Linear Regression Predictors Selection Using Genetic Algorithm



The selection of predictors in a linear regression model is often a challenging problem especially when the dataset under consideration has large number of attributes. Ideally, one would like to obtain the most predictive regression model with less number of predictors. Many techniques such as expert based selection, stepwise regression, stochastic search heuristics and Genetic Algorithms (GAs) have been applied to tackle this task. In this study, we investigate the performance of  GA for predictors selection (GAPS) by using Akaike information criterion (AIC) and Bayesian information criterion (BIC) statistical criteria. We have compared the performance of the proposed GA on four datasets. The performance of GAPS is found to be comparable in terms of Adjusted-R2, RSE, and predictive accuracy on AIC and BIC criteria, yet, BIC criteria selects less number of predictors than the AIC.


regression model, genetic algorithm, predictors selection, variable selection.

Full Text:



Subscribe to Print Journals

 IJAIKD is currently Indexed By  Journal Seek