Weight in competitive K-means algorithm

Abstract

K-means algorithm is a well-known clustering method. Typically, the k-means algorithm treats all features fairly and sets weights of all features equally when evaluating dissimilarity. However, experiment results show that a meaningful clustering phenomenon often occurs in a subspace defined by some specific features. Different dimensions make contributions to the identification of points in a cluster. The contribution of a dimension is represented as a weight that can be treated as the degree of the dimension in contribution to the cluster. This paper first proposes Weight in Competitive K-means (WCKM), which derives from Improved K-means and Entropy Weighting K-means. By adding weights to the objective function, the contributions from each feature of each clustering could simultaneously minimize the dispersion within clusters and maximize the separation between clusters. The proposed algorithm is confirmed by experiments on real data sets. © 2012 Springer-Verlag GmbH Berlin Heidelberg.