Available online at www.sciencedirect.

com

ScienceDirect
Procedia CIRP 107 (2022) 1126–1131
www.elsevier.com/locate/procedia

55th CIRP Conference on Manufacturing Systems

Review and analysis of artificial intelligence methods for demand
forecasting in supply chain management
Mario Angos Mediavillaa*, Fabian Dietricha,b, Daniel Palma,c
a
Reutlingen University, Alteburgstr. 150, 72762 Reutlingen, Germany
b
Stellenbosch University, 145 Banghoek Rd., 7600 Stellenbosch, South Africa
c
Fraunhofer Institute for Manufacturing Engineering and Automation, Alteburgstr. 150, 72762 Reutlingen, Germany

* Corresponding author. Tel.: +49 151 70147903. E-mail address: [email protected]

Abstract

The proper selection of a demand forecasting method is directly linked to the success of supply chain management (SCM). However, today's
manufacturing companies are confronted with uncertain and dynamic markets. Consequently, classical statistical methods are not always
appropriate for accurate and reliable forecasting. Algorithms of Artificial intelligence (AI) are currently used to improve statistical methods.
Existing literature only gives a very general overview of the AI methods used in combination with demand forecasting. This paper provides an
analysis of the AI methods published in the last five years (2017-2021). Furthermore, a classification is presented by clustering the AI methods
in order to define the trend of the methods applied. Finally, a classification of the different AI methods according to the dimensionality of data,
volume of data, and time horizon of the forecast is presented. The goal is to support the selection of the appropriate AI method to optimize
demand forecasting.
© 2022 The Authors. Published by Elsevier B.V.
This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0)
Peer-review under responsibility of the International Programme committee of the 55th CIRP Conference on Manufacturing Systems

Keywords: Demand Forecasting; Supply Chain Management; Artificial Intelligence; Machine Learning; Deep Learning; Review; Analysis

1. Introduction Different methods are adapted to forecast demand. These can

be statistical, AI-based, or hybrid methods, which combine the
The era of stable markets is history. Nowadays companies characteristics of a statistical model with a model from the AI
are confronted with highly complex scenarios characterized by domain [8]. Statistical methods provide accurate forecasting
unpredictable situations such as monetary crisis, pandemics, results and are very useful. Currently, due to increasing data
climate change, supply constraints. Additionally, growing dimensions and data volume, these classical methods face
demand for fully customized products in the shortest possible challenges and do not always meet the requirements of
time has aggravated this situation [1]. This increasing manufacturing companies [6,9,10]. With the use of AI in SCM
complexity hinders the desired transparency of material and new methods have been proposed, which combine traditional
information flow between supply chain members, which is time series forecasting with machine learning methods or use
essential to ensure the success of supply chain planning (SCP) artificial neural networks to refine and improve the demand
and to improve the overall value chain resilience [2]. Demand forecasting process [11]. Machine learning (ML) is a subarea of
forecasting represents a solid basis for planning and AI which works with self-learning algorithms. ML methods
procurement processes that make the supply chain more aim to improve their results based on experience gained from
responsive and efficient [3,4]. Therefore, the improvement of available historical data [12]. Furthermore, ML methods have
demand forecasting methods has become more and more been shown to perform well with large amounts of noisy data,
important for manufacturers, distributors and retailers [5–7]. such as those typically found in historical demand data. Due to

2212-8271 © 2022 The Authors. Published by Elsevier B.V.

