Adaptive robust variable selection (Q2448733)

The common situation encountered when using massive, large-scale data-sets is that the number of covariates greatly exceeds the number of observations, with the solution of selecting the most important variables. In this context, the paper deals with the adaptive robust variable selection, introducing the penalized quantile regression with the weighted \(L_1\)-penalty (WR-Lasso) for robust regularization. In order to make the bias reduction feasible, the adaptive robust Lasso (AR-Lasso) is also introduced. Numerical studies demonstrate the favorable finite-space performance of this approach.