Building Bayesian Network Based Expert Systems From Rules

Combining expert knowledge and user explanation with automated reasoning in domains with uncertain information poses significant challenges in terms of representation and reasoning mechanisms. In particular, reasoning structures understandable and usable by humans are often different from the ones for automated reasoning and data mining systems.Rules are a convenient and human understandable way to express domain knowledge and build expert systems. Adding certainty factors to these rules presents one way to deal with uncertainty in rule based expert systems. However such systems have limitations in accurately modeling the domain. Bayesian Network, on the other hand, is a probabilistic graphical model that allows accurate modeling of a domain and automated reasoning. But inference in Bayesian Networks is harder for humans to comprehend.In this thesis, we propose a method to combine these two frameworks to build Bayesian Networks from rules and derive user understandable explanations in terms of these rules. Expert specified rules are augmented with strength parameters for antecedents and are used to derive probabilistic bounds for the Bayesian Network's conditional probability table. The partial structure constructed from the rules is fully learned from the data. The thesis also discusses methods for using the rules to provide user understandable explanations, identify incorrect rules, suggest new rules and perform incremental learning.