Adaptive Neighborhood Search-Based Hybrid Clustering (ANSHC): A Robust Metaheuristic Framework for Partitional Data Clustering

A key objective of clustering, which is a basic operation in machine learning and data mining, is to divide data into significant groups according to similarity. Conventional clustering techniques like K-Means frequently encounter issues with local optima and are highly sensitive to initial conditions. This paper proposes a novel algorithm called Adaptive Neighborhood Search-based Hybrid Clustering (ANSHC) to address these limitations, which combines global meta-heuristic optimization with adaptive local search refinement. In the proposed ANSHC, an adaptive neighborhood search (ANS) is used so as to change the searching radius adaptively during the constrained optimization in a dynamic way between exploration and exploitation. Furthermore, we incorporate a hybrid local refinement operator to improve the precision of convergence and the compactness of clusters. The experimental result shows that the proposed method performs very well in terms of accuracy, convergence speed, and silhouette coefficient and also maintains high stability over more than one run. The Wilcoxon signed-rank test used to perform the statistical test then verifies the performance of the proposed method. The performance of ANSHC shows better results with 95% accuracy and efficiency and provides a solid and adaptable framework for enhancing the data clustering performance.