Succinct initialization methods for clustering algorithms

Abstract

In this paper, we focus on the problem of unsupervised clustering of a data-set. We introduce the traditional K-Means (K-means) cluster analysis and fuzzy C-means (FCM) cluster analysis of the principles and algorithms process at first, then a novel method to initialize the cluster centers is proposed. The idea is that the cluster centers' distribution should be as evenly as possible within the input field. A "Two-step method" is used in our evolutionary models, with evolutionary algorithms to get the initialized centers, and traditional methods to get the final results. Experiment results show our initialization method can speed up the convergence, and in some cases, make the algorithm performs better. © 2011 Springer-Verlag.