Support vector regression with penalized likelihood

Abstract

This paper is concerned with the method of support vector regression (SVR) with penalized likelihood. The ε-insensitive loss function utilized in SVR is naturally incorporated into the likelihood and is combined with the penalty for the vector of regression coefficients. We include all parameters necessary to implement SVR in the proposed penalized likelihood. An efficient algorithm to obtain estimators of parameters is provided and asymptotic results for the estimators are developed. We perform Monte Carlo simulations to confirm the effectiveness of the proposed method and report the results of applying the method to real data sets.