Fraud Detection: From Basic Techniques to a Multi-Agent Approach

According to KPMG figures, fraud represents a serious economical problem, which has been studied in different ways due to the fact that fraudsters are benefiting from the fast development of ICT and are developing their techniques. In this paper, after summarizing different fraud detection methods and tools proposed in the literature like meta-rules and tree-based detection, we will introduce a multi-agent system, called FIDES, which integrates the computational power of data mining tools and attack trees with experts´ judgments negotiated through a Delphi-based system. Two scenarios are described: in the first one FIDES, supported by cause-effect diagrams, is used to classify alarms generated by the system to help the experts to focus on the real dangerous ones; in the second one FIDES is used in a proactive way in order to block or prevent human based frauds.