An Improved Possibilistic Clustering Based on Differential Algorithm

A possibilistic clustering algorithm called unsupervised possibilistic clustering (UPC) was proposed in a previous paper. Although UPC is sound, the algorithm has the problem of generating coincident clusters. In this paper, we propose a new clustering model called improved unsupervised possibilistic clustering (IUPC) to overcome this weakness of UPC, and an efficient global optimization technique-differential evolution algorithm (DE) is introduced to optimize the proposed model. In IUPC, the optimal cluster centers are searched by the DE algorithm within a fixed feasible region, which is determined by the fuzzy c-means clustering algorithm. IUPC inherits the merits of UPC. In the meanwhile, it avoids the problem of generating coincident clusters by limiting the feasible regions of different clusters disjoint. The contrast experiments with PCM and its variants show the effectiveness of IUPC.