Abstract:

In order to reduce the computational complexity of SVC(Support Vector Clustering)on the basis of the proximity graph model developed by Yang et al.,the Euclidean distance in the Hilbert space is calculated by using a Mercer kernel,which is used as the weight criterion to generate a MST(Minimum Spanning Tree).The connec-tivity estimations are then lowered by only checking the linkages between the edges that construct the main stem of the MST.Moreover,the non-compatibility degree is defined to support the edge selection during linkage estima-tions.Experimental results confirm that.compared with the proximity graph model,the proposed approach is of fas-ter speed,optimized clustering quality and strong ability of noise suppression,which makes SVC advantageous in dealing with large data sets.

Key words: support vector machine, support vector clustering, proximity graph, minimum spanning tree