Interactive anomaly detection in large transaction history databases

Abstract

The scale of financial sector crime today makes the detection of anomalous financial flows into, out of and within, a nation one of the most important functions of modern government. The analysis necessary for detection of such criminal activity depends on the existence of a central IT infrastructure capable of maintaining historical transaction records and capable of enabling the application of advanced analysis techniques to large data volumes.

We describe a software tool developed to aid the rapid, error-free transformation of data held in aggregated transaction history databases into matrices for analysis by fraud detection experts. We also present some initial results of performance characterisation studies which will provide the basis for guidelines on how transformations can be tuned to make best use of underlying parallel database systems.