An Improved Possibilistic C-Means Algorithm Based on Kernel Methods
A novel fuzzy clustering algorithm, called kernel improved possibilistic c-means (KIPCM) algorithm, is presented based on kernel methods. KIPCM is an extension of the improved possibilistic c-means (IPCM) algorithm. Different from IPCM which is applied in Euclidean space, KIPCM can make data clustering in kernel feature space. With kernel methods the input data can be implicitly mapped into a high-dimensional feature space where the nonlinear pattern now appears linear. It is unnecessary to calculate in this high-dimensional feature space because we directly calculate inner products from the input data by kernel function. KIPCM can identify clusters of complex shapes and solve nonlinear separable problems better than IPCM and FCM (fuzzy c-means). Our experiments show that the proposed algorithm compares favorably with FCM and IPCM.