This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0)
Peer-review under responsibility of the International Programme committee of the 55th CIRP Conference on Manufacturing Systems
10.1016/j.procir.2022.05.119

This is a resupply of March 2023 as the template used in the publication of the original article contained errors. The content of the article has remained unaffected.
 Mario Angos Mediavilla et al. / Procedia CIRP 107 (2022) 1126–1131 1127

the large number of existing AI methods and the lack of comparison between supply chain (SC) members and how
transparency in their classification, Liu et al. propose a much data is collected for demand forecasting and shows which
taxonomy to classify them into traditional ML methods and related AI methods are applied.
deep learning methods. Subsequently, each category can be The first field in the table Area shows the industry sector of the
divided into supervised and unsupervised learning [13]. publications. The second field SCM defines the members(s) of
Demand forecasting represents a research field in continuous the supply chain involved in the forecasting process.
development [14], which increases the complexity of The field Data describes the dataset that is used for training and
identifying a suitable method for each scenario. In addition, the validation of the AI algorithms. This field has three subfields
current literature provides a limited overview of demand to describe the datasets in more detail to determine the volume
forecasting methods, especially in relation to manufacturing of the dataset. Each dataset can consist of several input
companies. Therefore, the aim of this paper is the review and attributes. For that reason, the dataset is analyzed according to
analyze currently used AI methods, focusing on demand dimensions, divided into univariate and multivariate, under the
forecasting in SCM of manufacturing networks. Furthermore, a field Dimensions. The Range of data collection is provided
classification is presented by clustering the AI methods in order consistently in months for better comparison between
to define the trend of the methods used. Finally, this publication publications. In addition, the Frequency of data collection can
presents a classification of different methods based on data be daily, weekly, monthly, or quarterly.
characteristics, which supports the AI method selection for Furthermore, all publications are analyzed according to the AI
demand forecasting based on user requirements. methods used to predict demand. In the AI field, the
publications are analyzed according to four subfields. All AI
2. Methodology methods used in the publications are listed in the subfield
Method. The publications use a single method (A), several
This paper uses a structured literature review to identify methods (A, B), and combined (A+B) methods. The Time
existing AI methods for demand forecasting. A structured horizon field describes the horizon forecasted by the AI
literature review is relevant for the analysis of a specific topic method. The time horizon is defined as short-term (hourly,
[15] and the identification of knowledge gaps [16]. Several daily, weekly), medium-term (monthly), and long-term
methods for conducting a literature review are currently (yearly). The metrics applied in the publications to evaluate the
available. This publication adopts the concept presented by AI methods are listed under the subfield Metrics. Tool
Snyder et al. [16] and follows the approach proposed by Patel represents the last subfield and lists the software or
et al. [17]. The objective of this review is to analyze the state- programming language used for the implementation.
of-the-art methods from different AI areas applied to demand
forecasting, specifically in the area of SCM, with a focus on 3.2 Overview and analysis of the literature
manufacturing. The databases used to collect the publications
are Web of Science, IEEE Explore and Springer Publishing. All According to the industrial sector, seven publications come
three databases are searched and accessed in October 2021. from the logistics area. After analyzing the supply chain
Subsequently, the following search sequence is defined in members involved in the forecasting process, four publications
combination of Strings with Boolean operators: ("demand analyze the situation from the retailer's point of view and seven
forecasting" OR "demand prediction") AND ("artificial from the manufacturer's perspective.
intelligence" OR "machine learning" OR "deep learning") Concerning the analyzed field of data, 55 percent of the data
AND ("supply chain management"). The search range is are univariate input data and only historical data sales are used
restricted to publications from the last five years (2017 - 2021). without additional attributes. Concerning multivariate inputs,
In the first stage, 92 publications that met the initial the publications consider a minimum of 3 attributes [5] and a
requirements are evaluated. In the following sections, the maximum of 16 additional attributes [6,18] in demand
authors consider only publications related to the manufacturing forecasting. Typical additional attributes used throughout the
domain. literature are climate, temperature, locality, etc. Furthermore,
the data ranged from 1 to 168 months and almost 50 percent of
3. Structured literature review the data were collected daily. The other half of the data is
collected on a weekly or monthly basis. All AI methods used
The structural literature search identified 23 publications in the publications are specified in the table to provide a general
that fulfilled the requirements defined in the methodology. First overview. They will be analyzed in detail in section 4. The
the terminology is defined and subsequently the characteristics analysis shows a tendency towards short- to medium-term
found are analyzed. forecasts. Only a single publication forecasted long-term
demands. The metrics most used in the publications are relative
3.1 Definition of terminology mean square error RMSE (x10), mean absolute percent error
MAPE (x6), mean absolute error MAE (x5) and mean square
This section defines the terminology used to analyze the error MSE (x5). Finally, the most used software is Python (x6),
publications. The analyzed fields are defined as Area, SCM, followed by R (x5) and MATLAB (x4).
Data, and AI (Table 1). The purpose of the table is to allow a

