Domain-Knowledge Based Feature Engineering in Fraud Detection Using Health Administrative Claims

Healthcare insurance fraud is the intentional act of deception to gain financial benefit. Insurance fraud is a serious crime, with health fraud being one of the most prevalent. It affects numerous victims, primarily by increasing costs. Healthcare fraud consists of various schemes. The complexity of health administrative data in Indonesia, such as that managed by BPJS, poses a major challenge in fraud detection. A domain-knowledge-based feature engineering method was developed to detect potential fraud in healthcare insurance claims using administrative health data. Feature engineering holds significant potential for identifying fraudulent transactions in real-world healthcare administrative claims datasets. Evaluation of 4 outlier detection models and 4 featureset segregated by fraud scheme resulting in DBSCAN as the best model. Furthermore the best approach to improve detection quality is to use featureset that are segregated by fraud scheme.