07-03-2016 04:08 AM
Historically most banking-related fraud detection systems have used a rules-based approach. This works OK for simple fraud detection. For example finding international ATM transactions done with ATM cards that are exclusively used in one country.
However for more sophisticated types of fraud predictive modelling can detect fraud where simple rules wont. One example I had involvement with is the so-called credit card "bin attack" fraud. This is where the fraudster may have stolen credit card numbers or may try to generate genuine credit card numbers using software (tools for this can be found on the Internet). The fraudster then proceeds to test these card numbers on insecure online merchant sites using "tester" transactions. By cycling through card numbers in sequence in combination with expiry dates it is possible a find card numbers that work. Then the big transactions are done.
By analysing all these transactions using a predictive approach it is possible to identify these fraudulent transactions very quickly and take steps to prevent bigger fraud happening.