Clustered knowledge representation: increasing the reliability of computerized expert systems.

Accuracy and reliability are major concerns for the developers of expert systems. We have developed a Bayesian expert system for medicine, Iliad. We have included in Iliad new types of decision frames, called clusters. Clusters are frames (usually Boolean) that contain groups of conditionally dependent findings. These groups often describe pathophysiologic concepts. Because clusters encapsulate groups of conditionally dependent findings, we expected clusters might increase Iliad's accuracy and reliability. We tested this hypothesis by measuring the reliability of pairs of clustered and nonclustered frames using real patient data.