Modified merging algorithm of fuzzy clustering with expected value

Abstract

The objective function of a fuzzy clustering algorithm is minimizing the sum of weighted distance function between data objects and cluster centers. Various cluster validity indexes evaluate the cluster's results from two aspects: compactness and separation measures. The fuzzy clustering algorithm with the expected value uses the merging algorithm to get separated clusters. The merging algorithm based on distance function. The distance function is not a good indicator to a separation measure. The clusters may be separated but these clusters are overlapped. This paper proposes a modified merging algorithm of fuzzy clustering with expected value (MCFEV). The MCFEV algorithm uses a similarity measure to merge overlapping clusters and to get the optimal number of clusters. The MCFEV algorithm is compared with the fuzzy c means (FCM) and possibilistic clustering algorithm (PCA) in the numerical examples and the real applications. A fuzzy cluster validity index (XB) is estimated for the above clustering algorithms. The results show the effective and accuracy of MCFEV algorithm.