Reducing False Positives in Fraud Prevention and Detection

To reduce the number of false positives in present fraud detection systems this project will focus on developing robust analytical techniques with tunable filters capable of focusing on high probability fraudulent events. We will develop visualization tools that give a more holistic understanding of fraud and fraud patterns and can be used to anticipate fraudulent exploits rather than to react to them. Present fraud detection systems use rule-based techniques. While these can reduce an organization’s attack surface, more sophisticated and computationally intelligent approaches are necessary to avoid the false positive dilemma. Aggregate behavior models, which identify hidden relationships between people, organizations, and events will be employed.

This project will develop (1) advanced analytical techniques to reduce the number of false positives, while preserving high probability fraud events, (2) visualization techniques that provide human operators a holistic view of fraud patterns to further improve the algorithms e.g. by using aggregate behavior models.