A method for training finite mixture models under a fuzzy clustering principle

摘要

In this paper, we establish a novel regard towards fuzzy clustering, showing it provides a sound framework for fitting finite mixture models. We propose a novel fuzzy clustering-type methodology for finite mixture model fitting, effected by utilizing a regularized form of the fuzzy c-means (FCM) algorithm, and introducing a proper dissimilarity functional for the algorithm with respect to the probabilistic properties of the model being treated. We apply the proposed methodology in a number of popular finite mixture models, and the corresponding expressions of the fuzzy model fitting algorithm are derived. We examine the efficacy of our novel approach in both clustering and classification applications of benchmark data sets, and we demonstrate the advantages of the proposed approach over maximum-likelihood.