Combating fraud detection without getting lost in a sea of transactions


You may have heard a lot in the news recently about novel algorithms for automatic fraud detection. Yet fraud detection in financial institutions is more than just application of an algorithm. Rather, it requires an enterprise-grade architecture that offers high speed, security and governance. IBM Safer Payments helps banks with a smarter way to battle fraud.

But what is the smarter way to counter fraud? Consider credit card transaction fraud. Suppose a midsize bank handles as many as 2,000 transactions per second during peak hours and millions of transactions per day, and suppose its analytical fraud model is 99 percent accurate at detecting fraud. That accuracy might seem high, but think again: 99 percent accuracy means 1 percent inaccuracy. Of the millions of transactions handled each day by this bank, 1 percent inaccurate detection could translate to up to one million transactions failing to be classified as fraudulent activity.

IBM Safer Payments can help with fraud detection capabilities. It scores transactions against a large and growing library of known fraud patterns. In addition, it allows the creation of detection mechanisms based on machine learning techniques informed by cases of identified fraud.

Although some machine learning systems are enigmas, merely outputting a score—and are thus difficult for operators to understand and manipulate—IBM Safer Payments is designed to be fully transparent. Accordingly, it acts as a virtual analyst, interactively suggesting countermeasures for the benefit of human experts. In addition, it runs simulations on past data to help it anticipate the workload likely to be incurred by each new rule. Detection patterns that produce workloads managed by the fraud investigation team become part of the scoring library. In this way, IBM Safer Payments can create methods of fraud detection that avoid false positives while keeping workloads manageable for smarter data security.

Follow @IBMAnalytics

This entry was posted in Big Data. Bookmark the permalink.