Modeling and predicting industrial carbon emissions in Nigeria using machine learning

Abstract

This study presents a comprehensive comparative analysis of machine learning and statistical modeling approaches for monitoring and predicting Nigeria’s industrial CO2 emissions in support of the nation’s 2060 net-zero target. A two-phase experimental design was implemented: the first phase established baseline performance for five models, Linear Regression, Prophet, ARIMA, Support Vector Regression (SVR), and Random Forest, while the second phase utilized hyperparameter optimization to improve predictive robustness. The results indicate that Multiple Linear Regression (MLR) and optimized SVR demonstrated the highest predictive accuracy, achieving R2 values of 0.932 and 0.923, respectively, with an optimized SVR RMSE of 2.229 Mt CO2. In contrast, ensemble and univariate models, including Random Forest and ARIMA, exhibited weak generalization capacity with negative R2 values, indicating limited predictive validity when applied to a constrained longitudinal dataset (n = 52). Statistical diagnostics confirmed the transport sector as the most significant exogenous driver of industrial CO2 emissions in Nigeria. The proposed comparative modeling approach establishes a robust methodological basis for emission forecasting, with projected industrial CO2 emissions for 2025 estimated at 123.79 Mt CO2 (MLR). These findings align with the objectives of Nigeria’s National Industrial Decarbonization Plan (NIDP), supporting evidence-based model selection and strategic sectoral prioritization for achieving long-term emission reduction goals.