Machine Learning Approaches for Predicting Transport-Related CO₂Emissions in North Africa

Najah Kechiche, Jawher Bouaziz, Mustapha Lamine, Walid Hassen

In this study, we address the gap in TCO₂emissions prediction in the North African region by developing a machine learning framework for Algeria, Morocco, and Tunisia. The correlation analysis revealed that population and urbanization rate are the main positive drivers of transport-related CO₂emissions, whereas renewable energy contributes to emission reduction. Four machine learning algorithms (RF, SVR, XGBoost, and MLP) were developed and evaluated for TCO₂ prediction. The developed models achieved strong predictive performance during training, with R² values exceeding 0.994 and very low prediction errors. On test dataset, SVR provided the best performance for Algeria (R² = 0.98), RF showed the highest accuracy for Morocco (R² = 0.96), and MLP outperformed the other models for Tunisia (R² = 0.99). These results demonstrate the capability of machine learning techniques to predict TCO₂emissions and emphasize the need for selecting the most appropriate model to each country.