A New Uncertainty Calculus For Rule-Based Expert Systems

Abstract

The solution of non-deterministic expert systems consists of two components –the solution reached and a calculated measure of belief in each solution. This measure of belief is often the most critical factor in analyzing the solution. Unfortunately, as this paper reviews, the issue of how best to implement uncertainty calculi in expert systems has never been settled. Some popular rule-based approaches have in fact been shown to produce results no better than random guessing. To improve the accuracy of rule-based systems, we propose a new calculus we call gamma factors. This calculus combines ideas from two popular certainty factor calculi the product method, and the probability sum method. It includes a tuning mechanism which the expert can use in a rule pre-processing step to compensate for dependent parallel evidence combination.