Density Ratio Estimation in Support Vector Machine for Better Generalization: Study on Direct Marketing Prediction

摘要

In this paper we show how to improve the generalization performance of Support Vector Machine (SVM) by incorporating density ratio based on Unconstrained Least Square Importance Fitting (uLSIF) into the SVM classifier. ULSIF function is known to have optimal non-parametric convergence rate with optimal numerical stability and higher robustness. The ULSIF-SVM classifier is validated using marketing dataset and achieved better generalization performance as compared against classic implementation of SVM.