Forecasting carbon dioxide emissions: application of a novel two-stage procedure based on machine learning models

Abstract Accurate forecast of carbon dioxide (CO2) emissions plays a significant role in China's peaking and neutrality policies. A novel two-stage procedure based on support vector regression (SVR), random forest (RF), ridge (Ridge), artificial neural network (ANN) is proposed evaluated by comparing it with the single-stage procedure. Nine independent variables’ data (study period: 1985–2020) are used to CO2 China. Our results reveal that, when time gap, h increases from 1 8, average root mean squared error (RMSE) absolute (MAE) SVR–SVR, SVR–RF, SVR–Ridge, SVR–ANN almost uniformly lower than errors arising their version, respectively. Among these models, exhibits lowest errors, whereas SVR–RF admits highest. The percentage decrease SVR–SVR vs. SVR, RF, SVR–Ridge Ridge, ANN 36.06, 5.98, 43.05, 14.81 for RMSE, 6.91, 43.27, 15.35 MAE. also suitable other variables, such as fossil fuel renewable energy consumption.