Improved Multi-objective Evolutionary Subspace Clustering

This paper presents a subspace clustering method using an evolutionary-based multi-objective optimization framework. Recently, subspace clustering techniques become popular in solving many clustering problems where the key task is to identify groups of objects where the objects in each group have some similar properties with respect to a subset of features which are relevant to the group. Again, the simultaneous optimization of multiple objective functions helps to identify the subspace clusters effectively. The proposed method optimizes multiple objective functions simultaneously so that it can generate good quality subspace clusters. Two cluster validity indices namely XB-index and PBM-index are modified to make them applicable for subspace clustering problem. The evolutionary-based technique is used to simultaneously optimize these two validity indices to generate the subspace clusters. Various mutation operators have been used to generate good offsprings and to explore the search space effectively. The proposed approach is tested on 7 real-life data sets and 16 synthetic data sets. The efficacy of the proposed method is shown by comparing the results with many state-of-the-art algorithms.