This is a resupply of March 2023 as the template used in the publication of the original article contained errors. The content of the article has remained unaffected.
 1128 Mario Angos Mediavilla et al. / Procedia CIRP 107 (2022) 1126–1131

Table 1. Overview of literature using AI-based forecasting methods in SCM in the manufacturing domain from 2017 to 2021.

Data AI
Area SCM
Range Time
Dimension Frequency Metho𝐝𝐝𝐝𝐝𝐚𝐚𝐚𝐚 Metric𝐬𝐬𝐬𝐬 𝐛𝐛𝐛𝐛 Tool
months horizon

Automotive Manufacturer Multivariate 108 Weekly MLP, RF, SVR Short-term NMAE R [19]
Automotive Manufacturer Univariate + 72 Quarterly AdaBoost, ANN, Medium-term RMSE, R² - [5]
Multivariate GB
Manufacturing Retailer Univariate 168 Monthly ANN, ARIMA + LR Medium- term AE, MAPE, TE R [20]

Logistics Retailer Multivariate 2 Daily Classification DT Short-term ACC, PR, RE WEKA [4]
Building and Manufacturer Multivariate 72 Monthly ARIMAX + ANN Medium-term FA, MAPE - [21]
Construction
Logistics Manufacturer Multivariate 4.5 Weekly K-means + QRF + RF Medium-term PICP, PINAW, R [22]
RMSE
Logistics Retailer Univariate 34 Daily ANN Medium-term MSE, R² MATLAB [23]

Manufacturing Manufacturer Univariate + 48 Monthly MLP Medium-term ACC MATLAB [18]

Multivariate
Logistics Retailer Multivariate 1 Daily DT, KNN, Naive Bayes Short-medium- MAPE - [24]
term
Tele- Manufacturer Univariate 36 Quartal GB Medium-term MAD, MAPE, - [25]
communication MSE, RMSE
Warehouse Distributor Univariate 43 Daily LSTM Short-term MSE Python [9]
Logistics Retailer Multivariate 30 Daily and ANN, LSTM + RF, Short- MAE, MSE, Python, R [6]
weekly LSTM, RF medium-term RMAE, RME,
RMSE
Manufacturing Retailer Univariate 36 Monthly LSTM, BLSTM Medium-term MAE, MASE, MATLAB [10]
RMSE, sMAPE
Service Provider Retailer Univariate 48 Daily ARIMA + Theta + Feed Short- medium- MAE, MAPE, Azure [8]
Forward MLP, term MASE
K-means
Store Retailer Multivariate 36 Weekly ANN Long-term MSE MATLAB, [26]
R
Logistics Retailer Multivariate - - ANN, NRS + SVM Short-term MAPE MATLAB [27]
Store Retailer Univariate 30 Daily DNN, GB Short-term MAE, RMSE Python [28]
Store Retailer Univariate 48 Monthly LSTM Medium-term RMSE Python [29]

- Retailer Univariate 72 Monthly LSTM, MLP, RF, Medium-term RMSE Python [30]
XGBoost
Electronics Distributor Univariate 26 Weekly ARIMA+RNN Medium-term MAE, MASE, R [31]
RMSE
Logistics Retailer Univariate 2 Daily ANN, DT, KNN, Part Short-term ACC, PR, RE WEKA [32]
classifier, RF, SVM
Logistics Retailer Multivariate 3 Daily DT, GB, LR, RF Short-term r, RMSE Python [33]

