Improving probabilities in a fuzzy clustering partition

Fuzzy clustering procedures based on the FCM algorithm calculate group membership probabilities or degrees taking into account the distance between objects and group prototypes. This paper seeks to improve the computation of such membership probabilities by a new membership function which also reflects the relative position of an object with respect to each group. By this way, some illogical results are avoided and a convex partition is provided. Finally, numerical examples illustrate the performance of the proposed algorithm.