Evolutionary search for optimal fuzzy c-means clustering

This paper introduces an evolutionary approach to automatically determine the optimal number and location of prototypes for the well-known fuzzy c-means (FCM) clustering algorithm. This approach is based on a clustering genetic algorithm (CGA) specially designed for clustering tasks. It uses context-sensitive genetic operators to globally explore the search space in such a way that the strong dependence of the FCM algorithm on adequate previous estimations of the number and initial positions of its cluster prototypes is avoided. In this case, FCM works as a local search engine to speed up convergence and improve accuracy of the overall evolutionary procedure. Two examples are presented to illustrate that the proposed algorithm is able to automatically find adequate clustering either starting from underestimated or overestimated initial number of clusters.