Defence Manufacturer Multivariate - - DT, MLP, RF, SVM, - ACC, PR, RE Python [34]
XGBoost

a
AdaBoost: Adaptive Boosting, ANN: Artificial Neural Network, ARIMA: Autoregressive Integrated Moving Average, ARIMAX: ARIMA model with an
exogenous variable, BLSTM: Long Short-Term Memory (Forward/Backward), DNN: Deep Neural Network, DT: Decision Tree, GB: Gradient Boosting,
LR: Logistic Regression, LSTM: Long Short-Term Memory (Forward), MLP: Multi-Layer Perceptron, NRS: Neighborhood Rough Set, QRF: Quantile
Regression Forest, RF: Random Forest, RNN: Recurrent Neural Network, SVM: Support Vector Machine, SVR: Support Vector Regression, XGBoost:
Extreme Gradient Boosting
b
ACC: Accuracy, AE: Absolute Error, FA: Forecast Accuracy, MAD: Mean Absolute Deviation, MAPE: Mean Absolute Percent Error, MASE: Mean
Absolute Scaled Error, MAE: Mean Absolute Error, MSE: Mean Square Error, NMAE: Normalized Mean Absolute Error, PICP: Prediction Interval coverage
probability, PINAW: Prediction interval normalized average width, r: Correlation coefficient , PR: Precision, RE: Recall, RMAE: Relative mean absolute
error, RME: Relative mean error, RMSE: Relative mean square errors, R²: Coefficient of determination, sMAPE: symmetric mean absolute percentage error,
TE: Total Error

This is a resupply of March 2023 as the template used in the publication of the original article contained errors. The content of the article has remained unaffected.
 Mario Angos Mediavilla et al. / Procedia CIRP 107 (2022) 1126–1131 1129

Table 2. Classification of the best performing AI methods rated by the authors in their AI areas.

Machine learning methods

Traditional machine learning methods Deep learning methods Ensemble methods
Unsupervised Learning Supervised Learning
Unsupervised
Dimension Supervised Learning Boosting Bagging
Clustering Classification Regression Learning
Reduction
K-means NRS [27] DT [4] LR [20] - ANN [21,23,26] AdaBoost [5] RF [6,22]
[8,22] Naive Bayes [24] QRF [22] LSTM [6,9,29] GB [25,28]
SVM [27] RF [33] BLSTM [10]; RNN [31] XGBoost [34]
Part classifier [32] MLP [8,18,19,30]

