Soft subspace clustering using differential evolutionary algorithm

Yangyang Li, Yujing Lu, Licheng Jiao

Abstract: This paper presents a differential evolutionary clustering approach to solve the optimization of the dimension weights in subspace, which is referred to as Soft Subspace Clustering Using Differential Evolutionary Algorithm (DESSC). The classical clustering methods can handle the low-dimensional rather than the high-dimensional data due to the curse of dimensionality. In addition, many subspace clustering approaches are sensitive to the initial points, and the results converge to local rather than global optimum. In the proposed algorithm, a novel technique for cluster data is developed to update the dimension weights by using differential evolution. DESSC combines the merits of differential evolution and the advantages of Soft Subspace Clustering. This contributes to avoiding trapping in local optimum and gets a stable clustering result. Moreover, it is robust and easy to implement. Experimental results on both synthetic and real data have shown that DESSC significantly outperformed several well-known algorithms, i.e., ESSC, FWKM, EWKM and LAC in almost all experiments.