The integration of machine learning techniques with optimization algorithms has garnered increasing interest in recent years. Two primary purposes emerge from the literature: leveraging metaheuristics in machine learning applications such as regression, classification, and clustering, and enhancing metaheuristics using machine learning to improve convergence time, solution quality, and flexibility. Machine learning techniques offer real-time decision-making capabilities, dimension reduction, and dynamic programming, contributing to robust decision-making processes capable of handling substantial data volumes and addressing stochastic events. To our knowledge, this paper represents the first comprehensive review that explicitly classifies and analyzes both fields, emphasizing their commonalities and delineating the area of their intersection. Through this exploration, our objective is to enrich the understanding of both metaheuristics and machine learning, foster interdisciplinary collaborations, and catalyze innovative approaches that harness the synergies between these two domains.
Top1. Introduction
Metaheuristic methods and machine learning have emerged as powerful tools for tackling complex optimization problems and making data-driven predictions, respectively. While these two domains have traditionally been treated as distinct fields of study, recent research has shown significant potential in their principles and methodologies (da Costa Oliveira et al., 2023; Eddaly et al., 2023; Houssein et al., 2022; Kalita et al., 2024; Karimi-Mamaghan et al., 2022; Valadi et al., 2024). By combining the strengths of both methods, researchers have achieved remarkable advancements in solving challenging optimization problems and enhancing predictive models’ performance. However, researchers who are not specialized in both domains sometimes overlook the commonalities between operational research and machine learning. To bridge this gap, the primary objective of this study is to provide a comprehensive exploration of the links and shared aspects between metaheuristic methods and machine learning. Through an extensive literature review, we seek to establish a solid foundation for understanding the potential synergies that arise from their integration. Furthermore, this study highlights notable contributions and identifies key research directions.
This work aims to provide a thorough mapping of methods and research endeavors, addressing the lack of a comprehensive analysis covering various aspects related to both metaheuristics and machine learning domains. The key contributions of this research include:
- •
Presentation of a comprehensive survey of metaheuristic algorithms.
- •
Discussion of challenges faced by metaheuristics researchers.
- •
Comprehensive survey of machine learning algorithms, including the classification of different categories and a historical overview of well-known algorithms.
- •
Discussion of challenges faced by machine learning community.
- •
Highlighting common aspects between both fields and emphasizing the synergistic power of their combination.
The organizational structure of this investigation is presented as follows: in Section 1 we have inaugurated the discourse with an introductory exposition elucidating both the conceptual foundations and the rationale underlying the exploration of synergies between metaheuristic methods and machine learning. Section 2 systematically expounds upon metaheuristic methods, providing a nuanced examination of their inherent characteristics, presenting some of the classical approaches, and culminating in a discussion on the principal challenges and impediments encountered by researchers within this domain. In Section 3, a comprehensive panorama of machine learning unfolds, encompassing diverse learning paradigms and algorithms. The section 3 further encapsulates an elucidation of prominent algorithms, their historical evolution, and concludes with an examination of the primary challenges confronting researchers in the field. Section 4 delves into the nexus of convergence between metaheuristic methods and machine learning, accentuating shared techniques and methodologies. This section serves as an encompassing review of extant research situated at the confluence of these two domains. Finally, Section 5 furnishes a concluding synthesis of the study, encapsulating key findings, and delineates prospective avenues for future research endeavors.
TopMetaheuristics (MH) constitute a broad class of algorithms introduced by Glover in the early 1980s. Tailored to address the complexities inherent in solving intricate optimization problems, especially Combinatorial Optimization Problems (COPs). MHs offer a potent approach where exact methods prove computationally infeasible, and classical heuristics falter in terms of effectiveness and efficiency. Notably, MHs have swiftly gained popularity, highlighting an impressive trade-off between solution quality and computational time.