Insurance Fraud Detection

Goal: Reduce Fradulent Death Claims

A major insurance company is concerned with finding patterns in death claims, which occur within and shortly after the two years contestability period.  During this time frame, the insurance company has the right to contest benefit payments.  Therefore, it is imperative that fraudulent claims be detected in a timely fashion.

The company database consists of over 100,000 policies.  A traditional auditing tool was first used as a preprocessor to manipulate the files and sort the data according to date clusters.  The contestability date was selected and the period two years before (back to the issuing date) and two years after formed the total four years period of interest.  The data were sorted into six months clusters or a total of eight clusters over the four years and the records containing death claims were selected.  Each resulting cluster of interest consisted of an average of about 550 records.

Once the preliminary step was taken with a traditional auditing tool, WizRule was used to mine the data.  Each cluster of interest was analyzed by WizRule for patterns in the data and for deviations.  Since WizRule finds all the rules and relationships in the data without requiring pre-knowledge of the auditor, interesting and unexpected results were obtained.

Among the unusual patterns detected by WizRule were cases where the death occurred in a state other than either the state of issuance or the sate the decedent lived in during the effect of the policy.  In some cases, a pattern of foreign issued death certificates pointed to potential fraud.

These cases are currently under investigation by the appropriate agencies.