A Methaheuristic approach for preference Learning in multi criteria ranking based on reference points.

International audienceIn this paper, we are interested in a family of multi-criteria ranking methods called Ranking with Multiple referencePoints (RMP). This method is based on pairwise comparisons, butinstead of directly comparing any pair of alternatives, it comparesrather the alternatives to a set of predefined reference points. Weactually focus on a Simplified RMP model (S-RMP) in which thepreference parameters include the criteria weights and the set of ref-erence points ordered by importance. Elicitation of the parameters(from the data provided by the decision makers) leads us to the pref-erence learning algorithms that cannot only be applied on relativelysmall dataset. Therefore, we propose in this work a preference learn-ing methodology for learning S-RMP models from a large set of pair-wise comparisons of alternatives. The newly proposed algorithm isa combination of an Evolutionary Algorithm and a Linear Program-ming approach. Empirical results and analysis are also presented