A Review of Metaheuristic Algorithms and Their Applications in Clustering

reihaneh zia¹ Ehsan soleimanidehkordi²

1) M.Sc. Student in Computer Engineering, Artificial Intelligence, Islamic Azad University, Borujen Branch, Iran Email:
2) Ph.D. Student in Artificial Intelligence, Islamic Azad University, Meybod Branch, Meybod,

IOI Code : ACCD-ACABAH
Publication : 10th International Conference on Applied Researches in Science & Engineering (carseconf.com)
Abstract :
    Clustering is one of the most important unsupervised learning techniques in data mining and artificial intelligence, aiming to identify hidden structures and group similar data points. Despite the widespread use of classical algorithms such as K-means, their sensitivity to initialization, susceptibility to convergence to local optima, dependence on the number of clusters, and difficulty in analyzing complex datasets have increased the need for advanced optimization methods. Metaheuristic algorithms, which employ population-based, evolutionary, and nature-inspired search mechanisms, offer considerable potential for improving clustering quality. This paper aims to review and analyze metaheuristic algorithms and examine their applications in clustering problems. Accordingly, algorithms such as Genetic Algorithms, Particle Swarm Optimization, Ant Colony Optimization, Simulated Annealing, Grey Wolf Optimization, and hybrid approaches are examined in terms of their operating principles, solution representation, objective functions, advantages, and limitations. Furthermore, clustering evaluation metrics, including the Silhouette Coefficient, Within-Cluster Sum of Squares, and Davies–Bouldin Index, are introduced to explain different dimensions of clustering performance. The review of existing studies highlights the importance of factors such as computational complexity, scalability, parameter tuning, the balance between exploration and exploitation, and robustness to noise. In addition, the role of hybrid algorithms and multi-objective optimization in developing more efficient clustering methods is discussed. The analytical findings indicate that selecting an appropriate algorithm depends on data characteristics, cluster structure, evaluation criteria, and computational resource constraints. Therefore, no single metaheuristic algorithm can be considered superior for all clustering problems. Finally, the development of adaptive, scalable, and reproducible methods is identified as an important direction for future research.

Keywords : metaheuristic algorithms clustering optimization unsupervised learning hybrid algorithms