4. Analysis of AI methods used for demand forecasting 4.2 Classification of AI methods by data characteristics
in SCM
To support manufacturing companies in selecting suitable
This section classifies the AI methods used by means of a AI methods, this section offers a classification of AI methods
taxonomy according to their learning approaches for further according to their characteristics. Punia et al. [6] propose the
analysis. Then, a classification of AI methods based on their idea of classifying demand forecasting methods according to
data characteristics is proposed. the dimensionality of data and volume of data used to train
and test the algorithms. This idea represents an initial
approach in identifying and suggesting a suitable method, but
4.1 Classification of the best performing AI methods only offers a general guideline for method selection. Based
on this understanding, Table 3 provides a more detailed and
In the previous section, Table 1 indicates all the AI complete classification of the 23 AI methods used in current
methods used and compared in each publication. For a more literature. The parameters to be considered are the
precise and detailed analysis, Table 2 shows only the method, dimensionality of the data, the volume of data, and the time
or combination of them, that provides the best results in each horizon over which the methods offer the best results.
publication. Therefore, Table 2 offers a classification of the According to the dimensionality, three classes are defined.
23 "winners" AI methods according to a taxonomy of AI Univariate dimensionality corresponds only to historical
disciplines. Due to the diversity of AI methods, these are demand data. Multivariate dimensionality is divided into AI
classified to provide a more clear overview. The authors aim methods using ten attributes or less and those using more
to provide transparency regarding the classification of the AI than ten attributes. The volume of the data represents the
methods used based on their learning approaches. In order to range of the data as a function of the frequency of the data.
evaluate the trend between traditional ML methods and deep Some publications report the volume of data for a set of
learning methods, the authors employ the taxonomy products. In order to make a fair classification, only the
proposed by Liu et al. [13]. Additionally, ensemble methods volume of data for a single product is analyzed. Four
are also considered. categories are defined according to the volume of data. The
In the field of supervised and deep learning, 12 low category includes methods that process up to 500 input
publications are observed. The most widely used methods in data. This is followed by the medium category with more
this field are Multi-Layer Perceptron MLP, Long Short-Term than 501 up to 1000 input data. In addition, the high volume
Memory LSTM, and Artificial Neural Network ANN. represents methods with more than 1001 data. Table 3 shows
Concerning traditional ML methods, higher use of a field with the designation "non-defined". This is because
supervised learning methods is noticed. Seven methods use not all publications provide a detailed statement of the data
classification and regression algorithms. Regarding used. Finally, the proposed classification considers the time
unsupervised learning, three publications previously applied horizon over which the methods perform well. This is
dimension reduction and clustering of their data. These indicated in brackets next to each AI method in Table 3. For
methods are not used to directly forecast demand. They are more information about the definition of the time horizon see
applied to the data to improve demand forecasting. section 3.1.
Dimension reduction is used to eliminate unnecessary
dimensions and thus improved the accuracy of the forecast 4.3 Analysis of AI methods
[27]. Clustering is used to group products with similar
properties to forecast the demand of an entire cluster [8,22]. The success of SCM depends on the effective exchange of
In addition, ensemble methods are used in six publications. information between members in the SC. When it comes to
The general purpose of an ensemble method in ML is to demand forecasting, upstream members are highly
combine the predictions of several regression or dependent in terms of information provided by the
classification methods to achieve a better prediction. downstream members.

This is a resupply of March 2023 as the template used in the publication of the original article contained errors. The content of the article has remained unaffected.
 1130 Mario Angos Mediavilla et al. / Procedia CIRP 107 (2022) 1126–1131

Table 3. Classification of the best performing AI methods rated by the authors by data characteristics.
Dimension
Univariate Multivariate (Attributes: 2-10) Multivariate (Attributes: 11-n)
Volume
Non-defined - [34] XGBoost (-) [27] NRS+SVM (Short-term)

Low [32] Part classifier (Short-term) [33] RF (Short-term) [6] LSTM+RF (Short-term)
(1-500) [25] GB (Medium-term) [4] DT (Short-term) [18] MLP (Medium-term)
[30] MLP (Medium-term) [24] Naïve Bayes (Short-medium-term) [26] ANN (Long-term)
[29] LSTM (Medium-term) [19] MLP (Medium-term)
[10] BLSTM (Medium-term) [21] ARIMAX*+ANN (Medium-term)
[20] ARIMA*+LR (Medium-term) [22] K-means+QRF (Medium-term)
[31] ARIMA*+RNN (Medium-term) +RF
Medium
[28] GB (Short-term) [5] AdaBoost (Medium-term) [6] LSTM+RF (Medium-term)
(501-1000)
High [9] LSTM (Short-term) - -
(From 1001) [8] ARIMA*+MLP (Short-medium-term)
+Theta*
[23] ANN (Medium-term)
AI Methods: AdaBoost: Adaptive Boosting, ANN: Artificial Neural Network, ARIMA: Autoregressive Integrated Moving Average, ARIMAX: ARIMA
model with an exogenous variable, BLSTM: Long Short-Term Memory (Forward/Backward), DT: Decision Tree, GB: Gradient Boosting, LR: Logistic
Regression, LSTM: Long Short-Term Memory (Forward), MLP: Multi-Layer Perceptron, NRS: Neighborhood Rough Set, QRF: Quantile Regression
Forest, RF: Random Forest, RNN: Recurrent Neural Network, SVM: Support Vector Machine, XGBoost: Extreme Gradient Boosting.
* Statistical method used for hybrid

