Outlier Detection in Inpatient Claims Using DBSCAN and K-Means

Health insurance helps people to obtain quality and affordable health services. The claim billing process is manually input code to the system, this can lack of errors and be suspected of being fraudulent. Claims suspected of fraud are traced manually to find incorrect inputs. The increasing volume of claims causes a decrease in the accuracy of tracing claims suspected of fraud and consumes time and energy. As an effort to prevent and reduce the occurrence of fraud, this study aims to determine the pattern of data on the occurrence of fraud based on the formation of data groupings. Data was prepared by combining claims for inpatient bills and patient bills from hospitals in 2020. Two methods were used in this study to form clusters, DBSCAN and KMeans. To find out the outliers in the cluster, Local Outlier Factor (LOF) was added. The results from experiments show that both methods can detect outlier data and distribute outlier data in the formed cluster. Variable that high effect becomes data outlier is the length of stay, claims code, and condition of patient when discharged from the hospital. Accuracy K-Means is 0.391, 0.003 higher than DBSCAN, which is 0.389.