Chapter 8: NetFraud: A Case Study


8.1 Fraud Detection in Real Time

Credit-card fraud detection is an extremely difficult problem to solve. Several factors compound the difficulty of this criminal pattern—recognition problem. First of all, the data provided in an authorization is extremely limited, such as the amount, location, and perhaps type of product or service being purchased. Second, the patterns of fraudulent use are very diverse. A person can use a credit card at millions of different places and thousands of Web sites, making it extremely difficult to match a pattern. Third, a thief can be using a credit-card number concurrently with an authorized user, making the pattern detection more challenging. A fraud detection model must determine a way to classify transactions made by the authorized party and those made by a non-authorized party. Finally, there is the issue of time and the anonymity of the thief. If he or she can get away with charging a few hundred dollars in a few days on a stolen card number, chances are the thief will never be tracked down. Therefore, fraud detection has to be done in real-time.

At a Web site, real-time fraud detection has to be done right at the virtual checkout counter, when the e-commerce site sends the information to Visa, which can be through several channels, and then as it arrives at the bank that issued the card. Before the bank sends an authorization back to the merchant through Visa, it will run it through a real time detection system. These fraud detection systems are almost always based on a combination of neural networks and rules, most which banks today outsource to HNC, a manufacturer of fraud detection software. HNC has developed neural network-based decision systems over the years because they provide a faster, statically robust, less subjective way to assess certain kinds of business risk.

Most banks have issued millions of credit cards, so using a non-neural network would be too slow for processing such a large database in real time. While real-time fraud detection can keep a thief from making one more illegal purchase on a card, it took several years for these systems to be effective. One of the early problems banks had with fraud-detection systems was that they were plagued with false positives, that is, situations where fraud is suspected on a legitimate transaction. This was expensive for banks, since they would use costly call centers to phone consumers to confirm questionable transactions. However, to the consumer, nothing could be more annoying than having a card declined simply because a transaction did not fit the profile created and maintained by their bank.




Investigative Data Mining for Security and Criminal Detection
Investigative Data Mining for Security and Criminal Detection
ISBN: 0750676132
EAN: 2147483647
Year: 2005
Pages: 232
Authors: Jesus Mena

flylib.com © 2008-2017.
If you may any questions please contact us: flylib@qtcs.net