Understanding and predicting the decision making behaviour of individuals is a
subject of interest for marketers, strategists, economists and the computer scientists
alike. We develop an aproach to learn a decision maker (DM)'s behavioral process
by combining recent possibilistic discrete choice models with the emerging machine
learning methods. The proposed approach considers the utility values derived by a
DM from each of the attribute values (information source values). We take the
training information in the form of the exemplary multi-attribute preferences, and
the decision model is specified in terms of two vectors that are unique to a DM. The
experimental results on a set of 10 benchmark datasets suggest that our approach is
both intuitively appealing and competitive to state-of-the-art methods in terms of the
prediction accuracy.