There has been a recent increasing trend in fraudulent insurance claims. To detect fraudulent claims insurers use statistical predictive modeling. However, the fraudulent-claim data are complex, and the current standard regression methods are not powerful enough to uncover the “knotty” interrelationships among the fraudulent-claim data. The model builder needs a cutting-edge data mining technique to extract the quintessence of fraudulent-claim data to insure an accurate classification fraudulent claims model can be built to produce fraud propensity scores, which the insurance adjuster would then use to target the criminal claimants.

The purpose of this article is to present a machine learning approach – the GenIQ Model© – that has the data mining muscle for untying the knotty fraudulent-claims data to secure building an effective fraudulent claims model. The GenIQ Model is a machine learning alternative model to the statistical regression model that lets the data define the model – automatically data mines for new variables, performs variable selection, and then specifies the model equation. GenIQ, exclusive of statistical regression’s imposed data restrictions, assumptions, and pre-specified form, clearly offers a superior approach for identifying and understanding the key drivers of fraudulent claimants.