We propose an approach to learn a multiattribute utility function to model, explain or predict the value system of a Decision Maker.  More precisely, we consider the generalized additive decomposable utility model which allows interactions between attributes while preserving some additive decomposability of the evaluation model. We present a supervised learning approach able to identify the factors of interacting attributes and to learn the utility functions defined on these factors from a set of examples. The aim is to keep the model as simple as possible, while incorporating interactions when necessary to fit the observed preferences. The proposed approach does not make any prior restriction on the size of interactions. It relies on the determination of a sparse representation of the ANOVA decomposition of the multiattribute utility function using multiple kernel learning. It applies to both continuous and discrete attributes. We will report the results of numerical tests showing the practical efficiency of the proposed approach. 

(joint work with Margot Hérin and Nataliya Sokolovska) 

Retour page d'accueil du séminaire