eyeDES Solution resulted in More Fraud Hits with Fewer Fraud Alerts

CHALLENGE

A bank facing growing losses from credit card fraudulent transactions decided to incorporate a more efficient solution in a cost-effective manner.

The bank had an extensive portfolio in the credit card (CC) market, having hundreds of thousands of CC transactions occurring each day. Even having a very small percentage, the fraudulent transactions were resulting in large amounts of losses every year. In the attempt to reduce the losses due to fraud, the bank had been using a market established fraud detection system. The system had a rule based component which used  some knowledge about the typical types of pre-fraud or fraud transactions and/or some information on stolen or copied cards. The second component was the statistical component where the personal profiles of card holders were determined and the deviations from their profiles were evaluated to determine the possibility of fraud.

SOLUTION

eyeDES solution was tested for its performance as a statistical component. For this test, our team had access to the full bank card data mart containing 279 variables. The number of variables selected from the data mart was 72 where some of them were based on other variable combinations. After applying feature reduction and selection techniques our team ended up with 17 variables to be used in our eyeDES model. Using this strategy, eyeDES solution increased the detection rate of the fraud transactions that were identified as fraudulent and, at the same time, decreased the number of total fraud alerts produced daily.

RESULTS

IDES Technologies' team worked on fine-tuning eyeDES model parameters and ended up with twice the number of identified frauds with even five per cent less number of alerts. Our solution doubled the fraud hit rate  and reduced the number of fraud alerts. Based on these results, eyeDES solution was chosen to replace the statistical component that was used by the bank in their credit card fraud management system.