Method to Enhance the Recognition Performance of an SVM Based on the Altered Datasets

The class label of each feature vector in the dataset is respectively added in the corresponding feature vector as a feature value, which build a new vector called altered feature vector, all of which compose the altered dataset. It is demostrated that an SVM based on the altered dataset has advantages such as high generilization performance and little structure risk, compared with an SVM based on the original dataset. When predicting the unkown feature vector, different class labels (1 and -1) are respectively added to the unkown feature vector, and 2 altered feature vectors are got. Two hyper plane function values are obtained by substituting the 2 altered feature vectors into the hyper plane function respectively, and the symol (1 or-1) of the function value with larger absolute value is conducted as the class label of the unkown feature vector. Experiments results show that the proposed mehtod can enhance the recognition performance of an SVM effectively.