Christophe Labreuche (Thales)
Multi-Criteria Decision Aid (MCDA) aims at helping an individual to make choices among alternatives described by several criteria. Criteria are usually conflicting in the sense that it is not possible to maximize all of them simultaneously. A multi-criteria approach is classically decomposed into the following parts:
(1) Structuring phase and choice of the decision model. The aim is to identify the relevant criteria and attributes, and then select a decision model.
(2) Elicitation of the decision model, and then Generation of the best option. In a constraint approach, from a set of learning data (representing for instance comparisons of alternatives), one then looks for the value of the model parameters compatible with the learning data, which maximizes some functional, e.g. an entropy or a separation variable on the learning data. The comparisons among alternatives are then obtained by applying the model with the previously constructed parameters. As the set of model parameters are usually not uniquely set from the learning data, one can try to handle such imprecision in a more cautious way. One can cite the use of robust preference relations or min-regret approaches.
(3) Explanation of the results. A trend in Artificial Intelligence and other domains is that decision support is not a black box but rather should explain its recommendation. The difficulty is to explain the result of decision models that are more and more elaborate. We will review the most well-known approaches in these three steps.