Discovering Potential Internal Fraud Models in a Stream of Banking Transactions

摘要

Internal frauds in the banking industry cause huge costs to banks and this problem is particularly difficult to address since swindlers are very fanciful and thus constantly make fraud schemata evolve. Fraud detection systems are therefore difficult to design because they must learn continuously from the new fraud models. As a consequence, the detection and modeling of the internal fraud are mainly made by hand, entailing late detection and long delays for the modeling phase. This paper proposes a new theoretical and practical approach for designing fraud detection systems able to detect potentially fraudulent transactions and discover the corresponding fraud pattern used by the swindlers. This approach is based on a particular method of abstraction that reduces the complexity of the problem from O(n~2) to O(n), n being transactions number, making it possible to implement it on a mere professional personal computer as a Java program that works in real time and on-line. The paper describes this approach and its results on real-world internal fraud detection, analyzing a transaction database provided by a worldwide French bank.