A simplified technique for using necessity-possibility measures

AbstractA new technique of uncertainty management in expert systems is proposed. It is suggested that the possibility theory-based treatment of uncertainty (developed by D. Dubois and H. Prade) in inference engines be simplified by using a combination of possibility and necessity measures, namely N + ∏ − 1, rather than using the possibility measure ∏ and the necessity measure N separately. This combination is denoted by V, and its range is [− 1, 1]. First it is explained how this measure V is derived from the original necessity-possibility measures through a one-to-one function, and the meaning of V is shown precisely. Then the paper describes the adaptation of the different uncertainty propagation and combination methods of the Dubois-Prade model to the “newly” defined measure. In this conversion the implicit behavior of each procedure is shown, along with its connection with other uncertainty resolution methods. Particularly important is the fact that none of the advantages of the Dubois-Prade model are lost, especially in the handling of fuzzy facts. This is so because at any given moment it is possible to work with either V or the necessity-possibility pair. The proposed uncertainty resolution method offers distinct advantages over previous models, such as certainty factors of MYCIN