Semi-Supervised Medical Insurance Fraud Detection by Predicting Indirect Reductions Rate using Machine Learning Generalization Capability

Abstract

There is 10% fraud in medical insurance based on published statistics in Insurance Research Institute of Islamic Republic of Iran in 1399 –solar system eq. 2020 in the Gregorian calendar-which cost about 28 thousand billion RIALs –the official currency of Iran eq. to about 320 million dollars-. This study proposes a machine learning-based technique to predict the claim cost based on other patients’ history and predict fraud or abnormal costs in claims that significantly differ from other claims. Besides, a new data sampling approach is proposed to lead the machine learning algorithms that focus on exceptional cases. A real-world private dataset is used to evaluate 700,000 claims of the RASA web portal, used for supplementary insurance by famous companies like Day. The proposed data sampling approach reduced absolute error in exceptional cases from 35 to 23 errors for deduction rate. The evaluation results show about 0.5% of abnormal cases in the dataset with a higher than 20% absolute error. The abnormal rates can be adjusted to a lower or higher range.