| Popis: |
Fraud is burgeoning throughout the world, making it necessary for institutions and payment processors to deploy computer-based fraud detection systems. In transactional fraud detection, such systems detect fraud in two principal ways: a) transactions may be processed by a series of user-specified rules that generate fraud alerts; and/or b) they are processed by some kind of intelligent system that automatically generates a fraud score for each transaction. While smaller institutions may find that a purely rules-based approach is sufficient for their needs, larger institutions commonly adopt a combination of both user rules and automated, intelligent fraud detection. Most well-known, high-end fraud detection systems offer both rules and intelligent capabilities. Fraud is burgeoning throughout the world. The evidence seems to suggest that modern Bayesian methods are superior to their neural network counterparts when detecting fraud patterns. However, many of the larger and more traditional financial institutions are still clinging to neural network-based methods, resisting change. In some cases this exposes customers to a greater risk of fraud. Mike Alford of Alaric International, compares the two approaches to see if the Bayesian approach might help to redress the balance in producing effective and intelligent fraud detection systems. |
Full Text from ScienceDirect