A Clustering Algorithm Based on Probabilistic Crowding and K-means

This paper proposes a clustering method based on probabilistic crowding and k-means. The clustering problem is first converted to a multimodal function optimization with genetic niching. The peaks of multimodal function, which constitute the initial cluster centers for k-means, are identified by probabilistic crowding. The stability analysis proves that the algorithm can reliably converge to cluster centers. Improved k-means algorithm is presented and used for refinement of clustering centers and final clustering. Therefore it is unnecessary to predefine initial clustering centers and the number of clusters. The experimental results showed that the clustering algorithm could not only increase the convergence speed remarkably, but also was robust to the presence of noise. As a result, the proposed algorithm is superior to traditional genetic k-means algorithm in clustering results