SCEOMOO: A novel Subspace Clustering approach using Evolutionary algorithm, Off-spring generation and Multi-Objective Optimization

Abstract

Subspace clustering is an extension of conventional clustering techniques that seeks to find clusters in different subspaces within a data set with higher dimensions. But, recent research results indicate that there are few challenges to be addressed efficiently, such as finding number of subspace clusters automatically and getting optimal subspaces by satisfying multiple objective criteria. This paper focused on these two challenges and proposed a framework called Subspace Clustering using Evolutionary algorithm, Off-Spring generation and Multi-Objective Optimization (SCEOMOO) to find the optimal subspace clusters. The proposed SCEOMOO model is tested on six standard data sets with different validity indices and results are compared with well known existing subspace clustering methods. The comparative performance analysis reveals that the proposed SCEOMOO gives significant improvements in the performance with respect to some of the well known existing models.