Machine Learning and Deep Learning Models for Demand Forecasting in Supply Chain Management: A Critical Review

Abstract

This paper presents a comprehensive review of machine learning (ML) and deep learning (DL) models used for demand forecasting in supply chain management. By analyzing 119 papers from the Scopus database covering the period from 2015 to 2024, this study provides both macro- and micro-level insights into the effectiveness of AI-based methodologies. The macro-level analysis illustrates the overall trajectory and trends in ML and DL applications, while the micro-level analysis explores the specific distinctions and advantages of these models. This review aims to serve as a valuable resource for improving demand forecasting in supply chain management using ML and DL techniques.

1. Introduction

In an era of widespread global supply chain disruptions and acute volatility, demand forecasting has become more crucial than ever for any competitive business seeking to gain a firm foothold in the market [1] (Ivanov, 2022). This is primarily because traditional methods for forecasting demand are not sufficiently effective in capturing drastically shifting consumer patterns [2]. Artificial intelligence (AI) provides a revolutionary alternative that offers the most accurate and agile forecasting models [2]. Compared to reliance on conventional methods, AI-based forecasting is far more accurate, as it transcends the limitations of even the most sophisticated traditional models [3]. Importantly, AI-based forecasting strategies are dynamic, with models that continuously adjust to market fluctuations, making them an ideal tool for ensuring the alignment of strategies with actual consumer behavior [4,5]. Thus, AI not only enhances demand prediction but also introduces a new paradigm for developing forecasting strategies [6].

This paper aims to advance existing research by addressing the following key questions: How does AI-driven automation enhance demand forecasting efficiency in supply chains? Which AI projects apply to demand forecasting in supply chains? What are the future research gaps, and how should these be addressed? This paper provides a comprehensive review of the literature on AI-driven demand forecasting in supply chains. The literature review for this study is thorough, integrating the foundations of AI used in the demand forecasting supply chain (DFSC) framework to propose an analytical framework that provides a comprehensive perspective on the AI-DFSC challenges. The paper utilizes both macroscopic visual analysis and micro-level content analysis for clarity. At a macro-level, it presents descriptive visual research findings related to topics, thematic progress, and thematic evolution in the general literature. At a micro-level, this paper focuses on conducting a detailed content analysis of the literature. Equally important, this study identifies promising future research trends and addresses the gaps that need attention in future studies. The remainder of this paper is structured as follows: the background of the research on AI-DFSC is presented in this section to establish the importance of this study. The data sources and the methods used in data processing are described in detail in Section 3. Section 4 presents a macroscopic structural analysis of AI-DFSC using visual analytics. Section 5 provides a thematic discussion of AI-DFSC in detail.

2. Theoretical Background

Demand forecasting is an essential aspect of supply chain management because it helps businesses anticipate and meet consumer demands, especially in retail and the food industry. It enables businesses to optimize inventory levels, pricing, and promotions. Traditional regression models are commonly used for demand forecasting, but they often provide unstable estimations [7]. However, recent studies have shown that machine learning techniques, such as long short-term memory (LSTM) networks, can improve the accuracy of demand forecasting by learning non-linear relationships and time-series-specific information [8]. Despite the limited empirical data available, some mathematical models, like seasonal ARIMA, ARIMA with Fourier terms, ETS, and TBATS, have proven to be efficient in forecasting demand [9].

Additionally, demand forecasting in the automotive sector has greatly benefited from artificial intelligence algorithms, such as random forest, multiple linear regression, k-nearest neighbors, extreme gradient boosting, and support vector machines [10]. Notably, another study proposed a SARIMA–LSTM–BP model combination to predict new energy vehicle demand, demonstrating better predictions than traditional econometric and deep learning models [11]. Another article developed an improved model using deep learning methods aimed at enhancing supply chain management, thereby increasing accuracy in product delivery [12].

In the context of blood supply chain management, a smart-platform-based approach was proposed using machine learning and time-series forecasting models to reduce uncertainty around blood requirements, balancing collection and distribution [13]. The effectiveness of different machine learning methods, such as random forest, gradient boosting, and XGBoost, was analyzed to forecast demand, with XGBoost proving to be the most accurate in predictions [14]. Lastly, a study explored how artificial intelligence (AI) and machine learning (ML) can be integrated to manage and respond to energy demand in real time, emphasizing their potential to improve energy usage and balance power supply across different consumers [15].

3. Materials and Methods

The search methodology involved a detailed examination of the pertinent literature, incorporating both broad and detailed perspectives. Broad analysis was conducted visually, whereas detailed analysis concentrated on the document content, employing an in-depth review of general documents to confirm the accuracy of key documents [16] (see Figure 1). To be more specific, this procedure was divided into two different phases: the first aimed to acquire a huge number of exceptional documents on the target research topics while exploring the literature, and the second focused on these documents’ selection based not only on the target research topics but also on the significance of the sources.

3.1. Data Preparation

The quality of data decides on the depth and on the precision of the results. Thus, the problem of choosing a suitable database for the needs of this particular study is ongoing [17]. In this particular research, the main source of data was SCOPUS. The claim is fully justified in the literature. The currently analyzed database can be seen as a good choice to perform a vast analysis. This is because SCOPUS is among the biggest bases of literature references [18]. SCOPUS has been used in many different studies, which indicates that the quality of data is good [19,20]. The following search tags for the database were included: the level of digital advancement of business, articles available in the index, authors, issues, document types, areas of science, and key words. Employing descriptive and analytical methods, the study was conducted. It is an inductive–deductive study with many tags ignored in order to achieve complete outcomes. The focus was solely on the goals of specific documents selected based on the threshold due to the goal of the study. According to the set parameters, a SCOPUS search was performed to find useful information. The data selected for analysis were significant publications from 2015 to 2024.

Table 1. The search strategy.

Category	Keywords
Demand forecasting	(“demand forecasting” OR “demand prediction”) AND
Artificial intelligence	(“artificial intelligence” OR “machine learning” OR “deep learning”) AND
Supply chain management	(“supply chain management “ OR “supply chain “)

describes the search methodology employed for conducting a literature review. The present search created a corpus of document sets of papers compatible with the search strategy’s criteria of the document’s titles, abstract, or list of keywords. The existing research on the application of artificial intelligence and machine learning specifically within supply chain management for forecasting demand is vast. The application has increased since 2010, resulting in several successful examples and controversies for future employment and business management [21,22]. Nowadays, various corporations employing AI invest heavily in this technology to optimize their supply chain’s full operation scope [23], further leading to redirecting academic attention to update their core research to modern AI. On such a note, the exploration of academic directions with focus on AI for supply chain demand forecasting commences and becomes regular. Applying this selection, the literature review was limited to 2014–2024. Under these conditions and criteria, the research garnered in this project reached a total of 119 articles within recent research until 13 August 2024.

