Designing Bayesian networks for highly expert-involved problem diagnosis domains

Abstract

Systems for diagnosing problems in highly complicated problem domains have been traditionally very difficult to design. Such problem diagnosis systems have often been restricted to the use of primarily rule-based methods for problem diagnosis in cases where machine learning for probabilistic methods has been made difficult by limited available training data. The probabilistic diagnostic methods that do not require a substantial amount of available training data usually require considerable expert involvement in design. This thesis proposes a model which balances the amount of expert involvement needed and the complexity of design in cases where training data for machine learning is limited. This model aims to use a variety of techniques and methods to translate, and augment, experts' quantitative knowledge of their problem diagnosis domain into quantitative parameters for a Bayesian network model which can be used to design effective and efficient problem diagnosis systems.