Fuzzy c-means clustering with conditional probability based K–L information regularization

Fuzzy c-means with regularization by K–L information (KLFCM) is an objective function method for clustering, which is regarded as a fuzzy counterpart of Gaussian mixture models (GMMs) with EM algorithm when the regularization parameter λ equals 2. However, KLFCM method extracts very close or even coincident clusters in many cases because the K–L information term in its objective function is used to minimize the dissimilarity between the membership degrees and the proportions of data belonging to the clusters. To deal with this problem, we propose a new model called fuzzy c-means clustering with conditional probability based K–L information regularization (CKLFCM) which incorporates the conditional probability distributions and the probabilistic dissimilarity functional into the conventional KLFCM algorithm in order to assign appropriate membership degrees to each data point. CKLFCM technique does not suffer from obtaining the unexpected close or coincident clusters. Several experiments are presented to show the effectiveness of the proposed algorithm.