3.2. Data Filtering

The evaluation of potential research literature is based on a structured four-step process:

• Stage 1: Removal of duplicate literature;
• Stage 2: Selection of articles that are relevant to the research topic;
• Stage 3: Specific identification of articles that are most relevant to the research topic;
• Stage 4: Identification of articles that are important.

In this study’s literature review process, a four-stage selection process was used. With each stage, the precision was further refined, and consequently, a total of 119 papers relevant to this study were reviewed and evaluated. While a large number of papers were found to be accessible, many were extensive, and the proposal in the literature is to read all of them with care and accuracy. Therefore, the literature overall was utilized with maximum care and was expansive across this study. It can be considered that some recently published papers of higher quality have shown high preference and reference. See Figure 2 and

Table 2. The number of paper types.

Type of Paper	Numbers
Article	45
Conference paper	60
Book chapter	11
Conference review	3

for a summary of the categories of the selected papers.

3.3. Visual Representation Techniques

Utilizing visualization methods in the literature review is important because it allows researchers to gain a deeper, more comprehensive understanding of the literature that must be examined. Knowledge mapping is a novel approach developed within the fields of scientometrics and information metrology. This method illustrates the emergence and evolution of knowledge and reveals the structure of the knowledge landscape. CiteSpace version (6.2.6) is a prominent tool for visualizing citation networks. It is used to generate keyword atlases and time zone views of co-citation maps and to cluster citations, identifying nodes and their interconnections. Bibliometrix R Package (R 4.3.2) and VOSviewer software (1.6.20) are reliable tools commonly used to create co-occurrence maps for authors, citations, and keywords. These maps spatially show how nodes cluster together and are connected based on their proximity and density, clearly demonstrating the connections between authors and institutions. Additionally, CiteSpace allows for contrasting and comparing the strengths and weaknesses of relationships among different themes using trees and lines, enabling an extensive analysis of theme relations. Meanwhile, VOSviewer is particularly helpful in understanding connections within each cluster, providing complementary information. Furthermore, the Biblioshiny interface utilizes trend analysis, allowing researchers to track research trends over time and examine how interest in particular areas or topic keywords evolves. Similarly, virtual online platforms have been used to advance investigations by analyzing more complex data handling methods, making the analysis more in-depth than ever before.

4. Findings Results

The data used in this study were up to date as of October 2023. The first phase of the subsequent steps involved an initial descriptive analysis to gain a better understanding of the reviewed articles. This analysis examined the papers across several dimensions to uncover various aspects of the literature. Key components of this examination included the words used in titles, key terms in abstracts, years of publication, citation counts, bursts of interest for specific topics, centrality measures indicating the importance of papers within a citation network, and sigma values representing the overall impact of the publications. Additionally, specific terminology and thematic clusters were identified through this analysis. Each paper’s disciplinary field was assessed, and distinctions were made between article types (e.g., research articles versus reviews), as well as tracking their distribution across different academic journals. This approach provided the researchers with a multifaceted understanding of the literature, revealing important patterns, themes, and focal points within the discipline.

4.1. Descriptive Findings

4.1.1. Keyword, Title, and Abstract of Publication

This research focused on “supply chain management” and “demand forecasting”, using SCOPUS as the primary source database; consequently, 119 representative articles were selected. Using VOSviewer, a bibliometric analysis tool, keywords that appeared at least ten times were identified, and a keyword co-occurrence network diagram was constructed. The network diagram highlights frequently mentioned phrases, such as “supply chain management”, “machine learning”, “forecasting”, “artificial intelligence”, and “deep learning”. These phrases summarize the content of the articles and illustrate the composition of research within this field.

The prevalence of these keywords see Figure 3 suggests that there are numerous demand forecasting methods based on machine learning (ML) and deep learning (DL) solutions for supply chain management, as frequently discussed in the literature. This study highlights a significant emphasis on machine learning (ML) and deep learning (DL) tools to enhance the accuracy and efficiency of demand forecasting. This trend reflects a broader effort by researchers and industry professionals to integrate cutting-edge AI systems for improved supply chain management. The repeated appearance of these keywords indicates their growing importance in addressing complex prediction problems.

A closer examination of these keywords provides valuable insights into the key research issues and themes in the field. The analysis underscores the importance of advanced AI methods in contemporary supply chain demand forecasting, demonstrating the increasing adoption of such technologies for enhanced performance. This comprehensive investigation lays the foundation for further exploration of the use and impact of machine learning (ML) and deep learning (DL) technologies in supply chain management, emphasizing their potential to transform demand forecasting practices.

4.1.2. The Total Volume of Publications over Time

In the broader context, it is important to monitor the progress in the overall number of publications over time. From the analysis of Figure 4, we can identify several phases in the development of research output. In the emerging phase (2014–2017), very few articles were published each year—specifically, not more than 1 to 3 publications per year. The year 2018 saw no publications at all, indicating an early stage of limited interest or exploration in this field. This phase suggests either a lack of interest or merely tentative engagement with the topic.

Next, there was a transitional phase (2018–2020). In 2019, a noticeable increase occurred, with eight articles being published. During this period, the volume of publications remained small but consistent, with seven to eight articles published each year in 2019 and 2020. This phase marks the beginning of growing interest in the research area, as the community started to gain momentum.

A remarkable surge in publications is evident during the acceleration phase (2021–2024). The number of published papers increased sharply, rising from 17 in 2021 to 30 in 2024. In both 2023 and 2024, a total of 30 articles were published, indicating high research activity. This sudden rise suggests that the topic is becoming very popular within scientific circles, demonstrating its relevance and significance.

Overall, the data show remarkably rapid progress in this area, especially between 2021 and 2024, when nearly 73% of all 119 publications were produced. This trend indicates the growing importance and interest in the field, which has now become a major focus of academic research. The exponential growth in the volume of publications suggests that the subject has moved beyond its niche status to become mainstream within the scientific community. The sharp increase in publications during the acceleration phase may be due to a combination of factors, such as technological advances, increased funding opportunities, and greater societal recognition of the subject’s relevance.

