An Integrated BIM–One Click LCA–XGBoost Framework for Explainable Embodied Carbon Prediction and Waste-Aware Material Selection

The building sector is a major source of embodied carbon and construction waste. Design decisions made during the early stages strongly influence these impacts. This study presents an integrated framework that combines Building Information Modeling (BIM), One Click LCA, and explainable machine learning to predict embodied carbon. The framework also provides decision support for material selection that can help reduce construction waste during early design.

A three-dimensional BIM model of the KUET Textile Engineering Building was developed in Autodesk Revit. Material quantities and geometric properties were extracted through BIM-based quantity take-off. A life cycle assessment was performed in One Click LCA following EN 15804+A2. The assessment reported a total Global Warming Potential (GWP) of 4.44 × 10⁵ kg CO₂e, with the A1–A3 stage contributing 87.5% of the total impact. A material-level dataset containing 3,091 records was prepared through data cleaning, feature engineering, categorical encoding, and logarithmic transformation. The dataset was divided into 80% training and 20% testing. An XGBoost regression model was trained to predict embodied carbon, and SHAP analysis was used to explain feature contributions.

The proposed model achieved an R² of 0.997, MAE of 0.035, and RMSE of 0.141 on the test dataset. SHAP analysis identified material volume, quantity, and material area as the most influential variables. The results also confirmed that material production (A1–A3) dominates life-cycle carbon emissions, highlighting the importance of material selection during early design.

The proposed BIM–One Click LCA–XGBoost framework provides an interpretable decision-support approach for embodied carbon prediction. It supports carbon-aware material selection during the early design stage and provides practical guidance for reducing material-related environmental impacts. The findings are based on a single case study and provide a foundation for extending the framework to broader building types in future research.