A correlation based stochastic partitional algorithm for accurate cluster analysis

Abstract

Most partitional clustering algorithms such as K-means, K-nearest neighbour, evolutionary techniques use distance based similarity measures to group the patterns of a data set. However the distance based algorithms may converge to local optima when there are large variations in the attributes of the data set, leading to improper clustering. In this paper we propose a simple stochastic partitional clustering algorithm based on a Pearson correlation based similarity measure. Experiments on real-life data sets demonstrate that the proposed method provides superior performance compared to distance based K-means algorithm. Copyright � 2013 Inderscience Enterprises Ltd.