Classical and Robust Forward Selection: A Simulation Study and Real Data Application

Abstract

In order to use any method as a model selection algorithm, it is needed to check the adequacy and stability of a model selected by the algorithm. Adequacy of a robust model was not checked by giving outliers in various ways. In this paper, several contamination cases have been introduced to check the adequacy of the robust model selected by robust forward selection (RFS). In each of the contamination case, the performance of RFS has been compared to standard forward selection (FS) through a simulation study. The adequacy and stability of the robust model has also been checked through a real data application. Based on simulation study and real data application, RFS has much better performance compared to standard FS.