Abstract: A fast support vector machines mufti class classification was proposed. Firstly, the number of every class training samples is used as weight to construct Huffman binary tree, and then train sub-classifiers for every non leaf node in the binary tree. For the sample to be classified, the sub-classifiers that between the root and a certain left are　used to classify, the left is the class of the sample. The classification experiments on the Reuters 21578 database arc done with this algorithm. The experimental results show that it has better performance and partly overcomes the flaw of existing mufti class classification algorithm of support vector machines, which is slow in the process of classification.This algorithm can remarkably increase the speed of classification, especially in the case of more classes and the scale of　every class is uniform.

Key words: Support vector machine, Mufti class classification, Binary tree