A SimRank based Ensemble Method for Resolving Challenges of Partition Clustering Methods

Abstract

Traditional clustering techniques alone cannot resolve all challenges of partition-based clustering methods. In the partition based clustering, particularly in variants of K-means, initial cluster centre selection is a significant and crucial point. The dependency of final cluster is totally based on initial cluster centres; hence, this process is delineated to be most significant in the entire clustering operation. The random selection of initial cluster centres is unstable, since different cluster centre points are achieved during each run of the algorithm. Ensemble based clustering methods resolve challenges of partition-based methods. The clustering ensembles join several partitions generated by different clustering algorithms into a single clustering solution. The proposed ensemble methodology resolves initial centroid problems and improves the efficiency of cluster results. This method finds centroid selection through overall mean distance measure. The SimRank based similarity matrix find that the bipartite graph helps to ensemble.