DISCRIMINANT STEPWISE PROCEDURE

Abstract

Stepwise procedure is now probably the most popular tool for automatic feature selection. In the most cases it represents model selection approach which evaluates various feature subsets (so called wrapper). In fact it is heuristic search technique which examines the space of all possible feature subsets. This method is known in the literature under different names and variants. We organize the concepts and terminology, and show several variants of stepwise feature selection from a search strategy point of view. Short review of implementations in R will be given.