A Method of Diagnosis Using Possibility Theory
Kôichi Yamada, Mitsuhiro Honda · Industrial and Engineering Applications of Artificial Intelligence and Expert Systems · 2022
The paper proposes a diagnostic method using possibility theory. According to possibility theory, fuzzy set A on a set of possible causes U, B on possible symptoms V, and a fuzzy relation R on UxV could be understood as possibility distributions π u (u) 1 , π v (v) 1 and a conditional possibility distribution π v |u( v 1 | u 1 , respectively. In this paper, we define diagnosis as a problem to obtain possibility that a given crisp subset of U is the causes of a crisp subset of V observed as symptoms. Possibility distribution πon U and conditional possibility distribution π v |U v jπ u i are given as a priori knowledge. Then, a way to solve the problem is discussed and proposed. We also discuss a conventional approach using probability, and designate the difficulties that it suffers from, but our approach does not.