Predicting Building Primary Energy Use Based on Machine Learning: Evidence from Portland

Accurately predicting equivalent primary energy use (EPEU) in buildings is crucial for advancing energy-efficient design, optimizing operational strategies, and achieving sustainability goals in the built environment. This study aims to develop reliable prediction models for EPEU by leveraging a comprehensive and high-quality dataset from buildings in Portland, USA. To achieve this, a systematic machine learning framework is adopted, encompassing feature selection, data preprocessing, model training, and performance evaluation. Several state-of-the-art machine learning algorithms are applied, including Random Forest (RF), Gradient Boosting Decision Tree (GBDT), and Back-Propagation Neural Networks (BP). These models are trained using key features such as building type, gross floor area, construction year, and various operational characteristics that are known to significantly influence energy consumption patterns. The dataset is carefully cleaned and normalized to ensure model generalizability and minimize bias. Model performance is assessed using standard statistical metrics, including the coefficient of determination (R²), Mean Absolute Error (MAE), and Root Mean Squared Error (RMSE). Among the tested models, ensemble learning methods—particularly RF and GBDT—consistently outperform others in terms of prediction accuracy, robustness, and stability across different building types. The results of this study not only highlight the potential of machine learning in energy prediction tasks but also provide actionable insights for architects, engineers, facility managers, and policymakers. By identifying the most influential variables and employing effective predictive models, this research supports data-driven decision-making processes aimed at improving building energy performance.