Stepwise Selection on Steroids

Abstract

It’s likely that everyone reading this book is familiar with stepwise selection. Typically, you have a large set of candidates for some task, often prediction or classification. You test each individual candidate and select the one candidate that performs the task best. Then you test the remaining candidates, seeking the one that, in conjunction with the one already selected, performs best. This is repeated as desired. It is a fast, efficient, and usually fairly effective way of selecting a respectable subset of features from a potentially large population.