Density induced p-norm support vector machine for binary classification

Abstract

This paper presents a new version of support vector machine (SVM) named density induced p-norm SVM (0 <; p <; 1), DPSVM for shot. Our DPSVM introduces the density degrees into the standard p-norm SVM. It extracts the relative density degrees for the training examples and takes these degrees as relative margins for corresponding training examples. Our DPSVM not only inherits good performance of p-norm SVM which can realize feature selection and classification simultaneously, but also improves the performance of p-norm SVM. The numerical experiments results show that our DPSVM is more effective than some usual methods in feature selection and classification.