A disaggregation approach for indirect preference elicitation in Electre TRI-nC: Application and validation

Abstract

Multicriteria sorting methods are often used in decision aiding contexts where the objective is to assign alternatives to predefined ordered categories. The Electre Tri family of sorting methods is based on pairwise comparisons of the alternatives with some, possibly fictional, alternatives that are either upper or lower limits of the categories (Electre Tri-B), or one or more typical reference alternatives, that is, representative categories profiles (Electre Tri-C, Tri-nC). In this paper, we are interested in the Electre Tri-nC method and in indirect preference elicitation based on partial information provided by the Decision Maker. We therefore propose, apply and evaluate a preference disaggregation method for learning criteria weights and the credibility threshold used in Electre Tri-nC. The proposed disaggregation method is validated in an experiment using a climate classification problem for light tourism where 62,482 touristic locations are sorted into four categories. A robustness analysis of the method's performance using 150 learning sets is conducted and the results are presented and discussed.