Based on the information presented in this article, it is evident that, despite being relatively young, this research area has gained significant momentum in recent years. There are still possibilities for further growth as new challenges and opportunities arise. The rapid increase in scientific output suggests that this field will likely continue to evolve, with ongoing contributions expected from an increasingly diverse group of academics and organizations.

4.1.3. The Distribution of Publications by Field, Type, and Journal

The number of publications by area, type, and journal provides important insights into the evolution of this domain over time. According to Figure 5, the most published articles are in the fields of computer science, engineering, decision sciences, business management, and accounting. This suggests that the application of AI in demand forecasting is interdisciplinary, impacting multiple areas. The high volume of publications in computer science and engineering indicates that AI and machine learning models require a strong technical foundation, which is essential for advancements in algorithms, computational techniques, and system designs for effective demand forecasting.

On the other hand, publications in decision sciences, business, and management focus on the practical applications of these technologies to enhance decision-making processes, optimize operations, and improve overall supply chain efficiency. This interdisciplinary approach ensures that artificial intelligence solutions are not only technically robust but also aligned with business goals and management practices.

Moreover, demand prediction challenges have been addressed using various methods, such as machine learning, deep learning, blockchain, and time series forecasting, in these studies. Machine learning and deep learning are particularly noteworthy because they can process large amounts of data and recognize complex patterns, enabling more accurate and dynamic predictions. Although less widely adopted, blockchain technology holds significant potential to improve demand forecasting accuracy by enhancing transparency and traceability in supply chains. Time series forecasting remains a classic methodology that is often integrated into sophisticated AI models to enhance prediction quality.

Overall, the diversity of publications in these areas reflects the complex nature of AI research in anticipating supply chain needs, highlighting the collaborative effort to synchronize technological innovations with practical applications. This trend indicates an ongoing and extensive effort to leverage AI for solving complex prediction problems, ultimately making supply chains more resilient and efficient.

4.1.4. The Different Methodological Approaches and Article Proportion

Annotations regarding the titles and abstracts of the 119 reviewed publications will guide the reader toward identifying the methodological focus (see Figure 6). Among these articles, modeling as a system of symbols used to obtain reproducible results or data falls under various categories: 44 papers focus on modeling, 43 on case studies, 25 on surveys, and 7 on conceptual discussions. For instance, “modeling” accounts for around 37 percent, while “case study” encompasses about 36 percent of the entire literature on demand forecasting. This is significant because it suggests that these approaches are crucial in identifying the most reliable demand estimation techniques based on numerous performance metrics, with a substantial body of literature supporting them.

The prominence of modeling and case studies indicates that researchers are particularly interested in assessing the realism of models by testing their theories against real-world scenarios, thereby enhancing their reliability. As mentioned in the previous section, as of October 2023, the data utilized in this study from various articles constitutes 21%. Surveys are among these documents, providing a general overview of many modern trends, challenges, and improvements related to AI-driven demand prediction, thus offering informative perspectives on how users understand and apply these technologies.

Although limited in number, conceptual discussions play an important role in identifying new research ideas, frameworks, and theoretical foundations for future studies. This methodological diversity reaffirms that researchers adopt a holistic approach to tackling the issue of demand forecasting in supply chain management, combining theoretical exposition, empirical assessment, and comprehensive examination to advance this field.

4.1.5. The Distribution of Publications among Countries, and Collaborative Efforts between Them

As depicted in Figure 7, the distribution of publications by the countries of the authors and their contributions to the field provides insightful perspectives on the global nature of this research area and its interconnections with other fields. The collaboration network covers a wide range of countries from various continents, such as Asia, Europe, North America, and Africa. This broad range indicates diverse interests in the research topic, originating not only from developed countries but also from underdeveloped ones. The involvement of countries like China, the USA, Germany, and India suggests that the scope of the research area is not confined to a single region but represents a cumulative scientific endeavor of worldwide concern.

The figure shows that certain countries have formed multiple collaborations, indicating strong academic links and shared research agendas. For example, the USA collaborates with countries like Bangladesh, while Japan has agreements with both China and Indonesia, demonstrating a leading role in research within this field. Collaborations are not limited to neighboring nations but also include those geographically distant, such as Germany–South Africa and India–Morocco, effectively erasing socio-economic and technological barriers to knowledge sharing.

Moreover, partnerships involve countries that were not previously considered significant players in international scientific research, such as Bangladesh, Iraq, and Jordan, highlighting an increasing interest in these regions. This growing recognition indicates a trend toward mainstreaming research efforts across different regions, informed by diverse global developments. The relatively high frequency of collaborations in countries like China and the USA has established these nations as hubs for multinational networks, facilitating academic endeavors due to their enhanced resources and infrastructure.

Global collaborations showcase a diverse and vibrant research sector, fostering creativity and encouraging growth. The integration of previously separate research avenues could lead to groundbreaking breakthroughs. The international scope of this work suggests that the findings are relevant globally and continuously.

In conclusion, the collaboration data reveal a sprawling and well-integrated international network in this research area, with numerous countries actively involved, underscoring its global significance. The findings demonstrate that cooperation among research teams from various regions is essential for collective progress.

4.1.6. The Analysis of Scientific Output by Country

In this context, the distribution of scientific works across numerous nations, as shown in Figure 8, provides a clear picture of the global research landscape. India stands out as the leading contributor, with 35 publications. This demonstrates not only a strong presence in academic circles but also indicates that significant importance is placed on this research area in India. This prominence can be attributed to substantial funding driven by evolving academic and industrial interests in the country.

China’s holistic policy of promoting innovative research and development in modern technologies and Morocco’s focus on national interests, such as Technological Industrial Revolution 4.0 and logistics, support their roles in these fields, with each contributing 12 publications. Although the number of contributors from these regions is relatively small, their focus aligns with strategic national priorities. Germany and the United States are also key players in the global research ecosystem, contributing 10 and 9 publications, respectively, highlighting their strong research foundation compared to their partners known for major technological innovations.

France, Bangladesh, and the United Kingdom follow, with 7 and 4 publications, respectively. While France and the UK have traditionally been strong in scientific research, Bangladesh reflects academic growth driven by collaborations with more established research nations. Several other countries, including Austria, Iran, Australia, Brazil, Canada, Italy, Japan, and Saudi Arabia, have contributed one or two publications, indicating emerging research communities or specialized interest areas, suggesting the presence of extensive research networks.

