Cluster Analysis of BPJS Kesehatan Claim Data in Madiun City to Identify High Claim Patterns and Fraud Indications

Abstract

The increasing number of BPJS Kesehatan (Health Social Security) service claims in Madiun City poses significant challenges to financing efficiency and raises concerns about potential irregular or fraudulent claims. This study aims to identify high-claim patterns and detect indications of fraud using a data mining approach through the K-Means and Hierarchical Clustering methods. The research employed secondary data consisting of 309 hospital claim records from Madiun City in 2025. The primary variables were the number of claims and total claim costs, supported by additional variables such as age, gender, occupation, type of service, and disease diagnosis. Data analysis involved three main stages: preprocessing, clustering, and cluster quality evaluation using the Silhouette Score, Davies-Bouldin Index, and Calinski-Harabasz Index. Ths Study further compared the performance of both clustering methods, revealing that K-Means achieved superior validity scores across major evaluation metrics. The K-Means method produced the best performance, with a Silhouette Score of 0.617 and a Calinski-Harabasz Index of 419.581, reflecting well-separated and compact cluster structures. Three main clusters were identified-low, medium, and high. The high-claim cluster consisted of participants aged 55 years and above, with a claim frequencies of 2 to 7 claims and total claim costs exceeding IDR 20 million. This cluster was dominated by retirees, housewives, and private-sector employees utilizing inpatient services. Although categorized as a high-risk group, verification results revealed no signs of fraud but rather complex medical needs. These findings suggest that integrating clustering analysis into BPJS Kesehatan’s claim monitoring system can support early anomaly detection and enhance both financing efficiency and claim management integrity.