Predicting Electric Vehicle Ranges Based on Cities Using Machine Learning Algorithms

Abstract

This study introduces a novel, data-driven framework for predicting electric vehicle (EV) range by incorporating city-level geographic and infrastructural variables alongside advanced machine learning (ML) algorithms, thereby addressing the key limitations of traditional models that focus solely on internal vehicle metrics. Using a comprehensive dataset of more than 65,000 real-world vehicles across the United States, the research highlights that urban context—including climatic factors, traffic flow, and regional infrastructure—is a crucial determinant of EV energy consumption. A rigorous performance evaluation comparing Random Forest (RF), Gradient Boosting (GB), and Neural Networks (NN) identifies RF as the most effective approach, delivering exceptional predictive accuracy with an R2 of 0.994 and an RMSE of 7.387 miles, significantly outperforming existing models in the literature. Ultimately, these results demonstrate that systematically integrating urban-level data substantially enhances the precision of range predictions, offering a practical solution to alleviate range anxiety, guide consumer expectations, and equip urban planners and policymakers with a reliable, data-backed foundation for optimizing charging networks and advancing sustainable electric mobility.