Cluster Validity Using Modified Fuzzy Silhouette Index on Large Dynamic Data Set

Abstract

Cluster validity index is applied to evaluate clustering results. It can be performed based on different measures and it can accomplish at data point level or cluster center level. In distance-based clustering methods, silhouette is an efficient point to point index measure, which defines relation based on compactness and separation distances. To validate fuzzy partitions, fuzzy silhouette index is used by applying defuzzification process. One of the applications of cluster validity is finding an optimal number of clusters in distance-based methods. As data size increases, point-wise index measure calculation takes more execution time. Hence, we proposed approaches to reduce time complexity by modifying fuzzy silhouette index at center to center and center to mean levels. All these methods are applied to find the right number of cluster and they are giving correct value in minimum execution time. All work is implemented in Matlab and effective results are given by our proposed methods.