Robust Fuzzy-Possibilistic C-Means Algorithm

In allusion to the disadvantages that fizzy c-means algorithm is sensitive to noise and possibilistic c-means is easy to generate superposition cluster center, a novel algorithm (FPCM) which simultaneously produces both memberships and possibilities was proposed in 1997. However, FPCM still uses a norm-induced distance, as a consequence, its performance on the noisy data is not strong enough. In this paper, a new algorithm using the "kernel method" based on the classical FPCM is presented and called as robust fuzzy-possibilistic algorithm (RFPCM). RFPCM adopts a new kernel-induced metric in the data space to replace the original Euclidean norm metric in FPCM. Experiments on the artificial and real datasets show that RFPCM has better clustering performance and is more robust to noise than FPCM and PCM.