Abstract

　　A novel clustering method called the distance-based spectral clustering is proposed. The proposed clustering method is complete and total different from traditional spectral clustering method since it makes no assumption on regarding the suitable similarity measure and the prior-knowledge of cluster number. The pairwise distance matrix can be directly employed without transformation as a pairwise similarity matrix in advance. Besides, the Laplace operator is successfully applied to the pairwise distance matrix in the distance-based spectral clustering of which the inter-cluster structures and the intra-cluster relationships are both under consideration to increase the discrimination capability of resulting clusters. Experimenting on various datasets shows that the correctness of automatic cluster extraction and the robustness of noises even with the high level of noises. The systematic analysis of the proposed clustering method shows the outstanding performance as well. Moreover, the distance-based spectral clustering is applied to the real problem in the field of bioinformatics, which the experimental results verify its advantages including reliability, feasibility, and adaptability. According to the distinguishing characteristics, the distance-based spectral clustering can be assured that it is a remarkable clustering method.