A Self Organizing Map Based Multi-objective Framework for Automatic Evolution of Clusters

Abstract

The current paper reports about the development of an automatic clustering technique which builds upon the search capability of a self-organizing multi-objective differential evolutionary approach. The algorithm utilizes new search operators which are developed after considering the neighbor-hood relationships of solutions of a population extracted using a self organizing map (SOM). Variable number of cluster centers are encoded in different solutions of the population which are evolved using the new search operators of differential evolution to automatically determine the number of clusters. Two cluster validity indices capturing different goodness measures of partitioning are used as objective functions. The effectiveness of the proposed framework namely, self organizing map based multi-objective (MO) clustering technique (SMEA_clust) is shown for automatically partitioning four artificial and four real-life data sets in comparison with a multi-objective differential evolution based clustering technique (similar to our proposed approach but without using SOM concept), two recent multi-objective clustering based techniques, VAMOSA and MOCK. Results are further validated using statistical significance tests.