Modified fuzzy c-mean for custom-sized clusters

Abstract

Fuzzy c-mean (FCM) is one of the widely used data clustering methods. FCM method not only divides a data set into several clusters but also determines the potential belongingness of each data in different clusters. The size of clusters generated by FCM cannot be controlled by the inherent mechanism. However, sometimes real life situations demand that the clusters should have some pre-specified size. In this study, the FCM method is further extended to obtain clusters with specified size. In the first step of the proposed method, FCM algorithm is executed; later the potential belongingness matrix passes through an optimization model to yield clusters with specified sizes. In the proposed technique, the centres of the clusters obtained from FCM are considered but the boundary elements are redistributed to achieve equal or custom-sized clusters. The methodology has been explained further with examples.