Hybrid Neural Network and Time Series Approaches for Forecasting CO2 Emissions: Iraq as a Case Study

This study shows the forecasting of carbon dioxide (CO2) emissions in Iraq by combining neural network and time series approaches. Using annual data from 1937 to 2023, CO2 emissions were modeled as the dependent variable, with air temperature and precipitation as explanatory variables. To ensure comparability and minimize scale effects, all variables were standardized prior to analysis. Three forecasting models were evaluated: the Auto-regressive Integrated Moving Average model with Exogenous variables (ARIMAX), Recurrent Neural Networks (RNN), and a hybrid ARIMAX-RNN model. Forecast performance was evaluated using multiple accuracy metrics, including the coefficient of determination (R2), Mean Square Error (MSE), Mean Absolute Error (MAE), and Root Mean Absolute Error (RMAE). The results demonstrated that while both the ARIMAX and RNN models provide reasonable predictions, the hybrid ARIMAX-RNN model consistently outperforms them across all evaluation criteria, capturing both the linear and nonlinear structures of the data. Regrettably, the hybrid model shows huge forecasted value for CO2 emissions for 2024 and 2025 and then ratio indicate only a slight decrease in other forecast years. However, despite this modest downward trajectory, the predicted values remain significantly higher than those recorded in the historical dataset until now Iraq not recorded this high record in previous years, indicating that Iraq’s CO2 emissions are still on an overall upward trajectory compared to previous levels. This result underscores the dual messages of the forecasts: while a short-term decline may occur, the long-term cumulative emissions burden continues to grow, and highlight the methodological advantage of hybrid forecasting methods to the urgent need for effective climate and environmental policies to mitigate Iraq’s future carbon dioxide emissions trajectory.