Predicting Carbon Emissions with Explainable Machine Learning Models: Applications for China’s Provinces

China has set “dual carbon” targets that involve achieving peak carbon emissions by 2030 and carbon neutrality by 2060, which are closely linked to carbon emissions. The projection of carbon emissions and the clarification of the factors that have a substantial influence on them are crucial in effectively controlling carbon emissions. In this analysis, the K-means method is implemented to sort the carbon emissions of the 30 provinces in China into three distinct classes: high, median, and low. Additionally, we examine the macro indicators in each group, and employ an explainable machine learning model to anticipate carbon emissions. This chapter analyzes the magnitude of influence exerted by each factor in the model and investigates the non-linear effects of macro indicators on carbon emissions with the Restricted Cubic Spline (RCS) methodology. The research findings suggest that the Support Vector Regression (SVR) model demonstrates the highest level of precision in predicting carbon emissions, with a 21.55% increase in Mean Absolute Error (MAE). Notably, natural gas, diesel oil, and raw coal are identified as the most influential factors in the model. Additionally, the influence of the four macroeconomic factors on carbon emissions vary, while the two macroeconomic indicators on carbon emissions present a converging pattern. It is worth highlighting that the proposed explainable machine learning model holds promise in predicting carbon emissions and positively contributes to enhancing environmental sustainability.