Quantitative Models for Reasoning Under Uncertainty in Knowledge-Based Expert Systems

The solutions to many real-world problems require reasoning under uncertainty (also known as inexact, approximate, or evidential reasoning). In knowledge-based expert systems, uncertainty can manifest itself in data and in the inference process. Well-known techniques for drawing inferences under uncertainty include Bayesian Probability Theory, Dempster-Shafer Theory of Belief Functions. MYCIN's model of Confirmation and Disconfirmation, and Zadeh's Fuzzy Set Theory. The quantitative nature of these four approaches allows us to represent various degrees of uncertainty. In this paper, we summarize the basic concepts of these techniques, apply them to solve the same, typical expert-system problem, and address the advantages and drawbacks of each approach. We also compare these four theories and provide some guidelines for selecting the appropriate techniques to model inexact reasoning in knowledge- based expert systems.