Additionally, countries like Argentina, Bahrain, Belgium, Colombia, Denmark, Ecuador, Finland, Hong Kong, Indonesia, Iraq, Jordan, Malaysia, the Netherlands, New Zealand, Pakistan, Peru, Poland, Portugal, Romania, Rwanda, Slovenia, South Korea, Sri Lanka, Thailand, Ukraine, the United Arab Emirates, and Uzbekistan each contributed at least one publication. This widespread participation shows that there is substantial potential for growth and increasing involvement in these regions as research infrastructure develops and international collaborations expand.

Four documents are categorized as “Undefined”, which may include international collaborations or multidisciplinary studies. This broad distribution of contributions from both developed and developing nations indicates a global research effort with opportunities for increased participation in the future. The presence of countries like India, China, and Morocco suggests that these nations could shape future research trends and innovations, while countries with moderate to emerging contributions could increase their output, leading to a more balanced global research pattern.

The wide spectrum of country participation shows that the research subject is increasingly recognized as important globally. This acknowledgment emphasizes the need for unity in promoting knowledge and innovation through global partnerships, ensuring that significant findings are achieved and substantial developments occur within this discipline.

4.1.7. Insights from the Most Globally Cited Documents

Key insights from the results can be obtained from the analysis of the most globally cited documents on the impact and trends in the field of demand forecasting using AI and ML techniques see Figure 9 and Table 3. Seyedan [24] is the most cited paper, with 117 citations, demonstrating its strong influence and consistent interest within the research community due to a high total citation (TC) per year of 23.40 and a normalized TC of 3.60. Although published more recently, Hu [25] already has 34 citations, with a maximum normalized TC of 17.38, indicating a potentially emerging impactful area. Papers such as those by Aamer [26] and SARDAR [27] show extremely high TC per year rates of 10.25 and 9.50, respectively; therefore, their influence is rapidly growing within the community. Meanwhile, Chien [28] presents a strong citation rate, with a TC per year of 14.20, highlighting its current relevance in discussions.

The heavy citation of papers, such as Hu [29], in a very short period may serve as an indicator of sources that are likely to be important for those interested in fast-moving and emerging topics. The more specialized interest in clean logistics and sustainable supply chains is reflected in Shokouhifar [30], with a normalized TC of 4.64. On the other hand, Nikolopoulos [31] has an average TC per year of 4.11 with 37 citations, suggesting more enduring research that remains relevant for influencing the field over time.

For those aiming to research studies that will have a significant impact, papers with a high normalized TC, such as Seyedan and Hu, should be prioritized, as they are highly likely to introduce new methodologies or insights. For a more historical perspective or foundational knowledge, Nikolopoulos could significantly contribute to understanding long-standing theories or practices. Some recommendations for future research include focusing on high-impact papers like Seyedan and Hu, as they have contributed massively, monitoring emerging trends observed in recent high-citation papers, balancing these with long-term impactful studies such as Nikolopoulos, and covering a diversity of journals to tap into the interdisciplinarity of the research area.

Regarding research goals and methods, papers like Seyedan integrated different ML models, such as random forest, SVM, and neural networks, to enhance the accuracy of demand forecasting using large-scale retail datasets. Chien [28] formed hybrid models that combine regression and neural networks to decrease forecasting errors related to data from industrial production and sales. Aamer et al. explored deep learning models, including CNN and LSTM, in detail for application to optimizing demand forecasting in operations management. Other notable papers include Sardar, Sarkar, and Kim, who worked on process optimization within supply chains using AI-driven models, showing improved efficiency in inventory management. Nikolopoulos, Babai, and Bozos (2016) compared traditional versus ML-based forecasting methods, highlighting the competitiveness of ML models in terms of accuracy and scalability.

Gonçalves [32] employed ensemble learning methods in decision-support systems, demonstrating improved robustness and accuracy in supply chain datasets. Hu presented several state-of-the-art DL models like transformers and demonstrated their applicability to dynamic supply chains, achieving high accuracy for real-time applications despite implementation complexity and computational costs. Leung [33] analyzed data-driven demand forecasting using clustering and decision trees to improve the management of large datasets. Abolghasemi [34] assessed the impact of AI on demand forecasting in production economies and highlighted timely and accurate predictions across various industries. Shokouhifar [29] discussed the sustainability dimensions of demand forecasting using DL models, considering environmental factors to balance economic and environmental practices in green supply chains.

This comprehensive analysis underscores the progress from traditional methods to sophisticated AI/ML techniques and the growing importance of sustainability in supply chain management. Each paper has its strengths but also faces challenges, such as computational costs, complexity, and data requirements, which provide strategic direction for both researchers and practitioners keen to stay at the forefront of developments in demand forecasting.

Table 3. Overview of AI/ML techniques in supply chain demand forecasting.

Author(s) and Year	Research Goals	ML/DL Models Used	Dataset Characteristics	Key Findings	Advantages	Limitations
Seyedan [24]	To integrate demand forecasting with AI and ML for enhanced accuracy in supply chain management	Various ML models like random forest, SVM, and neural networks	Large-scale retail dataset with diverse product categories	AI/ML models significantly outperform traditional forecasting methods	Improved accuracy and adaptability	High computational cost and complexity
Chien [28]	To apply ML models to predictive analytics in supply chain demand forecasting	Hybrid models combining regression and neural networks	Industrial production and sales data	Hybrid models reduce forecasting errors up to 20% compared to standard methods	Effective in capturing complex patterns	Requires extensive data preprocessing
Aamer [26]	To explore the role of AI in optimizing demand forecasting in operations management	Deep learning models (CNN, LSTM)	Supply chain operations data	DL models enhance demand prediction accuracy, especially for volatile demand patterns	Handles large datasets effectively	Long training times and potential overfitting
Sardar [27]	To investigate process optimization in supply chains through AI-driven demand forecasting	Gradient boosting, decision trees	Manufacturing and logistics datasets	AI-driven models increase efficiency in supply chain management by optimizing inventory levels	Reduces human intervention in forecasting	Model interpretability can be challenging
Nikolopou [31]	To evaluate traditional and ML-based demand forecasting methods	Traditional methods vs. ML models (e.g., ARIMA, SVM)	Historical sales and demand data	ML models provide competitive accuracy with better scalability	Balances traditional and modern approaches	ML models require more data and fine-tuning
Gonçalves [32]	To implement ML techniques in decision-support systems for supply chain demand forecasting	Ensemble learning methods (e.g., bagging, boosting)	Diverse supply chain datasets	Ensemble methods outperform single models in demand prediction	Enhances robustness and accuracy	Requires more computational resources
Hu H, 2023 [25]	To explore cutting-edge AI/ML techniques in demand forecasting for dynamic supply chains	Advanced DL models (e.g., transformers)	Real-time and historical data integration	Transformer models capture temporal dependencies better than traditional methods	High accuracy in real-time applications	Complex implementation and high computational cost
Leung [33]	To analyze data-driven demand forecasting using AI in supply chains	Clustering, decision trees	Retail and distribution data	Data-driven approaches improve demand forecasting by identifying underlying patterns	Effective in managing large datasets	May require domain-specific tuning
Abolghase [34]	To assess the impact of AI on demand forecasting in production economies	Neural networks, regression models	Production and demand datasets from various industries	AI models lead to more accurate and timely predictions in production planning	Scalable across industries	Requires continuous model updating
Shokouhif [29]	To incorporate sustainability into demand forecasting using AI	DL models with environmental factors	Supply chain data with sustainability metrics	Sustainable AI models align demand forecasting with green supply chain practices	Balances economic and environmental goals	May face trade-offs between accuracy and sustainability objectives

