Adaptive Optimization of the Number of Clusters in Fuzzy Clustering

Abstract

In this paper, we present a local, adaptive optimization scheme for adjusting the number of clusters in fuzzy C-means clustering. This method is especially motivated by online applications in which a potentially changing clustering structure must be maintained over time, though it turns out to be useful in the static case as well. As part of the method, we propose a new validity measure for fuzzy partitions which is a modification of the commonly used Xie-Beni index and overcomes some deficiencies thereof.