ALL SUBSETS REGRESSION USING A GENETIC SEARCH ALGORITHM

Subset regression procedures have been shown to provide better overall performance than stepwise regression procedures. However, it is diffi cult to use them when a large number of candidate variables exists. Th is is due to the high computational costs associated with the combinat orial nature of evaluating each potential subset. To resolve this diff iculty, the use of a ''Genetic Algorithm'' (GA), a global optimization search procedure, is proposed to reduce the number of subsets which m ust be evaluated. Any of a number of popular criteria, including Mallo ws' Cp, MSE, R(2), AIC, etc., can be used to drive the search strategy associated with GA. Several illustrated examples on its use are provi ded.