Table 3. Overview of AI/ML techniques in supply chain demand forecasting.

Author(s) and Year	Research Goals	ML/DL Models Used	Dataset Characteristics	Key Findings	Advantages	Limitations
Seyedan [24]	To integrate demand forecasting with AI and ML for enhanced accuracy in supply chain management	Various ML models like random forest, SVM, and neural networks	Large-scale retail dataset with diverse product categories	AI/ML models significantly outperform traditional forecasting methods	Improved accuracy and adaptability	High computational cost and complexity
Chien [28]	To apply ML models to predictive analytics in supply chain demand forecasting	Hybrid models combining regression and neural networks	Industrial production and sales data	Hybrid models reduce forecasting errors up to 20% compared to standard methods	Effective in capturing complex patterns	Requires extensive data preprocessing
Aamer [26]	To explore the role of AI in optimizing demand forecasting in operations management	Deep learning models (CNN, LSTM)	Supply chain operations data	DL models enhance demand prediction accuracy, especially for volatile demand patterns	Handles large datasets effectively	Long training times and potential overfitting
Sardar [27]	To investigate process optimization in supply chains through AI-driven demand forecasting	Gradient boosting, decision trees	Manufacturing and logistics datasets	AI-driven models increase efficiency in supply chain management by optimizing inventory levels	Reduces human intervention in forecasting	Model interpretability can be challenging
Nikolopou [31]	To evaluate traditional and ML-based demand forecasting methods	Traditional methods vs. ML models (e.g., ARIMA, SVM)	Historical sales and demand data	ML models provide competitive accuracy with better scalability	Balances traditional and modern approaches	ML models require more data and fine-tuning
Gonçalves [32]	To implement ML techniques in decision-support systems for supply chain demand forecasting	Ensemble learning methods (e.g., bagging, boosting)	Diverse supply chain datasets	Ensemble methods outperform single models in demand prediction	Enhances robustness and accuracy	Requires more computational resources
Hu H, 2023 [25]	To explore cutting-edge AI/ML techniques in demand forecasting for dynamic supply chains	Advanced DL models (e.g., transformers)	Real-time and historical data integration	Transformer models capture temporal dependencies better than traditional methods	High accuracy in real-time applications	Complex implementation and high computational cost
Leung [33]	To analyze data-driven demand forecasting using AI in supply chains	Clustering, decision trees	Retail and distribution data	Data-driven approaches improve demand forecasting by identifying underlying patterns	Effective in managing large datasets	May require domain-specific tuning
Abolghase [34]	To assess the impact of AI on demand forecasting in production economies	Neural networks, regression models	Production and demand datasets from various industries	AI models lead to more accurate and timely predictions in production planning	Scalable across industries	Requires continuous model updating
Shokouhif [29]	To incorporate sustainability into demand forecasting using AI	DL models with environmental factors	Supply chain data with sustainability metrics	Sustainable AI models align demand forecasting with green supply chain practices	Balances economic and environmental goals	May face trade-offs between accuracy and sustainability objectives

Figure 9. The most globally cited documents [24,25,26,27,28,29,31,32,33,34].

Figure 9. The most globally cited documents [24,25,26,27,28,29,31,32,33,34].

4.1.8. Keywords and Subject Clusters within the Publications

At a micro-level, one can observe the changes in a subject over time, which reveal temporary landmarks in the development of this discipline and highlight important temporal trends. The keyword co-occurrence map, arranged in a time-series fashion, illustrates which research hotspots were at the forefront during each period. Researchers can trace these hotspots by sequentially examining keyword clusters, from 0 to 7, reflecting increased scholarly attention on forecasting demand using machine learning (ML) technologies (clusters 1 to 5). A cluster analysis revealed a variety of patterns within the field (see

Table 4. Summary of the 5 largest clusters.

#	Size	Silhouette	Label (lsi)	Label (llr)	Label (mi)	Average Year
1	18	0.988	Supply chain management	Exploring application (6.19, 0.05)	Review (0.4)	2019
2	18	0.967	Intelligent system	Vaccine supply chain management (6.19, 0.05)	Using deep learning technique (0.4)	2020
3	15	0.924	Machine learning technique	Machine learning technique (6.62, 0.05)	Learning approach (0.34)	2016
4	14	1	Predictive big data analytics for supply chain demand forecasting: methods, applications, and research opportunities	Supply chain demand forecasting (5.22, 0.05)	Supply chain demand forecasting (0.11)	2020
5	7	1	Modeling wholesale distribution operations: an artificial intelligence framework	Modeling wholesale distribution operations (6.95, 0.01)	Modeling wholesale distribution operations (0.08)	2021

and Figure 10).

Cluster 1 is mainly devoted to supply chain management and review articles, which mostly appeared after 2019, indicating that foundational scholarship and critical evaluations were being enhanced during this period. Cluster 2 focuses on smart systems, covering vaccine supply chain management and the use of deep learning in other domains, with a strong emphasis on developing AI models for practical applications, particularly in response to global issues like vaccine allocation, which became prominent between 2020 and 2022. Cluster 3 explores different machine learning techniques and innovative learning methods, yet it still aligns with the themes discussed in Cluster 2, thereby showing an expanding yet deepening understanding of various ML types during this period.

In addition, Cluster 4 is primarily about forecasting using big data analytics for predicting demand within a supply chain, reaching its peak in 2016. This suggests that big data should be increasingly utilized for making predictions through improved techniques. This cluster also includes findings related to the creation and use of everyday prediction systems based on artificial neural networks for supermarket sales floors around 2012, implying early experimentation with neural networks in real-world forecasting scenarios during that time.

