Abstract:

In this paper, three novel fuzzy clustering models are proposed based on the principle of cluster center separation. In the investigation, first, a fuzzy clustering model named FCM CCS is proposed by adding a cluster center separation item to the objective function of fuzzy C-means （FCM） algorithm to extend FCM algorithm based on the principle of cluster center separation. FCM_CCS enlarges the distance between cluster centers during the clustering and results in a better clustering effect. Then, a possibilistic clustering model named PCM_CCS is pre- sented to overcome the noise sensitivity of FCM_CCS. Finally, PCM_CCS is extended to its possibilistic fuzzy clustering model named PFCM_CCS. PFCM_CCS is of good performance in dealing with noisy data and in overcoming co- incident clusters. The test results of data sets show that PFCM_CCS simultaneously produces fuzzy membership values and typicality values, and possesses larger cluster center distance than FCM and higher clustering accuracy than FCM_CCS.

Key words: cluster center separation, fuzzy clustering, fuzzy C-means clustering, clustering model