Time series analysis of carbon dioxide emission: a comparison of statistical, machine learning, and deep learning...

Carbon dioxide (CO₂) is one of the infamous greenhouse gases, resulting in increased global temperature and climate change. The steady rise in atmospheric CO₂ levels, primarily driven by anthropogenic activities, poses serious environmental and socio-economic challenges. Understanding and forecasting CO₂ emission trends are essential for guiding global mitigation efforts and assessing progress toward climate commitments. In this study, we aim to investigate the monthly CO₂ emission trend from direct measurement data collected by the National Oceanic and Atmospheric Administration and build predictive models from three different modeling approaches: statistical: autoregressive integrated moving average, seasonal autoregressive integrated moving average; machine learning: Random Forest, adaptive boosting; and deep learning models: long short-term memory, gated recurrent unit. We constructed and trained multiple model configurations through a data-driven approach, then evaluated and selected a top-performing model from each category, enabling a robust performance comparison. Based on the experimental results on test data, the Random Forest model with 250 decision trees outperformed all other models with the best scores: root mean square error 0.2401, mean absolute percentage error 0.0005, and directional accuracy 0.9544. Forecasting was performed for the next 3 years from the top-performing models in each category. Experimenting from statistical to state-of-the-art deep learning models, this study serves as a baseline case study for developing advanced computational frameworks on emission data forecasting for the future. Overall, this research provides valuable tools and perspectives for climate scientists, stakeholders, and policymakers aiming to combat climate change through informed, predictive strategies.