Detecting medical prescriptions suspected of fraud using an unsupervised data mining algorithm

Nowadays, health insurance companies face various types of fraud, like phantom billing, up-coding, and identity theft. Detecting such frauds is thus of vital importance to reduce and eliminate corresponding financial losses. We used an unsupervised data mining algorithm and implemented an outlier detection model to assist the experts in detecting medical prescriptions suspected of fraud. The implementation ran medicine code, patients’ sex, and patients’ age variables through three successive screening steps. The proposed model is capable of detecting 25% to 100% of cases violating the standards for some medicines that are not supposed to be prescribed at the same time in one single prescription. This model can also detect medical prescriptions suspected of fraud with a sensitivity of 62.16%, specificity of 55.11%, and accuracy of 57.2%. This paper shows that data mining can help detecting potential fraud cases in medical prescriptions more quickly and accurately than by the manual inspection as well as reducing the number of medical prescriptions to be checked which will result in reducing investigators heavy workload. The results of the proposed model can also help policymakers to plan for fighting against fraudulent activities. [Figure: see text]