A dissimilarity measure based Fuzzy c-means (FCM) clustering algorithm

Abstract

According to the definition of cluster objects belonging to same cluster must have high similarity while objects belonging to different clusters should be highly dissimilar. In the same way cluster validity indices for analyzing clustering result are based on the same two properties of cluster i.e. compactness intra-cluster similarity and separation inter-cluster dissimilarity. Most of the clustering algorithm developed so far focuses only on minimizing the within cluster distance. Almost all clustering algorithms ignore to include the second property of a cluster i.e. to produce highly dissimilar clusters. This paper recommends and incorporates a dissimilarity measure in Fuzzy c-means FCM clustering algorithm, a well-known and widely used algorithm for data clustering, to analyze the benefit of considering second property of cluster. Here we also introduced a new effective way of incorporating the effect of such measures in a clustering algorithm. Experimental results on both synthetic and real datasets had shown the better performance attained by the new improved Fuzzy c-means in comparison to classical Fuzzy c-means algorithm.