Costs sensitive to examples learning in reduced support vector machine
代价与样本相关的简约核支持向量机

In many machine learning domains, misclassification costs are sensitive to examples. As an extension of class dependent costs, a cost-sensitive reduced Support Vector Machine (SVM) named sd2sSVM that aimed at minimizing all costs was introduced. Firstly, through the use of Generalized SVM (GSVM) framework, the optimization object was converted into unconstrained mathematical programming problems. Secondly, based on smooth piecewise polynomial function that was used to approach the plus function, the unique optimization solution can thus be gained by Newton-YUAN method. Finally, reduced kernel was employed to improve the solution of nonlinear problem. The experimental results show that sd2sSVM is comparable or choicer than traditional example dependent cost-sensitive SVM. It was also discussed that how parameter C influenced the performance of sd2sSVM.