An Explainable Machine Learning Framework for Healthcare Building Energy Demand Forecasting

This paper presents an explainable machine learning framework for short-term healthcare building energy demand forecasting using the ASHRAE Great Energy Predictor III dataset. Five forecasting approaches, including Random Forest, XGBoost, Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and a naïve persistence baseline, are systematically evaluated using multiple performance metrics. The study integrates SHAP-based interpretability, ablation analysis, cross-building validation, statistical significance testing, and residual diagnostics to provide a comprehensive assessment of model performance and reliability. Results show that ensemble learning methods, particularly Random Forest, achieve the highest forecasting accuracy, while feature importance analysis highlights recent energy consumption and building size as the dominant predictors. The findings demonstrate the value of explainable machine learning for transparent and reliable healthcare building energy forecasting, providing practical insights for intelligent building energy management.