Anomaly Detection of Hospital Claim Using Support Vector Regression

Abstract

BPJS Kesehatan plays a crucial role in providing affordable access to healthcare services and reducing individual financial burdens. However, deficit issues can disrupt the sustainability of the program, making anomaly detection highly important to conduct. Previous research on unsupervised anomaly detection in BPJS Kesehatan revealed a limitation with Simple Linear Regression (SLR), which only accommodates linear relationships among independent variables and the target variable of BPJS Kesehatan claim values. Minister of Health Regulation No. 52 of 2016 identified eight influential non-linear independent variables, leading to the proposal of Support Vector Regression (SVR) to address SLR's shortcomings.Research findings demonstrate SVR's superior anomaly detection performance over SLR. Interestingly, the SVR model excels in anomaly detection but lacks in prediction. Optimal tuning of SVR hyperparameters (C=9, epsilon=90, gamma=0.009, residual anomaly definition > 0.5*RMSE for both datasets) yields impressive metrics: Accuracy=0.97, Precision=0.84, Recall=0.97, and F1-Score=0.90. The anomaly detection results are expected to greatly support the sustainability of the BPJS Kesehatan program in Indonesia.