Multi-class Classification of Support Vector Machines Based on Double Binary Tree

To solve the problems of ´irreversibility´, ´error accumulation´ and randomicity of classification order in multi-class classification of support vector machines based on binary tree (BT-SVM), the paper proposes a multi-class classification method of support vector machines based on double binary tree (DBT-SVM). According to the method, each sub-classifier of BT-SVM is modified. After unknown samples are classified by the modified BT-SVM, the negative output of its final sub-classifier can be classified again by adding an Auxiliary BT-SVM so that the misclassified samples mixed in the negative output can be classified correctly. Experiment results show that the classification accuracy of earlier classified samples can be improved using DBT-SVM method, while the general classification accuracy does not decrease.