VARIABLE SELECTION WITH AKAIKE INFORMATION CRITERIA : A COMPARATIVE STUDY

Abstract

In this paper, the problem of variable selection in linear regression is considered. This problem involves choosing the most appropriate model from the candidate models. Variable selection criteria based on estimates of the Kullback-Leibler information are most common. Akaike’s AIC and bias corrected AIC belong to this group of criteria. The reduction of the bias in estimating the Kullback-Leibler information can lead to better variable selection. In this study we have compared the Akaike Criterion based on Fisher Information and AIC criteria based on Kullback-Leibler.