Linear penalization Support Vector Machines for feature selection

Abstract

Support Vector Machines have proved to be powerful tools for classification tasks combining the minimization of classification errors and maximizing their generalization capabilities. Feature selection, however, is not considered explicitly in the basic model formulation. We propose a linearly penalized Support Vector Machines (LP-SVM) model where feature selection is performed simultaneously with model construction. Its application to a problem of customer retention and a comparison with other feature selection techniques demonstrates its effectiveness. © Springer-Verlag Berlin Heidelberg 2005.