Regularization path for linear model via net method

Abstract

We investigate a net regularization method for variable selection in the linear model, which has convex loss function and concave penalty. Meanwhile, the net regularization based on the use of the Lr penalty with 1/2 ≤ r ≤ 1. In the simulation we will demonstrate that the net regularization is more efficient and more accurate for variable selection than Lasso. © 2012 Springer-Verlag.