A spectral clustering algorithm for automatically determining clusters number
Bin Chen, Yalin Wang, Fan-Ying Gong, Xiaoli Wang, Chunhua Yang · 2014
Ascertainable clustering number is one of the vital problems of spectral clustering. To solve this problem, a spectral clustering algorithm automatically determining the clustering number is proposed. By mapping the sample point of the data set into feature space, the orthogonal positional relationship of sample points between different clusters in the feature space can be determined. Based on the orthogonal relationship, the proposed spectral algorithm calculates and analyses the angle between mapping points to determine the optimum clustering number. Simulation results show that: the proposed algorithm not only can get the correct number of clusters both on multiple artificial date sets and on practical data set of the alumina evaporation process, but also has the least calculating time comparing with Self-Tuning and SASC algorithm.