Cluster 5 focuses on modeling wholesale distribution activities using intelligent artificial neural computer systems, with notable developments recorded in 2021. These developments highlight continued advancements in neural computing to improve complex distribution networks. All these clusters consistently show very high silhouette values, demonstrating coherence and high research quality.

This comprehensive analysis of different research patterns, techniques, and time trajectories in thematic areas like supply chain management, intelligent systems, and advanced analytics provides valuable insights; thus, further study into what each distinct cluster uniquely contributes is necessary. The evolution from ML applications in demand forecasting to data-driven methodologies reflects how much this field has developed over time. Future research should build on existing frameworks to enhance methods and broaden their applications for anticipating demand.

An analysis of keyword-rich content and the relationships between various topics generally shows how knowledge production has changed over time, the focus of studies, and the development of the field. By considering these clusters, scholars can identify key topics or new developments that can direct future research while also helping them better understand the historical growth of the discipline. This ongoing journey will continue to generate more robust and adaptable prediction models, contributing to higher efficiency and stability of supply chains worldwide.

4.2. Trajectory and Trends in ML and DL Applications

4.2.1. Thematic Trends and Patterns

Thematic clustering and a trend analysis have revealed several key patterns illustrating how demand forecasting for supply chain management is evolving through artificial intelligence (AI) applications (Figure 11). There is a clear and consistent trend toward the greater adoption of AI technologies in supply chain management, driven by the need for more accurate and agile demand forecasting mechanisms. This shift enables better anticipation of demand changes, stock level optimization, and overall operational performance improvement for firms. Traditional forecasting practices are being reassessed with the integration of AI-driven models, which show significant accuracy advancements, as highlighted by Ivanov [1].

The literature also indicates a noticeable shift towards deep learning models, such as long short-term memory (LSTM) networks, due to their unique ability to manage complex temporal patterns and non-linear relationships embedded within demand data. Advanced forecasting systems increasingly rely on deep learning models like LSTM as a foundational element. For example, according to Phyu and Khine [8], these models are more accurate and adaptable, allowing supply chains to adjust their strategies to real-time market conditions. This adaptability results in reduced forecast errors and improved customer service levels.

Additionally, traditional statistical methods are being combined with advanced AI techniques to create hybrid models, leveraging the strengths of both approaches to develop more robust and accurate forecasting solutions. One example is the SARIMA–LSTM–BP model, which combines the seasonal pattern detection of SARIMA with the temporal pattern learning capabilities of LSTMs and the optimization features of backpropagation. Falatouri et al. [11] demonstrated that these hybrid models significantly outperform conventional standalone AI algorithms, offering greater precision and reliability across various prediction scenarios.

There is also a significant emphasis on implementing AI in decision-support systems in real-time within supply chain management. This approach allows for quick and accurate forecasting, even under changing conditions typical of fast-paced global value chains. Real-time AI-based decision-support systems enhance agile response capabilities, enabling supply chains to adapt quickly to shifts while making intelligent decisions within constrained time frames. Specifically, Al-Musaylh et al. [35] documented how SVM-based real-time forecasting systems have greatly improved decision-making, resulting in more agile and responsive supply chains.

By examining how different keywords interrelate in the field of supply chain management through this thematic map, one can understand the terms clustered according to their frequencies and relevance in the existing literature. “Supply chain management”, “demand forecasting”, “machine learning”, and “big data” are central topics in Cluster 1, reflecting core interests in the discipline. Terms like “machine learning techniques”, “neural networks”, and “long short-term memory” suggest a focus on advanced predictive and analytical methods. Cluster 2 includes keywords such as “blockchain”, “life cycle”, and “behavioral research”, indicating an emerging interest in integrating blockchain technology into supply chain management and addressing global issues like COVID-19 and vaccine distribution.

Cluster 3 covers methodological aspects, such as “design/methodology/approach”, “fuzzy neural networks”, and “outsourcing”, which emphasize the design and optimization of supply chain processes. Cluster 4 highlights the role of AI in optimizing inventory management, focusing on themes like “inventory control”, “decision-making”, and “artificial intelligence”. Terms such as “predictive analytics” and “cost reduction” reflect the ongoing need for efficiency and sustainability.

By analyzing these clusters in detail, we gain insights into the evolving priorities and areas that require critical attention, particularly the interrelationship between AI and advanced analytics driving contemporary supply chain management (SCM). As a result, we are likely to see more sophisticated and flexible forecasting models emerging from this ongoing research, ultimately leading to improved global logistics.

4.2.2. Thematic Evolution

From the thematic evolution analysis, we gather significant insights that demonstrate how dynamic this area of research is (Figure 12). Notably, artificial intelligence (AI) shifted its focus from risk assessment to supply chain management from 2014 to 2024, marking a significant turning point. This transformation indicates a strong association with risk assessment (inclusion index score of 0.50) while also showing a moderate connection to inventory control and optimization. This suggests an increased emphasis on reducing risks and improving operational efficiency.

Deep learning has maintained continuity, particularly in its application to supply chain management and commerce, despite having a high stability index. This indicates that while it may not be the most prominent theme in the research landscape, it remains consistent and integral within the domain. Supply chain management continues to be a key topic across all time periods, increasingly interacting with emerging technologies, such as deep learning and learning algorithms. This ongoing integration demonstrates strong thematic growth, with supply chain management maintaining its original focus on predictive methods and machine learning while also incorporating a variety of emerging technologies and approaches.

The steady growth trend, initially present from 2014 to 2023, began to be associated with deeper learning by 2024. Although this relationship is still developing, it suggests a growing integration of sustainable practices with advanced AI technologies. A nearest-neighbor search and production control highlight the thematic dynamics of marketing and supply chain processes in later temporal periods. Although these correlations are not very stable, they indicate areas that may require further investigation and stabilization.

Overall, this study shows that artificial intelligence, deep learning, and supply chain management are interconnected. New applications and methodologies, particularly those related to risk assessment and optimization, are emerging, driving the future of supply chain management toward more efficient, streamlined, and sustainable practices. This continuous improvement underscores the importance of flexible predictive models and innovative approaches to addressing today’s supply chain challenges.

