Penalized Linear Regression Methods where the Predictors Have Grouping Effect

  • Kanyalin Jiratchayut
  • Chinnaphong Bumrungsup
Keywords: Adaptive elastic net, correlation based penalty, elastic net, SCAD-L2, variable selection

Abstract

The aim of this paper is to study the performance of four different penalized linear regression methods: elastic net, adaptive elastic net, L1CP, and SCAD-L2.Simulation studies show that the adaptive elastic net performs best in variable selection and parameter estimation, while the SCAD-L2 has prediction accuracy better than the other methods. When sample size is large, the L1CP has a prediction performance close to the prediction accuracy of the SCAD-L2.

Published
2019-07-10
Section
Articles