Comparison of a probabilistic network and an expert system

The authors present a medical application of a belief network. Belief networks are direct acyclic graphs which can be used to represent knowledge of a domain expert. In the medical application network (diagnosis of severe infectious diseases), the strengths of the graph links were defined as conditional probabilities. These probabilities are estimated from statistical studies of the clinical database for the SES expert system, where 1400 cases are recorded. The results of this belief network are compared to those for SES.<<ETX>>