A new initialization method for the Fuzzy C-Means Algorithm using Fuzzy Subtractive Clustering

Fuzzy C-means (FCM) is a popular algorithm using the partitioning approach to solve problems in data clustering. A drawback to FCM, however, is that it requires the number of clusters and the clustering partition matrix to be set a priori. Typically, the former is set by the user and the latter is initialized randomly. This approach may cause the algorithm get stuck in a local optimum because FCM depends strongly on the initial conditions. This paper presents a novel initialization method using fuzzy subtractive clustering. On both artificial and real datasets, this algorithm is able, not only to determine the optimal number of clusters, but also to provide better clustering partitions than standard algorithms. Availability: The supplementary documents and the method software are at http://ouray.ucdenver.edu/~tnle/fzsc.