Correctness, transparency and effectiveness are the principal attributes of knowledge derived from databases using data mining. In the current data mining research there is a focus on efficiency improvement of algorithms for knowledge discovery. However, improving the algorithms is often not sufficient. The limitations of data mining can only be dissolved by the integration of knowledge of experts in the field, encoded in some accessible way, with knowledge derived from patterns in the databases. In this paper we discuss an approach for combining expert knowledge and knowledge derived from transactional databases. The approach proposed is applicable to a wide variety of risk management problems. We illustrate the approach with a case study on fraud detection in an insurance company. The case clearly shows that the combination of expert knowledge with monotomic neural networks leads to significant performance improvements.
Daniels, Hennie and Dissel, Han van, "Risk Management Based on Expert Rules and Data Mining: A Case Study in Insurance" (2002). ECIS 2002 Proceedings. 30.