Prediction of Carbon Dioxide Emissions in China Using Shallow Learning with Cross Validation

Accurately measuring carbon dioxide (CO 2 ) emissions is critical for effectively implementing carbon reduction policies, and China’s increased investment in reducing CO 2 emissions is expected to significantly impact the world. In this study, the potential of shallow learning for predicting CO 2 emissions was explored. Data included CO 2 emissions, renewable energy consumption, and the share of primary, secondary, and tertiary industries in China from 1965 to 2021. These time-series data were converted into labeled sample data using the sliding window method to facilitate a supervised learning model for CO 2 emission prediction. Then, different shallow learning models with k-fold cross-validation were used to predict China’s short-term CO 2 emissions. Finally, optimal models were presented, and the important features were identified. The key findings were as follows. (1) The combined model of RF and LASSO performed best at predicting China’s short-term CO 2 emissions, followed by LASSO and SVR. The prediction performance of RF was very fragile to the window width. (2) The sliding window method is used to convert time series predictions into supervision learning problems, and historical data can be used to predict future carbon dioxide emissions. To ensure that the feature data are real, the model can predict CO 2 emissions for up to six years ahead. (3) Cross-validation and grid search were critical for optimizing China’s CO 2 emissions prediction with small datasets. (4) By 2027, carbon dioxide emissions will continue to grow and reach 10.3 billion tons. It can be seen that the task of China to achieve its carbon peak on schedule is very heavy. The results indicate that an increase in renewable energy consumption and adjustments in industrial structure will continue to play an important role in curbing China’s CO 2 emissions.