Multi-class Classification Algorithm Based On Decision Tree Twin Support Vector Machine

As a new machine learning theory based on V-C dimension and structural risk minimization theory of statistical learning methods, Support Vector Machine(SVM) is usually applied in classification or regression, and because of its excellent learning ability, it has become a research hotspot. Jayadeva and his colleagues proposed twin support vector machine(TSVM), for two class classification problem, the idea of TSVM is to construct two non-parallel hyper planes, which were fitted a sample while away from the other samples, and the new samples are attributed to the class which it is close to the hyper plane. For the same training samples, TSVM have very good promotion in training speed that is about four times the traditional SVM, and compared with the traditional SVM which has better generalization ability. As a usual method of solving multi class classification problem, SVM decision tree has a fast calculation speed and no refused classification phenomena. Combining with the advantages of TSVM and decision tree, a pair of decision tree based on twin support vector machine(DT-TSVM) for multi classification is proposed in this paper. At first, the separability between classes is defined, secondly, all samples are divided into two with the highest separability by TSVM classifiers, and then divide each of the two sub-categories into two categories in the same way, finally all samples well be classified in different class. In addition, the test data from the UCI database are used in the LIBSVM toolbox for experiment, the result turn out that the DT-TSVM has faster training speed and higher accuracy.