5. Discussion

This section aims to provide comprehensive insights that address the following key questions: How does the integration of AI-powered automation contribute to enhanced demand forecasting in the supply chain? What specific AI techniques are closely associated with demand forecasting in supply chain management? Additionally, what gaps exist in the current research environment that necessitate further investigation?

5.1. Comprehensive Insights from AI-Driven Automation in Supply Chain Management

This study shows that artificial intelligence (AI) significantly impacts demand forecasting in supply chain management. While some researchers focus on the specific applications of machine learning (ML), others agree that AI is revolutionary in addressing existing challenges.

Table 5. Supply chain applications and corresponding AI methods.

#	Supply Chain Application	AI Methods
1	Blood supply chain management	Machine learning, time series forecasting
2	Automated blood bank management	Automated machine learning, forecasting
3	Blockchain-enabled demand forecasting	Blockchain integration, machine learning
4	Supply chain optimization	Evolutionary algorithms, support vector regression (SVR)
5	AI in small- and medium-sized enterprises (SMEs)	Artificial intelligence (AI), machine learning, deep learning
6	Supplier selection and order allocation	Machine learning, optimization models
7	E-healthcare supply chain management	Deep learning, reinforcement learning
8	Predictive analysis in supply chain management	Machine learning, predictive analytics
9	Retail demand prediction with machine learning	Tree-based ensembles, long short-term memory (LSTM)
10	AI-based time-series forecasting for inventory	AI-based time-series forecasting, anomaly detection

presents various methods and trends in demand forecasting, highlighting successful strategies and associated probabilities.

For example, machine learning and scheduling prediction show tremendous potential in blood donation management (BSC) by predicting blood product demand and improving automated blood bank scheduling through proper sampling. Blockchain-enabled demand forecasting optimizes supply chain management, while combining machine learning and evolutionary algorithms with support vector regression (SVR) enhances demand forecasting and optimizes supply chain processes. AI also assists small- and medium-sized enterprises (SMEs) in automating supply chain planning and decision-making, particularly in supplier selection and order allocation.

Additionally, AI addresses anomalies in e-healthcare supply chain management (e-HSCM) by integrating predictive analytics to enhance deep learning and reinforcement learning processes. Furthermore, AI-based time series forecasting enables anomaly detection, leading to more accurate demand forecasts.

5.2. Technical Analysis and Comparative Effectiveness of ML and DL Models

To deepen the understanding of the application of machine learning (ML) and deep learning (DL) models in demand forecasting, the findings of previous studies were analyzed through technical and comparative analyses.

5.2.1. Long Short-Term Memory (LSTM) Networks

Technical Details: An LSTM network is a type of recurrent neural network (RNN) that uses memory cells to learn and retain long-term dependencies across broad sequences. This design enables LSTM networks to handle sequences of varying lengths and capture complex time series, making them more efficient for sequential time forecasting.

Performance: Studies, such as those by Phyu and Khine, have shown that LSTM networks outperform traditional models in capturing nonlinear relationships in demand data, resulting in higher prediction accuracy. LSTM models reduce forecast errors by 15–20% compared to ARIMA models in terms of retail demand forecasts.

5.2.2. Random Forest (RF)

Technical Details: Random forest is an ensemble learning method that constructs multiple decision trees during training. It makes predictions based on the majority class for classification tasks or the average prediction of individual trees for regression tasks.

Performance: According to Ji et al. (2019) [14], random forest (RF) models are robust against overfitting and provide high accuracy for demand forecasting. In vehicle demand forecasting, the RF model showed a 10% improvement in accuracy compared to traditional regression models.

5.2.3. Support Vector Machine (SVM)

Technical Details: Support vector machines (SVMs) are supervised learning models used for classification and regression analysis. They work by finding the optimal hyperplane that maximizes the margin between different classes. SVMs are particularly effective at handling high-dimensional data and can manage nonlinear relationships using kernel functions.

Performance: As highlighted by Al-Musaylh et al. [35], support vector machines (SVMs) perform well in demand prediction by finding the optimal hyperplane that maximizes the separation between groups. In electricity demand forecasting, the SVM model achieved approximately a 12% reduction in the forecast error rate compared to conventional linear models.

5.2.4. Extreme Gradient Boosting (XGBoost)

Technical Details: XGBoost is a highly efficient and scalable implementation of the gradient-boosting framework. It supports parallel tree development, allowing it to solve many data science problems quickly and accurately by optimizing both the loss function and model complexity.

Performance: According to Ji et al. [14], XGBoost models consistently achieve higher accuracy and faster computation times compared to other ensemble methods in demand forecasting tasks. For instance, in sales forecasting for e-commerce platforms, XGBoost models outperformed traditional boosting methods by 8–10% in terms of accuracy and reduced training times by 30%.

5.2.5. SARIMA–LSTM–BP Combination Model

Technical Details: This new model combines the strengths of SARIMA, LSTM, and backpropagation. SARIMA captures seasonal trends, LSTM learns and predicts temporal patterns, and backpropagation (BP) adjusts the model to improve its reliability and accuracy.

Performance: The SARIMA–LSTM–BP model, as discussed by Falatouri et al. [11], surpasses traditional econometric models and standalone LSTM models in predicting new energy vehicle demand, offering higher reliability. The performance of this hybrid model in prediction is also improved, with an 18% reduction in error compared to standalone LSTM models.

5.2.6. Comparative Effectiveness

Accuracy: Generally, models using LSTM and XGBoost significantly improve accuracy over traditional methods like SARIMA. For example, LSTM models reduce forecast errors by 15–20%, while XGBoost models improve accuracy by 8–10% [8,14].

Computation Time: XGBoost has proven to be faster in both training and prediction, making it highly useful for real-time forecasting applications [14]. Additionally, XGBoost’s support for parallel computing significantly shortens calculation times, enhancing its utility in dynamic environments.

Complexity: On the other hand, support vector machines (SVMs) and random forest classifiers are comparatively easier to implement and interpret than LSTM networks, which are more computationally expensive. This was affirmed by Al-Musaylh [35], who reported that Ji et al. [14] noted the complexity of the architecture and training of LSTM models, which require specialized knowledge and significant computational power.

Robustness: Random forest models resist overfitting and perform exceptionally well when there are a large number of predictors involved, whereas LSTM networks excel when the dataset exhibits temporal dependencies. The ensemble nature of random forest models makes them highly resilient to changes in data.

