A multiclass decision rule minimizing a loss function based on one class SVM

Abstract

A multiclass algorithm based on ν-1-SVM which minimizes a loss function is introduced. The loss function allows to use decision costs which depend on the classes, and to consider partial and total rejection. The algorithm is based on deriving the ν-1 SVM regularization path for each class. The decision rule is determined by tuning all the ν-1-SVM parameters and all the other decision parameters together in order to minimize the loss function. Experimental results on artificial data sets and some benchmark data sets are provided to assess the effectiveness of this approach.