Selection of variables in multiple regression - an analysis of contemporary techniques

The problem of determining the best subset has two important aspects: the choice of a criteria defining what is meant by 'best' and a computational procedure to identify the best subset. The choice of a combination of a criteria and a computational procedure depends on the intended use of the subset model selected. Numerous criteria and computational procedures exist in literature. In this study a few of the more often used criteria and computational procedures were compared to evaluate their relative merits based on the intended use of the selected subset model. The study also incluhs an evaluation of a few methods for estimating the regression coefficients.