Hence, it is important to choose an appropriate AI method to technological requirements in terms of capacity, storage,
avoid demand distortion. The demand forecasting process is processing and data analysis techniques [35].
additionally influenced by several attributes such as location, Finally, this analysis points out that most of the examined AI
traffic. Thus, the dimension of data is important not only for methods forecast the demand in the short- and medium-term
the representation of real scenarios but also for the selection with a good performance.
of a suitable AI method. This research shows the clear and
successful trend of applying AI methods with multivariate 5. Conclusion
datasets to improve forecasting accuracy [6,22–25,29,33,34].
Almost 50 percent of the analyzed publications use a Accurate demand forecasting enables manufacturing
multivariate data input. In comparison, the study by companies to increase overall supply chain resilience. AI
Gonçalves et al. [19] analyzed 15 publications between 2004 methods alone or in combination with statistical methods
and 2016. 90 percent of these publications worked with significantly improve the accuracy of demand forecasting
univariate data. Especially deep learning methods perform methods. Additionally, in order to avoid demand distortion -
well with unstructured and high-volume data [35]. Table 3 bullwhip effect - and thus ensure the success of the supply
confirms that most AI methods that use multivariate data are chain, transparent communication between the members
based on deep learning techniques. involved in the demand forecasting process is essential.
The volume of data plays an important role in the selection However, this publication shows that most of the literature
of an AI method. This paper points out that currently, most reviewed concentrates only on demand forecasting from the
AI methods work with a low volume of data. This is due to retailer's perspective. This shows the absence of
the difficulty in accessing real data and technical limitations "collaborative forecasting" which is indispensable for
in terms of software and hardware. Additionally, the few upstream members of the supply chain.
methods that use a high volume of data belong to the deep This publication analyses 23 different methods successfully
learning area. applied in demand forecasting between the years 2017 and
Furthermore, the use of hybrid methods (marked with an 2021. This analysis shows the clear trend of using deep
asterisk in Table 3) significantly increases the demand learning techniques. The methods most used are Multi-Layer
forecasting process. Although statistical methods are Perceptron MLP, Long Short-Term Memory LSTM, and
generally used with linear data, they perform well in Artificial Neural Network ANN, all of them corresponding
combination with AI methods. This means that hybrid to the deep learning area.
methods allow modeling demand data with linear and non- With respect to the data, techniques such as clustering and
linear behavior. Such data behavior is characteristic of dimension reduction are used to improve data quality and
customer demand in complex supply chains. thus demand forecasting. Furthermore, the consideration of
Table 3 indicates the total absence of methods in the field of additional attributes such as weather, location, and events, in
big data, i.e., data characterized by multivariate dimension most cases, improves the accuracy of demand forecasting.
and high volume of data. This is because the handling of Big This publication also shows a clear gap in the existence of AI
Data requires unique architectures that exceed the typical methods that perform in the field of big data, i.e., data with

This is a resupply of March 2023 as the template used in the publication of the original article contained errors. The content of the article has remained unaffected.
 Mario Angos Mediavilla et al. / Procedia CIRP 107 (2022) 1126–1131 1131

