Using Rough Set Theory for Detecting the Interaction Terms in a Generalized Logit Model

Although logit model has been a popular statistical tool for classification problems it is hard to determine interaction terms in the logit model because of the NP-hard problem in searching all sample space. In this paper, we provide another viewpoint to consider interaction effects based on information granulation. We reduce the sample space of interaction effects using decision rules in rough set theory, and then use the procedure of stepwise selection method is used to select the significant interaction effects. Based on our results, the interaction terms are significant and the logit model with interaction terms performs better than other two models.