Sliced Inverse Regression (SIR) via group lasso

Abstract

In this paper the authors propose a group-lasso for sliced inverse regression (group lasso-SIR). This proposed method can deal with the problem of correlation existence between predictor variables. Simulation is used to investigate the performance of proposed method comparing with ridge and lasso in sliced inverse regression (lasso-SIR). The results show that the group lasso-SIR method is performs well comparing with other methods depending on Mean Square Errors (MSE) criterion.