Classification performance of support vector machine with ε-insensitive loss function

The ε-insensitive loss function generally employed in support vector regression is introduced into support vector classification, and the support vector classification with ε-insensitive loss function (ε-SVC) is proposed. Compared with the standard support vector classification method (C-SVC) and the least square support vector classification method (LS-SVC), the experimental result indicates that the classification accuracy ratio of ε-SVC is slightly lower than that of C-SVC and LS-SVC when ε sufficiently approaches to 1, but the training, testing and parameter selecting rates of ε-SVC are superior to that of C-SVC and LS-SVC, especially for large scale problem. Through accurate search of the parameter ε, the ε-SVC is endowed with higher classification accuracy than C-SVC and LS-SVC, however, the training, testing and parameter selecting rates decrease with smaller ε.