A Novel Alternative Exponent-Weighted Fuzzy C-Means Algorithm

摘要

Under noisy environment and uneven data distribution, Fuzzy C-Means (FCM) algorithm and some of its advanced algorithms give large miss-clustering result or become malfunction. This paper proposes a novel Alternative Exponent-weighted Fuzzy C-Means (AEFCM) algorithm which introduces exponent-weight matrix and defines a new metric space. During iteration, the exponent-weight matrix gives every data sample a difference weight based on difference cluster center. Meanwhile, new metric space can efficiently restrain the bad influence produced by noisy samples during the iteration. Experiments have proved that AEFCM algorithm may overcome the bugs of FCM algorithm in a certain extent, with favorable convergence and robustness.