New Algorithm NDR-SVM to Improve Accuracy

A new algorithm based on neighbor domain rate NDR-SVM is proposed.Construct a neighbor domain for each point in the training set.Figure out the ratio of samples in this sphere,which has the same label with its center, and reserve or delete its center according to the relationship between the ratio and the eliminating range.Then the re- mained are used for SVM classification.Through pruning the training set,the proposed algorithm decreases the effects of noises on the generalization ability of SVM.Numerical experiments show that,compared with existing algorithm, NDR-SVM has higher accuracy and greatly improves the training speed.