A referential scheme of fuzzy decision making and its neural network structure

The author introduces a method for dealing with imprecise objectives involved in the process of decision-making. A three-stage form of the system is proposed. It comprises three basic functional components realizing matching, nonlinear transformation, and inverse matching. The proposed scheme has a referential structure which shows that the fuzzy set of a decision is not determined by the objectives themselves, but by the levels of the matching with some prototype decision situations. Both matching and inverse matching procedures involve some logic-based mechanisms (equality indices). Neural nets are used to realize the nonlinear mapping indicated in the general scheme. Several advantages of the referential model, including exhaustive usage of knowledge about the decision problem conveyed by prototype situations, and an introduction of mechanisms of evaluation of the relevancy of fuzzy decisions, are highlighted. Additional indices expressing consistency of decision scenarios are developed. Detailed numerical studies demonstrate the performance of the method and provide some additional background concerning an evaluation of the results