A NOISE INSENSITIVE SVM MULTI CLASS CLASSIFIER

Classifier based on SVM theory has been developed as a general purpose two class classifier. But it is very sensitive to noise and incapable of multi class classification. In this paper, the standard PCA is extended to the more general field. And a new SVM learning structure is proposed, which uses the extended PCA algorithm to map the training set to a new de noised data set, and then combines many two classifiers via dissymmetry matrix into a multi class classifier to deal with the data set. The theoretical analysis and experimental results show that this classifier is noise insensitive and of high learning efficiency.