Modified global k-means clustering algorithm using mutual information

Abstract

In this paper, we propose a new unsupervised K-means algorithm to cluster data without initial guesses of the locations of the cluster centers or a priori information about the number of clusters. Cluster centers are incrementally obtained by adding one cluster center at a time with the modified global K-means clustering. The number of clusters is obtained by measuring the mutual information between clusters. Starting with one cluster, an optimal set of clusters can be iteratively found by adding until at least one pair of clusters displays positive mutual information. Then, the design parameter, β representing the clustering boundary, is used to reduce the computational load. Compared to the K-means, global K-means, and self-organizing map (SOM) methods, the proposed method shows as a similar performance while requiring less computation time. © 2013 American Scientific Publishers All rights reserved.