A New Multi-Classification Method Based on Binary Tree Support Vector Machine

Binary tree support vector machine, which combines support vector machine and binary tree, is an effective way for solving multiclass problems. Classification accuracy and decision speed of the classifier relate closely to the structure of the binary tree. To maintain high generalization ability, most separable classes should be separated at upper nodes of a binary tree. And in order to obtain classification results rapidly, levels of the binary tree should be fewer. In this paper, a new binary tree with fewest levels based on clustering method is established. The efficiency of the improved binary tree support vector machine multiclassifier is proved by the results of experiment.