Abstract: The traditional "fuzzy" clustering(FCM) converges slowly when confronted with a large number of data points,meanwhile it can't deal with non-hyper spherical data structure,which compel us to present a new fuzzy clustering algorithm-the KPIM algorithm based on partition index maximization(PIM) algorithm and the kernel method.As well the paper proves convergence theorem of the new algorithm.The results of experiments on the real data show that the KPIM algorithm can effectively cluster on data with diversiform structures while guaranteeing the computation time in contrast to other previous algorithms.

Key words: fuzzy, clustering, PIM, kernel method, KPIM algorithm