Carbon emissions prediction based on ensemble models: An empirical analysis from China

Abstract

The global warming problem has seriously threatened the sustainable development of human society. In order to effectively control carbon emissions, this study integrates economic, social, energy, and environment (ESEE) factors to develop a comprehensive, multi-dimensional carbon emissions prediction (CEP) index system, crucial for analyzing the determinants of carbon emissions and forecasting future emissions. Then employ a grey correlation analysis (GRA), followed by a genetic algorithm-based (GA-based) feature extraction method to demonstrate the strong correlation between selected factors and carbon emissions and refine the input of single and ensemble machine learning models for predicting carbon emissions in China. The number of selected economic factors is the highest, followed by energy factors, with only one social and environmental factors being selected. Meanwhile, the prediction results show that Bagging-ANN outperforms other algorithms with the lowest R² value of 0.8792, followed by Voting, Stacking, ANN, Bagging-SVR and SVR.