A comparative study of statistical and machine learning models on carbon dioxide emissions prediction of China

The escalating levels of carbon dioxide (CO2) emissions represent the primary driver of global warming, and addressing them is of paramount importance. Timely and accurate prediction, as well as effective control of CO2 emissions, are pivotal for guiding mitigation measures. This paper aims to select the best prediction model for near-real-time daily CO2 emissions in China. The prediction models are based on univariate daily time-series data spanning January 1st, 2020, to September 30st, 2022. Six models are proposed, including three statistical models: grey prediction (GM(1,1)), autoregressive integrated moving average (ARIMA), and seasonal autoregressive integrated moving average with exogenous factors (SARIMAX), and three machine learning models: artificial neural network (ANN), random forest (RF), and long short-term memory (LSTM). The performance of these six models is evaluated using five criteria: mean squared error (MSE), root-mean squared error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and coefficient of determination (R2). Our findings reveal that the three machine learning models consistently outperform the three statistical models across all five criteria. Among them, the LSTM model demonstrates exceptional performance for daily CO2 emission prediction, boasting an impressively low MSE value of 3.5179e-04, an RMSE value of 0.0187, an MAE value of 0.0140, an MAPE value of 14.8291%, and a high R2 value of 0.9844. This underscores the robustness of the LSTM model in capturing and predicting complex emission patterns, positioning it as the most suitable option for near-real-time daily CO2 emission prediction based on the provided daily time series data. Moreover, our study's results provide valuable insights into emissions forecasting, enabling data-driven decision-making for policymakers and stakeholders. The accurate and timely predictions offered by the LSTM model can aid in the formulation of effective strategies to mitigate carbon emissions, contributing to a more sustainable future. Furthermore, the findings of this study can enhance our understanding of the dynamics of CO2 emissions, leading to more informed environmental policies and actions aimed at reducing carbon emissions.