The purpose of this study is to detect the intentional or unintentional abuse of insurance companies producing health insurance by insured and contracted health institutions in advance with machine learning. Although the insured do not have check-up coverage in group policies, it is a common type of abuse to have a free check-up independent of the insurance company and to reimburse the cost to the insurance company. The caseloads that take place between the insured persons of the group policies and the institutions have been examined as a decriminalization scenario in this context. For this abuse scenario, a data labeling study based on business knowledge and K-Means clustering method was conducted. KNIME and SPSS Modeler tools were used in the study. Random Tree (RT), Random Forest (RF), XGBoost (XGB), K Nearest Neighbor (KNN) models were developed using the labeled data set and the results were compared with Prediction, Recall, Specificity, F1 Score and Accuracy classification performance measures. It has been seen that the RF model works with higher performance than other models. As a result of the study, the abuses detected between the group policies and the health institutions were shared with the insurance companies, enabling them to improve the damage management.
Anahtar Kelimeler: Anomaly Detection, Fraud Prediction, Healthcare Insurance, Machine Learning