Optimizing Public Hospital Budgets Using Ensemble Machine Learning and SHAP Analysis for Interpretable Cost Prediction

Public hospitals are in a position of growing economic pressure, and frugal resource management is necessary. Unfortunately, most traditional cost forecasting models do not capture healthcare costs' dynamic and non-linear nature. This paper offers a financial optimization framework based on AI with Ensemble Machine learning techniques that are interpretable. This methodology identifies the data preprocessing, feature engineering, and model training with the optimized Random Forest and XGBoost algorithms and SHAP (Shapley Additive exPlanations) analysis for model interpretability. The results report that generating our optimized XGBoost model led to an R² score of 0.89, outperforming Random Forest (R² = 0.88) and our baseline models. It also achieved a Mean Absolute Error (MAE) of 2502.36 and a Mean Squared Error (MSE) of 11230456.12, which is very high in predictive accuracy. Interpretability is achieved using SHAP (Shapley Additive exPlanations) analysis, which identifies key cost-driving factors such as smoking status, BMI, and age, enabling more transparent and informed decision-making by stakeholders. With the framework, we present a scalable predictive budgeting and decision-making solution in public healthcare institutions