Using mixed integer optimisation to select variables for a store choice model

Abstract

This paper develops a store choice model to investigate consumers' store choice behaviour through the use of actual scanner panel data. For this purpose, we use the variable selection method proposed by Sato et al. 2015, which is based on mixed integer optimisation for logistic regression. Computational results demonstrated that when the Akaike information criterion is used as a goodness-of-fit measure, our approach yields a predictive performance better than that obtained using the common stepwise method of variable selection. Moreover, we clarified store choice factors by analysing the results of variable selection.