Nature-Inspired Metaheuristics Algorithm for Data Clustering

Abstract

In this work, we proposed an optimized version of the K-Means clustering algorithm. Specifically, we improved the random initialization step of the centroids by replacing it with a strategy guided by the Black Hole Algorithm (BHA), a metaheuristic inspired by astrophysics. This led to the development of a hybrid algorithm named KMBHA, which combines the strengths of both K-Means and BHA. Instead of randomly selecting initial centroids, BHA searches for optimal starting points, which K-Means then uses to perform clustering. A comparative evaluation between KMBHA and standard K-Means was conducted on both synthetic and real datasets. The results demonstrated that KMBHA consistently outperforms K-Means, especially on complex real-world datasets such as Digits, where the high dimensionality and overlapping clusters highlight the limitations of conventional K-Means. These findings confirm the potential of our approach for more robust unsupervised clustering tasks. However, further improvements are possible and are discussed in the general conclusion.