Variable selection in high-dimensional partly linear additive models

Abstract

Semiparametric models are particularly useful for high-dimensional regression problems. In this paper, we focus on partly linear additive models with a large number of predictors (can be larger than the sample size) and consider model estimation and variable selection based on polynomial spline expansion for the nonparametric part with adaptive lasso penalty on the linear part. Convergence rates as well as asymptotic normality of the linear part are shown. We also perform some Monte Carlo studies to demonstrate the performance of the estimator.