Energy use intensity analysis of office buildings using green BIM-integrated Interpretable machine learning

• Published in: Journal of Building Engineering Vol. 108; p. 112760
• Main Authors: Nguyen, Ngoc-Mai, Cao, Minh-Tu
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 15.08.2025
• Subjects: Building information modelling; Energy use intensity; Energy-efficient buildings; Ensemble model; Machine learning; Taiwan office buildings; Energy use intensity; Energy-efficient buildings; Ensemble model; Building information modelling; Taiwan office buildings; Machine learning
• ISSN: 2352-7102; 2352-7102
• DOI: 10.1016/j.jobe.2025.112760

Accurately predicting the energy use intensity (EUI) of office buildings is essential for optimizing energy efficiency and implementing sustainable building practices. This study integrated green building information modelling (green BIM) with advanced machine learning (ML) techniques to develop EUI prediction models tailored to Taiwanese office buildings. A data set of 1006 unique BIM models was generated using Green Building Studio, which calculated the EUI for diverse configurations of architectural and environmental variables. The forensic-based investigation algorithm was employed for hyperparameter optimization, and it considerably enhanced the performance of the developed boosting-based ensemble models. Of all models, XGBoost exhibited the highest accuracy and robustness, achieving the lowest root-mean-square error (37.98 ± 11.81 MJ/m2/year) and mean absolute percentage error (1.86 % ± 0.50 %) for the test set, thereby outperforming the random forest, bagging, and adaptive weighted blended ensemble models. Feature importance analysis, conducted using the SHapley Additive exPlanations technique, revealed that the number of stories, geographical location, and floor area are the factors most strongly affecting EUI, consistent with real-world expectations. Overall, this study highlights the effectiveness of combining green-BIM-driven simulations, advanced ML models, and interpretability techniques to address energy efficiency challenges in office buildings. The findings provide actionable insights for architects, engineers, and policymakers seeking to design energy-efficient buildings and implement targeted energy-saving strategies. •Green BIM and machine learning are integrated for office building EUI analysis.•Fusing FBI algorithm with ensemble ML models to enhance predictive accuracy.•A reliable dataset was created using Green BIM for enhanced energy modelling.•FBI-XGBoost achieved the lowest MAPE (1.86 %) in EUI prediction.•SHAP identified stories, location, and floor area as key EUI-influencing factors.