Machine learning models for predicting rural residential carbon emissions and optimising spatial forms

Global warming is exacerbating its effects on ecosystems and human populations, with carbon emissions identified as the primary cause. As a crucial aspect of urban development, spatial form significantly impacts energy efficiency and carbon emissions. However, research on rural areas has been limited compared to that on metropolitan and central cities. This study investigates carbon emissions and spatial form in rural residential areas. It employs Random Forest, XGBoost, and BP Neural Network methodologies to predict carbon emissions and optimise spatial form. The findings indicate that (1) spatial form factors, including floor area ratio, number of floors, and building orientation, exhibit a strong correlation with carbon emissions; (2) the XGBoost model demonstrates superior prediction accuracy and generalization ability, achieving a reduction of over 10% in carbon emissions under the optimized spatial form; (3) optimization strategies, such as regulating the floor area ratio and minimizing the building shape coefficient, are proposed. These results provide a scientific foundation for low-carbon rural development and facilitate a green transition.