SLMBSVMs: a structural-loss-minimization-based support vector machines approach

Abstract

Existing approaches,for constructing SVMs are based on minimization of structural risk where the generalization error loss is treated equivalently for each training pattern. Considering that error loss of one pattern is generally different to the other's in real binary classification problems, we propose a reformulation of the minimization problem such that generalization error rate for-each training pattern are treated respectively to minimize total generalization loss, which we call the structural-loss-minimization-based support vector machines (SLMBSVM). We. show experimentally that SLMBSVMs is potential.