Sustainability assessment on green construction using a novel analytical framework integrating machine learning and emergy analysis

Carbon emissions from building construction phase contribute 26.6 % of China's energy-related carbon emissions, yet their assessment remains a key scientific and engineering challenge. Existing systems depend on subjective weighting and linear assumptions, failing to capture the nonlinear relationship between resource inputs and emissions. This study establishes a data-driven sustainability assessment framework for green construction integrating emergy theory and machine learning. Based on data from 5110 buildings across 190 projects, six algorithms (DT, LightGBM, CatBoost, XGBoost, RF, ET) were used to build emergy–carbon emission regression models. The ET model achieved the best performance ( R 2 = 0.9714), effectively characterizing the nonlinear interactions between emergy inputs and emissions. Purchased and non-renewable emergy drive emissions, while renewable emergy mitigates them. The emergy sustainability index correlates strongly with emissions, serving as an objective evaluation metric. This framework connects emergy theory with empirical modeling, supporting a scientific basis and practical tool for low-carbon construction and policy standardization. • An emergy-carbon model based on 190 projects was established to reveal nonlinear resource-emission links. • A 17-indicator framework for assessing construction phase emissions was developed. • Indicator weights and their impact on emission levels were quantified using machine learning. • The framework was applied to support practical carbon reduction and resource optimization.