Possibilistic approach to kernel-based fuzzy c-means clustering with entropy regularization

Abstract

The fuzzy c-means (FCM) is sensitive to noise or outliers because this method has the probabilistic constraint that the memberships of a data point across classes sum to one. To solve the problem, a possibilistic c-means clustering (PCM) has been proposed by Krishnapuram and Keller. An advantage of PCM is highly robust in a noisy environment. On the other hand, some clustering algorithms using the kernel trick, e.g., kernel-based FCM and kernel-based LVQ clustering, have been studied to obtain nonlinear classification boundaries. In this paper, an entropy-based possibilistic c-means clustering using the kernel trick has been proposed as more robust method. Numerical examples are shown and effect of the kernel method is discussed. © Springer-Verlag Berlin Heidelberg 2005.