[Identification of Urban Road Traffic Carbon Emission Driving Mechanisms Based on Interpretable Machine Learning]

Under the background of accelerated urbanization， the continuous increase in the number of motor vehicles has led to a rapid rise in road traffic carbon emissions， becoming an important factor restricting the green transformation of cities. In order to systematically identify the key driving factors of road traffic carbon emissions， a "bottom-up" approach was used to construct a road traffic carbon emission inventory for the urban agglomerations in Fujian Province from 2003 to 2022， with motor vehicles being subdivided into 13 types for accounting purposes. Secondly， correlation analysis and Lasso regression were combined for variable selection， and multiple machine learning algorithms were used to build carbon emission prediction models. Finally， SHAP values were employed to enhance model interpretability and quantify the contributions of driving factors. The results show that： ① The output value of the transportation industry， urban green space area， and the number of invention patents were the main factors affecting carbon emissions. ② Among all models， XGBoost performed the best， with the test set R² of 0.992 and MAE and RMSE of 7.393×10⁴ t and 8.803×10⁴ t， respectively. ③ SHAP analysis revealed the positive and negative impact pathways of key factors on carbon emissions. This study， starting from the perspective of interpretability， reveals the driving mechanisms of urban road traffic carbon emissions dominated by motor vehicles， providing scientific support for the formulation of low-carbon transportation policies.