Integrating BIM and machine learning to predict carbon emissions under foundation materialization stage: Case study of China's 35 public buildings

Abstract

For the significant energy consumption and environmental impact, it is crucial to identify the carbon emission characteristics of building foundations construction during the design phase. This study would like to establish a process-based carbon evaluating model, by adopting Building Information Modeling (BIM), and calculated the materialization-stage carbon emissions of building foundations without basement space in China, and identifying factors influencing the emissions through correlation analysis. These five factors include the building function type, building structure type, foundation area, foundation treatment method, and foundation depth. Additionally, this study develops several machine learning-based predictive models, including Decision Tree, Random Forest, XGBoost, and Neural Network. Among these models, XGBoost demonstrates a relatively higher degree of accuracy and minimal errors, can achieve the RMSE of 206.62 and R2 of 0.88 based on testing group feedback. The study reveals a substantial variability carbon emissions per building's floor area of foundations, ranging from 100 to 2000 kgCO2e/m2, demonstrating the potential for optimizing carbon emissions during the design phase of buildings. Besides, materials contribute significantly to total carbon emissions, accounting for 78%–97%, suggesting a significant opportunity for using BIM technology in the design phase to optimize carbon reduction efforts.