An Enhanced Metaheuristic Approach for Optimizing Graph-Based K-Means Clustering through Collective Intelligence

Abstract

Data clustering is a fundamental aspect of data mining, serving as a crucial tool for data description and organization. The objective of clustering is to group data points into distinct or overlapping clusters, where data within each cluster exhibit high similarity while being dissimilar to data in other clusters. The emerging clustering technique, known as clustering based on graph K nearest reciprocal neighborhood, has demonstrated notable advancements by addressing issues associated with conventional clustering algorithms, such as the reliance on predefined cluster counts and initial center selection. However, a notable challenge remains in determining the optimal neighborhood parameter 'K' that governs the selection of nearest neighbors. In this paper, we introduce a solution leveraging the power of collective intelligence and metaheuristic algorithms to effectively address the sensitivity issue surrounding the 'K' parameter. Specifically, we employ the Jaya algorithm, distinguished by its minimal control parameters, which not only facilitates population convergence towards optimal solutions but also efficiently distances itself from suboptimal ones. This unique characteristic contributes to a heightened convergence rate when compared to similar metaheuristic methods. We conduct rigorous experiments utilizing datasets from the UCI repository, demonstrating the superiority of our proposed approach over existing methods in terms of clustering accuracy. Our research offers a robust and innovative solution for refining graph-based K-means clustering, further enhancing its applicability and effectiveness in real-world data analysis scenarios.