Support Vector Machine Algorithm Based on Kernel Hierarchical Clustering for Multiclass Classification

Decision-tree-based multiclass support vector machine (DTSVM) can solve the problem of unclassified regions that exists in the conventional SVM. But the classification precision and generalization ability of DTSVM classifier depends on the structure of the decision tree. In addition, the training speed of DTSVM becomes slower for more training samples. In this paper, a new measurement of inter-class separability is defined, and an kernel hierarchical clustering algorithm is given by using kernel function to hierarchical clustering, then a fast training algorithm based on K nearest neighbours is given, at last, a SVM algorithm for multiclass classification based on kernel hierarchical clustering (KHC-SVM) is proposed. Experiment results proved the effectiveness of KHC-SVM.