Abstract: Fuzzy C-mean algorithm is sensitive to initial centroid and the choice of fuzzy weighting exponent m plays an import role in clustering result PSO has the advantage of global optimization and good convergence speed. This paper proposed a new method by combining PSO and fuzzy GMeans to solve those problem. I3y a simple and effective particle encoding method, the best initial centroid and fuzzy weighting exponent were both searched in the process of PSO. Ex- periments on synthetic data sets and several UCI data sets achieve good results.

Key words: Clustering, FCM, PSO, Particle coding, Initial centroid