Wavelet L_p-Norm Support Vector Regression with Feature Selection

Abstract

This paper proposes wavelet L_p-norm support vector regression (L_p-WSVR) to solve feature selection and regression problems effectively. Unlike conventional support vector regression (SVR), linear L_p-WSVR ensures that useful features are selected based on theoretical analysis. By using the wavelet kernel, L_p-WSVR approaches any curve in quadratic continuous integral space that leads to improving regression performance. Results of experiments show the superiority of L_p-WSVR in both feature selection and regression performances. Applying L_p-WSVR to Chinese real estate prices shows that the most significant and powerful factor contributing to Chinese housing prices is monetary growth.