several variables and a high volume of data. Hybrid methods [18] Amirkolaii KN, Baboli A, Shahzad MK, Tonadre R. Demand
represent hereby a good candidate to deal with these issues. Forecasting for Irregular Demands in Business Aircraft Spare Parts
Supply Chains by using Artificial Intelligence (AI). 2017;50(1):15221–
Finally, this publication classifies the AI methods analyzed 6. https://doi.org/10.1016/j.ifacol.2017.08.2371.
throughout this literature review by their characteristics such [19] Gonçalves JN, Cortez P, Carvalho MS, Frazão NM. A multivariate
as dimensionality of data, volume of data, and time horizon approach for multi-step demand forecasting in assembly industries:
of the forecast. This classification supports manufacturing Empirical evidence from an automotive supply chain. Decision Support
companies in the process of selecting an appropriate AI Systems 2021;142:113452. https://doi.org/10.1016/j.dss.2020.113452.
[20] Francis H, Kusiak A. Prediction of Engine Demand with a Data-driven
method for forecasting customer demand. Approach. Procedia Computer Science 2017;103:28–35.
https://doi.org/10.1016/j.procs.2017.01.005.
References [21] Feizabadi J. Machine learning demand forecasting and supply chain
performance.2020:1–24.
[1] Bauernhansl T (ed.). Management in der Produktion. Berlin, Heidelberg: https://doi.org/10.1080/13675567.2020.1803246.
Springer Vieweg; 2020. References [22] van Steenbergen RM, Mes M. Forecasting demand profiles of new
[2] Kummer S. Grundzüge der Beschaffung, Produktion und Logistik. 4th products. Decision Support Systems 2020;139:113401.
ed. München: Pearson Studium; 2018. https://doi.org/10.1016/j.dss.2020.113401.
[3] Chopra S, Meindl P. Supply Chain Management: Strategy, Planning, and [23] Pereira MM, Frazzon EM. Towards a Predictive Approach for Omni-
Operation; 2016. channel Retailing Supply Chains. IFAC-PapersOnLine
[4] Alsanad A. Hoeffding Tree Method with Feature Selection for 2019;52(13):844–50. https://doi.org/10.1016/j.ifacol.2019.11.235.
Forecasting Daily Demand Orders. IEEE; 2020, p. 223–227. [24] Arif MAI, Sany SI, Nahin FI, Rabby ASA. Comparison Study: Product
[5] Bottani E, Mordonini M, Franchi B, Pellegrino M. Demand Forecasting Demand Forecasting with Machine Learning for Shop. IEEE; 2019, p.
for an Automotive Company with Neural Network and Ensemble 171–176.
Classifiers Approaches. Springer International Publishing; 2021, p. 134– [25] Wisesa O, Adriansyah A, Khalaf OI. Prediction Analysis Sales for
142. Corporate Services Telecommunications Company using Gradient
[6] Punia S, Nikolopoulos K, Singh SP, Madaan JK, Litsiou K. Deep Boost Algorithm. IEEE; 2020, p. 101–106.
learning with long short-term memory networks and random forests for [26] Chawla A, Singh A, Lamba A, Gangwani N, Soni U. Demand
demand forecasting in multi-channel retail 2020;58(16):4964–79. Forecasting Using Artificial Neural Networks—A Case Study of
https://doi.org/10.1080/00207543.2020.1735666. American Retail Corporation. In: Malik H, Srivastava S, Sood YR,
[7] Carbonneau R, Laframboise K, Vahidov R. Application of machine Ahmad A, editors. Applications of Artificial Intelligence Techniques in
learning techniques for supply chain demand forecasting. European Engineering. Singapore: Springer Singapore; 2019, p. 79–89.
Journal of Operational Research 2008;184(3):1140–54. [27] Xue H, Jiang C, Cai B, Yuan Y. Research on demand forecasting of
https://doi.org/10.1016/j.ejor.2006.12.004. retail supply chain emergency logistics based on NRS-GA-SVM. In:
[8] Benhamida FZ, Kaddouri O, Ouhrouche T, Benaichouche M, Casado- 2018 Chinese Control And Decision Conference (CCDC). IEEE; 2018,
