Title : KNOWLEDGE DISCOVERY WITH AVERAGE COMPRESSED ENTROPY

Text: Recently, Minagawa proposed a knowledge refinement method for crack diagnostic expert system. The inference engine was constructed for a reciprocal network based on min-max composition algorithm. The knowledge refinement function was installed into the engine by using the concept of back propagation algorithm. First the inference and refinement method are applied to the rule-base system for selecting the retrofitting method of steel bridges damaged by fatigue. It is confirmed that the inference engine can be used for any particular domain. Second, for the purpose of knowledge discovery, we evaluated average compressed entropies for a case-base virtually constructed through some inferences with the inference system that we proposed for selecting the retrofitting method. It is found from the analyses that the average compressed entropy is an effective measure for the discovery of knowledge that is implicitly buried into databases or case-bases.