A total multi-class support vector machine

A new multi-class classification algorithm, called total K-SVCR, is proposed in this paper. The proposed algorithm not only makes the correctly classified samples as far as possible from the classification hyper-plane, but also makes the misclassified samples as close as possible to it, which could avoid the disturbance of noises or outliers to a certain degree. Compared with K-SVCR on four benchmark datasets, the proposed algorithm yields higher prediction accuracy and costs lower. It demonstrates the feasibility and validity of the proposed algorithm. Copyright © 2011 Binary Information Press.