On learning of weights through preferences

We present a method to learn the criteria weights in multi-criteria decision making (MCDM) by applying emerging learning-to-rank machine learning techniques. Given the pairwise preferences by a decision maker (DM), we learn the weights that the DM attaches to the multiple criteria, characterizing each alternative. As the training information, our method requires the pairwise preferences of alternatives, as revealed by the DM. Once, the DM’s decision model is learnt in terms of the criteria weights, it can be applied to predict his choices for any new set of alternatives. The empirical validation of the proposed approach is done on a collection of 12 standard datasets. The accuracy values are compared with those obtained for the state-of-the-art methods such as ranking-SVM and TOPSIS.