An Enhanced Interpretable ML Approach for Forecasting Long-Term CO2 and N2O Emissions via Optimized Random Forest Modeling

Abstract

Accurate long-term forecasting of greenhouse gas emissions is essential for climate mitigation planning, yet many existing studies either focus on a single gas or prioritize predictive accuracy without sufficient interpretability. To address this gap, this study develops an interpretable machine learning (ML) framework for the joint modeling of carbon dioxide (CO2) and nitrous oxide (N2O) emissions using country-level socio-economic indicators. Five regression-based algorithms, namely Decision Tree (DT), K-Nearest Neighbors (KNN), AdaBoost, Support Vector Machine (SVM), and Random Forest (RF), were benchmarked, and the best-performing model was further optimized using Random Search. The optimized Random Forest achieved the strongest predictive performance for both gases, reaching R2 values of 0.989 for CO2 and 0.982 for N2O, with corresponding Root Mean Square Error (RMSE) values of 40.09 and 3.24, respectively. Cross-validation results also confirmed strong stability, with mean R2 values of 0.987 +/- 0.004 for CO2 and 0.979 +/- 0.006 for N2O. Interpretability analyses based on SHapley Additive exPlanations (SHAP), Local Interpretable Model-agnostic Explanations (LIME), and ablation testing showed that Gross Domestic Product (GDP) is the dominant driver of CO2 emissions, whereas population has a relatively stronger contribution in N2O prediction. In methodological terms, the study contributes a transparent multi-gas framework that integrates model optimization, robustness assessment, and dual-layer explainability. Substantively, the results indicate that CO2 mitigation is more tightly associated with economic decoupling and energy transition, while N2O mitigation requires greater emphasis on agriculture-linked demographic pressures and nitrogen management. The framework is suitable for conditional forecasting and scenario-based policy analysis when future socio-economic pathways are supplied exogenously.