Case mining from large databases

This paper presents an approach of case mining to automatically discover case bases from large datasets in order to improve both the speed and the quality of case based reasoning. Case mining constructs a case base from a large raw dataset with an objective to improve the case-base reasoning systems' efficiency and quality. Our approach starts from a raw database of objects with class attributes together with a historical database of past action sequences on these objects. The object databases can be customer records and the historical action logs can be the technical advises given to the customers to solve their problems. Our goal is to discover effective and highly representative problem descriptions associated with solution plans that accomplish their tasks. To maintain efficiency of computation, data mining methods are employed in the process of composing the case base. We motivate the application of the case mining model using a financial application example, and demonstrate the effectiveness of the model using both real and simulated datasets