5.3. Refined Impact Assessment of AI on Supply Chain Demand Forecasting

Artificial intelligence in demand forecasting has ushered in a new era of precision and speed, particularly in the healthcare and retail sectors. Advancements in machine learning (ML) and deep learning (DL) have brought significant improvements in predictive analytics, leading to refined operational strategies. However, these developments are not without challenges: data quality issues, algorithmic biases, and integration with existing systems demand a balanced view on the impact of AI on core systems.

The demand forecasting process can be significantly enhanced by various AI techniques and tools. Long short-term memory (LSTM) networks are well-suited for handling complex time-series data, while support vector machines (SVMs) excel in both classification and regression tasks. Additionally, deep learning tools like convolutional neural networks (CNNs) and generative adversarial networks (GANs) fine-tune estimation processes, improving the depth and accuracy of training datasets and enhancing the analysis of sequential data. Furthermore, blockchain technology offers robust and transparent record-keeping, which enhances reliability in demand forecasts across complex supply chains. AutoML democratizes the use of AI and scales solution availability by simplifying model selection and deployment for advanced forecasting, making these tools accessible to non-experts.

This detailed perspective not only highlights the role of AI in improving supply chain efficiency but also connects specific AI technologies to their practical applications. It offers valuable insights for stakeholders who seek to make informed decisions regarding the strategic implementation of AI in supply chain management.

5.4. Addressing the Research Gaps and Future Directions

Although there has been significant growth in AI-based demand forecasting, some gaps still remain. Future studies should focus on the development and implementation of models that integrate more classical forecasting methods with AI applications across various industries. Additionally, research is needed to evaluate the efficiency of AI in geologically underexplored areas. Future research should also aim to establish standards for compatibility and to identify the benefits of incorporating AI into forecasting models. Improving algorithm sensitivity to more complex and unstructured data in supply chains, often considered “noisy”, is another area that requires attention [25,36].

5.4.1. Strategic Implications and Practical Applications

The findings reveal various ways in which AI has been utilized in supply chain management. For example, AI is used in the blood supply chain for time-series-based demand forecasting to improve inventory management of blood products [7,8]. Other AI methodologies, such as LSTM networks, are employed in the retail sector to predict consumer demand, enhance inventory levels, and meet customer expectations. Many applications also benefit from blockchain technology, which provides higher levels of trust and transparency. This trend has enabled many firms to gain new strategic advantages through post-implementation improvements.

5.4.2. Challenges and Future Directions

Machine learning (ML) techniques and deep learning (DL) have emerged as some of the most pervasive tools in demand forecasting across various industries. However, their use presents several challenges. Firstly, handling time series data is a significant issue because traditional machine learning algorithms, such as random forest regressor and support vector regressor, often fail to capture complex temporal dependencies within the data. This limitation has driven the development of novel methodologies, including the integration of K-means clustering, LASSO regression, and LSTM deep learning models, to more effectively address the complexities of demand forecasting.

Moreover, many sources of demand forecasting data are unreliable, either due to a lack of historical data, intrinsic biases, or both, which greatly hinders their adoption in business scenarios [37]. Despite these challenges, the potential of combining machine learning and deep learning techniques to enhance the accuracy of demand forecasting is promising. This potential is further realized when these techniques are synergistically integrated with complementary methods, such as time series analysis, data transformation, imputation of missing data, and feature extraction and selection, all of which leverage advances in technology and data processing [25].

Expanding from the microcosm of machine learning applications to the broader landscape of supply chains, numerous challenges impede the seamless adoption and effective implementation of these technologies. Chief among these challenges is the pivotal role of data quality and accessibility, with issues related to real-time data quality posing significant obstacles to the success of machine learning models [36,38]. Integrating machine learning and deep learning solutions into existing supply chain frameworks is a complex endeavor, often complicated by interoperability issues and the challenge of seamless integration [37].

Furthermore, the interpretability of complex machine learning and deep learning models remains a persistent challenge, especially in industries governed by stringent regulatory compliance, where it is crucial to explain the rationale behind predictions [39]. Additionally, implementing machine learning and deep learning solutions can be costly, posing a significant barrier for smaller enterprises that must balance the anticipated benefits against the financial investment required [40].

The dynamic nature of supply chains, which are prone to unexpected shocks and disruptions, adds further complexity. This requires continuous improvements and adjustments to machine learning models in response to rapidly evolving situations and sudden events. Looking forward, some emerging trends pose additional challenges for machine learning and deep learning in demand forecasting for supply chains. Explainable AI (XAI) is becoming increasingly important to address the challenge of interpretability by emphasizing transparency in decision-making processes alongside model accuracy [41,42]. Additionally, integrating machine learning and deep learning models with edge computing technologies is crucial for real-time decision-making. Future challenges will continue to revolve around optimizing efficiency and scalability [43].

This underscores the importance of cross-sector collaboration among academia, industry, and policymakers, necessitating interdisciplinary partnerships to bridge the gap between theoretical innovation and practical, scalable implementation. Moreover, future research needs to focus on sharpening ethical scrutiny regarding how automated decision-making systems function, striving for greater accuracy and resilience in forecasting.

6. Conclusions

In conclusion, this comprehensive literature review elaborates on how artificial intelligence is revolutionizing supply chain management through meticulous demand forecasting. The review analyzed 119 papers published over a decade, from 2015 to 2024, identifying important trends, emerging methodologies, and potential future challenges. The main findings from the analysis suggest a paradigm shift in supply chain dynamics driven by the exponential growth of AI-driven solutions. Notably, the increased research interest after 2021 indicates a recognition of the transformative potential of AI in demand forecasting within the supply chain. Various methodologies, ranging from machine learning to integration with blockchain, have been adopted, each offering specific advantages in enhancing predictive accuracy and operational efficiency.

Additionally, this paper emphasizes that AI research in supply chain management (SCM) should extend beyond the traditional domains of computer science, engineering, and business. Interdisciplinary cooperation is crucial to addressing the numerous challenges related to data quality, model interpretability, and ethical considerations associated with AI adoption.

However, several challenges remain to be addressed to fully realize these potential benefits. Issues related to data access, model transparency, and scalability will continue to be pressing and require coordinated efforts from researchers, practitioners, and policymakers to effectively integrate AI into supply chains. Moreover, increasing attention must be directed toward the ethical implications of AI-driven decision-making in supply chain contexts to ensure responsible and fair deployment of advanced technologies.

Looking ahead, the accomplishments identified in this review pave the way for future research endeavors that could unlock the full potential of AI in supply chain management. By addressing the challenges outlined and fostering collaborative innovation, the field can drive substantial improvements in efficiency, resilience, and sustainability within global supply chain operations.