Development of Machine Learning Models to Predict Health Insurance Claim Costs Among Older Indonesians: A Retrospective Predictive Modeling Study

Objectives The objective of this study was to develop machine learning models to predict health insurance claim costs among older adults in Indonesia.

Methods This study utilized secondary data from the Indonesian National Health Insurance program (Jaminan Kesehatan Nasional [JKN]) spanning 2017 to 2023. Three modeling techniques-linear regression, random forest, and XGBoost-were employed to predict individual claim costs. Model performance was assessed using the root mean square error (RMSE), coefficient of determination (R2), and mean absolute error (MAE). Additionally, variable importance analysis was conducted to identify key predictors.

Results XGBoost with 500 boosting rounds yielded the best performance, with an RMSE of 11,360,283, an R2 of 0.81, and an MAE of 4,485,917, outperforming both linear regression (RMSE, 13,710,035; R2=0.72) and random forest (RMSE, 12,434,238; R2=0.78). Notably, outpatient care was identified as the most consistent predictor across all models. Other significant predictors included length of stay (LOS), diagnosis type (International Classification of Diseases, 10th Revision chapter), facility type, facility classification, and severity of illness, particularly for moderate cases. Although LOS and diagnosis type were important predictors, these findings should be interpreted in the context of Indonesia's fixed Indonesian Case-Based Groups payment system.

Conclusions XGBoost provides reliable predictions of claim costs among older adults, capturing clinical, utilization, and structural drivers. These findings can inform targeted interventions, improve chronic disease management, optimize the referral system, and support integration of predictive tools into JKN to enhance responsiveness and promote sustainable, equitable financing.