Forecasting CO 2 emissions in Iraq using ARIMAX and artificial neural networks: a comparative modeling approach

Climate change is a critical global challenge driven by rising greenhouse gas emissions, particularly carbon dioxide CO $_2^{}$ . Accurate forecasting of CO $_2^{}$ emissions is essential for developing effective mitigation strategies. This study focuses on modeling and forecasting CO $_2^{}$ emissions in Iraq based on data from 1937 to 2023, incorporating climatic variables such as temperature and precipitation as exogenous variables to enhance forecast accuracy using multiple models, including traditional time series ARIMAX, Feedforward Neural Networks (FNN), Recurrent Neural Networks (RNN), and hybrid FNN-RNN. ARIMAX requires the assumption of linearity, FNN alone can model complex nonlinear interactions for each observation, while the RNN capture temporal relationships in sequential data. The hybrid configuration combining FNN and RNN models provides a learning of both linear and nonlinear structures. Empirical results indicate that the hybrid FNN-RNN model outperforms other models using key evaluation metrics, including $R^2$ , MSE, RMSE, and MAE. The hybrid model shows that both training and validation losses decrease steadily and converge to very low values without overfitting. The close alignment of the two curves indicates good generalization, and the slight dip in validation loss suggests effective regularization. Additionally, the study forecasts a significant 9.18% rise in Iraq's CO $_2^{}$ emissions over the 5 years from 2024 to 2028, and the forecast showed its highest recorded value in 2028. These findings may support policymakers in designing more accurate and proactive emission control strategies. While focused on climatic variables, the model offers a strong basis for future research to focus on socioeconomic factors such as GDP and population growth.