Mansilla D, Lopez-de-Ipina D. Stock&Buy: A New Demand p. 3647–3652.
Forecasting Tool For Inventory Control. In: 2020 5th International [28] Wanchoo K. Retail Demand Forecasting: a Comparison between Deep
Conference on Smart and Sustainable Technologies (SpliTech). IEEE; Neural Network and Gradient Boosting Method for Univariate Time
2020, p. 1–6. Series. IEEE; 2019, p. 1–5.
[9] Hodzic K, Hasic H, Cogo E, Juric Z. Warehouse Demand Forecasting [29] Jain A, Karthikeyan V, B S, BR S, K S, S B. Demand Forecasting for
based on Long Short-Term Memory neural networks. In: 2019 XXVII E-Commerce Platforms. IEEE; 2020, p. 1–4.
International Conference on Information, Communication and [30] Moroff NU, Kurt E, Kamphues J. Machine Learning and Statistics: A
Automation Technologies (ICAT). IEEE; 2019, p. 1–6. Study for assessing innovative Demand Forecasting Models.
[10] Pacella M, Papadia G. Evaluation of deep learning with long short-term 2021;180:40–9. https://doi.org/10.1016/j.procs.2021.01.127.
memory networks for time series forecasting in supply chain [31] Fu W, Chien C-F. UNISON data-driven intermittent demand forecast
management. Procedia CIRP 2021;99:604–9. framework to empower supply chain resilience and an empirical study
https://doi.org/10.1016/j.procir.2021.03.081. in electronics distribution. Computers & Industrial Engineering
[11] Shen B, Chan H-L. Forecast Information Sharing for Managing Supply 2019;135:940–9. https://doi.org/10.1016/j.cie.2019.07.002.
Chains in the Big Data Era: Recent Development and Future Research. [32] Qasem S. A PART-Based Algorithm for Prediction of Daily Demand
Asia Pac. J. Oper. Res. 2017;34(01):1740001. Orders. International Journal of Computer Network and Information
https://doi.org/10.1142/S0217595917400012. Security 2019;19:91–5.
[12] Moroff NU, Sardesai S. Machine learning in demand planning cross- [33] Ma Y, Zhang Z, Ihler A, Pan B. Estimating Warehouse Rental Price
industry overview. epubli GmbH; 2019. using Machine Learning Techniques. INT J COMPUT COMMUN
[13] Liu H, Lang B. Machine Learning and Deep Learning Methods for 2018;13(2):235–50. https://doi.org/10.15837/ijccc.2018.2.3034.
Intrusion Detection Systems: A Survey. Applied Sciences [34] Pawar N, Tiple B. Analysis on Machine Learning Algorithms and
2019;9(20):4396. https://doi.org/10.3390/app9204396. Neural Networks for Demand Forecasting of Anti-Aircraft Missile Spare
[14] Zhu X, Zhang G, Sun B. A comprehensive literature review of the Parts. In: 2019 International Conference on Communication and
demand forecasting methods of emergency resources from the Electronics Systems (ICCES). IEEE; 2019, p. 854–859.
perspective of artificial intelligence. Nat Hazards 2019;97(1):65–82. [35] Najafabadi MM, Villanustre F, Khoshgoftaar TM, Seliya N, Wald R,
https://doi.org/10.1007/s11069-019-03626-z. Muharemagic E. Deep learning applications and challenges in big data
[15] Abdullah Ramdhani, Muhammad Ali Ramdhani, Abdusy Syakur analytics. J Big Data 2015;2(1). https://doi.org/10.1186/s40537-014-
Amin. Writing a Literature Review Research Paper: A step-by-step 0007-7.
approach 2014.
[16] Snyder H. Literature review as a research methodology: An overview
and guidelines. Journal of Business Research 2019;104:333–9.
https://doi.org/10.1016/j.jbusres.2019.07.039.
[17] Patel AB, Desai TN. A systematic review and meta-analysis of recent
developments in sustainable supply chain management.
2019;22(4):349–70. https://doi.org/10.1080/13675567.2018.1534946.

This is a resupply of March 2023 as the template used in the publication of the original article contained errors. The content of the article has remained unaffected.