Preferences-based learning of multinomial logit model

Abstract

We learn the parameters of the popular multinomial logit model to gain insights about a DM’s decision process. We accomplish this objective through the recent algorithmic advances in the emerging field of preference learning. The empirical evaluation of the proposed approach is performed on a set of 12 publicly available benchmark datasets. First experimental results suggest that our approach is not only intuitively appealing, but also competitive to state-of-the-art preference learning methods in terms of the prediction accuracy.