Modeling Decision-Maker Preferences through Utility Function Level Sets

摘要

In this paper, we present a method based on the multiat-tribute utility theory to approximate the decision-maker preference function. A feature of the proposed methodology is its ability to represent arbitrary preference functions, including functions in which there are non-linear dependencies among different decision criteria. The preference information extracted from the decision-maker involves ordinal description only, and is structured using a partial ranking procedure. An artificial neural network is constructed to approximate the decision-maker preferences, reproducing the level sets of the underlying utility function. The proposed procedure can be useful when recurrent decisions are to be performed, with the same decision-maker over different sets of alternatives. It is shown here that the inclusion/exclusion of information causes only local rank reversals instead of large scale ones that may occur in several existing methodologies. The proposed method is also robust to relatively large levels of wrong answers of the decision maker.