Induction of expert system rules based on rough sets and resampling methods

Automated knowledge acquisition is an important research issue in developing medical expert systems. While several methods of symbolic inductive learning have been proposed, most of the approaches focus on inducing some rules to classify cases correctly. On the contrary, medical experts also learn other information important for medical diagnostic procedures from clinical cases. In order to acquire both kinds of knowledge, we developed a program that extracts not only classification rules for differential diagnosis, but also other medical knowledge needed for diagnosis. This system is based on a diagnosing model of a medical expert system RHINOS, which diagnoses causes of headache and facial pain. We apply this program to the same domain and compared the induced results with expert rules. The results show that the combination of a rule induction method with resampling methods is effective to estimate the performance of induced results, especially when only small training samples are available without domain knowledge.