Enhancing Medical Expenditure Prediction Using Machine Learning on Claims Data for Better Healthcare Cost Management

The prospective disease group-based payment system, implemented in frameworks such as APR-DRG, seeks to standardize claims management but encounters several limitations. This study aims to enhance the accuracy of health insurance reimbursement claims by leveraging health claims data and machine learning techniques, specifically Extreme Gradient Boosting (XGBoost), Random Forest, and Linear Regression. The research identifies that the precision of health cost claims utilizing the APR-DRG coding system and CCS diagnosis codes can be significantly improved. By incorporating additional variables from claims data, such as demographics, diagnoses, and facility utilization, the study develops a robust predictive model for medical expenditures. The findings demonstrate that both XGBoost and Random Forest algorithms outperform traditional linear regression, providing high accuracy in predicting inpatient costs. This advancement has the potential to improve health cost management and reduce discrepancies in reimbursement processes. Future research should expand the predictive variables to include comorbidities and explore other coding systems, such as INA-CBG, to further enhance the accuracy of healthcare cost predictions in diverse contexts.