An incremental learning algorithm of multiple support vector machines

Abstract

Based on analyzing the construction process of HT-SVM, this paper proposes incremental learning algorithm of multi-class SVM based on Huffman tree. This method is to convert the incremental learning of multi-class SVM into the incremental learning of two-class SVM. Firstly, construct the multi-class SVM based on Huffman tree according to original training dataset. Then, according to the structure of HT-SVM, the new adding dataset is divided into multiple intersection subsets of two-class (If there are k classes of the training dataset, the number of the multiple intersection subsets of two-class is k-1). Finally, the k-1 subsets is send to k-1 two-class classifiers of HT-SVM to be learn using incremental learning algorithm of two-class SVM. Simulate with KDD CUP 1999 dataset, and the experiment results show the performance.