An unsupervised possibilistic c-means clustering algorithm with data reduction

Abstract

Because of using the possibilistic partition to describe the data set, possibilistic clustering algorithm is more robust to noises than hard and fuzzy clustering algorithms. But calculating the membership matrix also makes it has a low efficiency. Moreover, the performance of possibilistic clustering may be degreased if the cluster number is set wrongly. In this paper, we proposed a new possibilistic clustering algorithm named unsupervised possibilistic c-means clustering algorithm with data reduction (UPCMDR) to improve the efficiency of possibilistic c-means clustering algorithm (PCM). In UPCMDR, data reduction technique is introduced to speed up the process of estimation of the cluster centers. A new clustering algorithm called weighted possibilistic c-means clustering algorithm is proposed to estimate the positions of centers of PCM accurately. The contrast experimental results with conventional algorithms show that UPCMDR has a relatively high efficiency, and can execute unsupervised clustering task when combining with the generalized cluster validity index.