A method and system for progressive learning of expert experience knowledge in a single classification domain through the analysis of new cases is disclosed. Based on the pre-built ripple-down rules (RDR) knowledge base, it is evaluated whether the conclusion of the new case corresponds to a false conclusion inconsistent with the knowledge of the RDR knowledge base. Among the new cases evaluated, cases evaluated as false conclusions are collected. Pre-determined knowledge of the RDR knowledge base that caused the collected cases to be evaluated as false conclusions is traced. New knowledge is extracted from the collected cases evaluated as false conclusions and the predetermined set of knowledge that caused the cases to be evaluated as false conclusions. The extracted new knowledge is reflected in the RDR knowledge base. Even after the initial RDR knowledge base is established, the machine learning system can automatically extract the expert's experience knowledge from a large number of new cases, and automatically add the relevant experience knowledge to the appropriate location in the RDR knowledge base.
展开▼