Evaluating the machine learning-based models for predicting carbon neutrality in Sub-Saharan African Nations

Abstract

Achieving net-zero emission goals is a critical challenge for Sub-Saharan Africa, a region burdened with distinct economic and environmental pressures. This study employs a comparative machine learning (ML) framework, utilizing Decision Tree, XGBoost, Random Forest, Elastic Net, Lasso, and AdaBoost models, to predict CO2 emissions based on key socioeconomic and energy-related factors, including energy efficiency, government stability, clean energy, GDP per capita, and population. The Random Forest model outperformed others, achieving an RMSE of 0.310, MAPE of 2.73%, MAE of 0.228, and R2 of 0.954, indicating its robustness in handling nonlinear interactions. Sensitivity analysis and the SHAP model revealed that government stability has minimal influence on CO2 emissions, while feature importance analysis identified population as the most critical determinant. Additionally, the results highlight regional variations, with Ethiopia, Ghana, and South Africa showing consistent prediction trends, whereas countries like Burundi, Gabon, Mozambique, and Zimbabwe exhibit higher prediction uncertainties. These findings underscore the necessity for policymakers in Sub-Saharan Africa to prioritize targeted investments in clean energy, implement enhanced regulatory frameworks, and foster regional cooperation to achieve carbon neutrality. The study also emphasizes the value of machine learning techniques in providing actionable insights for sustainable development in resource-constrained regions.