Identification of Possibility Distribution and Its Application to Discriminant Analysis

Abstract

This paper proposes an identification method of possibility distribution based on the data given by experts and leads to a possibility discriminant rule from the viewpoint of possibility distributions. Since the proposed method resorts to linear programming, possibility distributions defined as quadratic functions can be easily obtained. This method is used to construct fuzzy if-then rules. The possibility discriminant rule corresponds to Bayes discriminant rule based on probability distributions. To illustrate the proposed method, a discriminant problem is shown as a numerical example.