Abstract: We present a fuzzy kernel C-means clustering algorithm(FKCM) which is a generalization of the conventional fuzzy C-means clustering algorithm(FCM). This new FKCM algrotihm integrates FCM with the Mercer kernel function and can cluster non-hyperspherical data structure, data with noise, mixed data structure of multi pattern prototypes, asymmetric data structure, etc. This generalization can obviously improve the performance of the fuzzy C-means clustering algorithm. It is pointed out that the FKCM algorithm with the first-order polynomial kernel function is equivalent to the FCM algorithm. The results of experiments on the artificial and real data show that the fuzzy kernel C-means clustering algorithm can effectively cluster on data with diversiform structures in contrast to the fuzzy C-means clustering algorithm.

Key words: clustering analysis, fuzzy C-means algorithm, kernel-based method, unsupervised learning