AI-Powered Modeling of Carbon Emissions and Renewable Energy Optimization in the U.S. Power Sector

¹ Escuela Superior Politécnica del Litoral, FCNM, 90902 Guayaquil, Ecuador
² Universidad de Guayaquil, FCE, 90514 Guayaquil, Ecuador
³ Universidad ECOTEC, FSC, 92301 Samborondón, Ecuador

^* Corresponding author: pramos@espol.edu.ec

Abstract

The transition to renewable energy is essential for mitigating climate change and achieving net-zero carbon emissions. This study employs the Emissions & Generation Resource Integrated Database (eGRID) 2023 dataset, which provides comprehensive data on power generation capacity, fuel types, and carbon emissions across the U.S. power sector. Advanced machine learning techniques, including Gradient Boosting and Random Forest models, were applied to explore key predictors of CO₂ emissions and their complex interrelationships. The analysis highlights that renewable energy integration significantly reduces emissions, with net generation and fuel type identified as the most influential variables. A robust preprocessing pipeline ensured data reliability by addressing outliers and missing values, while feature engineering captured critical insights. The Random Forest and XGBoost models demonstrated strong predictive capabilities, achieving R² scores of 0.91 and 0.93, and RMSE values of 58,123.4 and 52,876.2, respectively, with XGBoost slightly outperforming Random Forest, in both metrics. These findings emphasize the importance of renewable energy policies and offer actionable strategies for optimizing power generation. Policymakers and industry leaders can utilize these insights to prioritize investments in renewable energy infrastructure, enhance grid resilience, and foster sustainable growth. This study illustrates the transformative potential of machine learning in converting large-scale energy data into actionable intelligence, advancing the transition to a net-zero power sector.