A GREET-Derived Machine Learning Framework for Life-Cycle Emissions Prediction of Electric and Combustion Vehicles

A reproducible GREET-derived lifecycle dataset con-
Training 58,905 BEV, gasoline ICEV, and diesel ICEV
scenarios across U.S. electricity-grid regions, overcoming
the scalability limitations of process-based LCA. A feed-forward tabular neural network (MLPRegressor) that serves as a computationally efficient surrogate for predicting total life cycle GHG emissions from vehicle and regional electricity grid characteristics. Permutation feature importance analysis to identify the key drivers of life cycle emissions, improving model interpretability and linking ML-based prediction with LCA-based vehicle comparison.