Colin Shearer, VP Product Marketing, SPSS You Asked How can data mining be used as a tool to prevent fraud protection?

In many applications of data mining – for example, deciding who to target with marketing offers – it’s straightforward to decide how to approach the problem. There are reasonably large numbers of past cases we can look at and say, “Yes, she bought this when offered,” or “No, he didn’t buy”, and build models that tell buyers from non-buyers. In fraud, the situation is different.

Often, at the start of analysis, fraud is simply suspected, rather than known for sure – there are no confirmed historical cases to learn from. And the occurrence may be very low – the needles are few and the haystacks are large.

The first approach is usually “anomaly detection” – using data mining algorithms which identify cases that are different from most of the others, often in ways which are far too subtle to be detected manually. These “anomalies” are just that – cases which are unusual, but are not necessarily bad. Human follow up is essential – expert investigators study the anomalies, and decide which are truly fraudulent.

Once these genuine fraud cases have been identified, the “learning” capability of data mining comes into play – building models that ensure any similar type of fraud in future is detected automatically. Fraudsters constantly change their approaches to try to stay undetected, so anomaly detection to find new, emerging patterns is always important, with predictive models being built to recognize known fraud types.

Fraud detection models can give great value when deployed to the “front-line” – intercepting attempted fraud as it happens. For example, some insurance companies are using models which assess the “risk score” of a claim as it is entered, e.g,. in call centers. This not only means that more fraudulent claims can be detected (and at an earlier stage), but also that low-risk claims can be “fast-tracked." This leads to better customer satisfaction, and reduces claim handling costs by streamlining the process for “safe” claims.

SPSS