AI-Driven Forecasting of US CO₂ Emissions: Machine Learning Approaches for Sustainable Policy Development

Abstract

The increasing levels of CO₂ emissions in the United States pose significant challenges to environmental sustainability and climate change mitigation. This study explores the potential of machine learning (ML) models in forecasting CO₂ emissions and supporting sustainable policy formulation. Using historical emissions data, energy consumption patterns, and economic indicators, we develop and compare multiple ML models, including Long Short-Term Memory (LSTM) networks, Gradient Boosting Machines (GBM), and Random Forest regressors. The findings indicate that LSTM achieves the highest predictive accuracy with a mean absolute percentage error (MAPE) of 4.7%, outperforming traditional statistical approaches. Additionally, feature importance analysis reveals that fossil fuel consumption, GDP growth, and industrial activities are the most significant contributors to emission fluctuations. The proposed AI-driven forecasting framework provides policymakers with a data-driven approach to developing proactive strategies for emission reduction and climate sustainability. Future research should incorporate real-time data streams and hybrid AI models to enhance predictive capabilities and support more dynamic policy interventions.