Research on Multi-Class SVM Combining with LDA

In recent years, multi-class SVM has been one of the hot spots for many researchers, and multi-class classification based on clustering is one of strategies. Because the information of class-labels is not considered by clustering, too much branches of the binary-tree are formed, especially in the case of samples in different classes having similar features. To solve the problem, linear discriminant analysis is introduced to binary-tree, the pretreatment that training samples before clustering is done to find optimal feature space in which the samples in the same classes will be gathered together, while the samples in different classes will be loosed, so binary-tree is optimized and the implementation of the algorithm is improved. The experiment is carried out on the UCI data sets. The results show that this method reduces the branches of binary-tree and improves the accuracy of the algorithm.