Subspace clustering based on differential evolution

Abstract

The performance of soft subspace clustering largely depends on the objective function and the search strategy. This paper presents a differential evolution (DE) based algorithm for subspace clustering. In the proposed algorithm, a novel objective function is firstly designed by considering the fuzzy weighting within-cluster compactness and loosening the constraints of dimension weight matrix. Then, a novel membership between a data point and a cluster is proposed. At last, an efficient global search strategy, composite DE, is introduced to optimize the proposed objective function to search subspace clusters. The simulation results show that both the proposed objective function and the introduced DE search strategy contribute to the performance enhancement of soft subspace clustering, and thus the proposed algorithm is significantly better than existing algorithms.