A Comparative Machine Learning Analysis of Hospital Energy Use Intensity and Its Key Drivers

Among healthcare facilities, hospitals are particularly energy-intensive, consuming substantially more energy per unit area than many other commercial buildings. Most existing studies on hospital energy use focus on a single facility and short-term load forecasting. This study evaluates annual site energy use intensity (EUI) across 103 U.S. hospitals using operational, structural, and climate-related variables. Several statistical and machine learning approaches were compared, including OLS, Ridge, Lasso, Decision Tree, Random Forest, XGBoost, CatBoost, and LightGBM. Model performance was evaluated using train–test analysis and cross-validation, and feature importance was examined to identify the features most associated with energy intensity. Overall predictive performance was limited, reflecting the relatively small and heterogeneous dataset. The models consistently identified two highly correlated variables, number of beds and inpatient days, as influential predictors. Interpreted together, they indicate that hospital capacity and operational activity play an important role in explaining differences in energy use. Structural and geographical variables generally showed lower importance in prediction. These findings support the use of machine learning methods as benchmarking and screening tools, while also highlighting the need for larger and more detailed national datasets to improve hospital energy prediction and comparison.