Thursday, October 3
Thu, Oct 3, 2:45 PM - 3:47 PM
Evergreen I
Evergreen I
Speed Session 1
Weighted LAD-LASSO for Robust Variable Selection for Grouped Data (306419)
Nedret Billor, Auburn University*Kristin Lilly, Columbus State University
Keywords: Group variable selection, LASSO, Robust Statistics
Group variable selection is an interesting new problem, whereby the predictors in a multiple linear regression setting are non-arbitrarily grouped and selecting a subset of important groups of variable is of interest. In this paper, two new methods are proposed, the group WLAD-LASSO and the adaptive group WLAD-LASSO for selecting significant groups of variables when outliers are present in both the response and the predictor matrix. The theoretical properties of each of the methods is discussed. Preliminary results from a simulation study will be presented.