S., 2007. Effectiveness of Q-learning as a tool for calibrating agent-based supply network models. Enterprise Information Systems, [ejournal] 1(2), pp. 217–233. http://dx.doi.org/10.1080/17517570701275390.