Machine learning approach for analyzing feature importance in alternative fuel vehicle selection

Abstract

This study investigates the differences in the importance of various factors influencing vehicle preferences by fuel type. We analyzed four major fuel types: gasoline, diesel, electric, and hydrogen, aiming to enhance the effectiveness of zero-emission vehicle policies. Using Shapley additive explanations with an XGBoost classifier, we evaluated feature importance using conjoint survey data, considering vehicle attributes and owner characteristics, such as current vehicle usage. This approach not only identifies the most impactful criteria for more precise policy segmentation but also addresses the limitations of traditional methods that struggle to reveal differences in factor significance across fuel types. The results show that consumers choosing electric vehicles prioritize recharging infrastructure availability and economic factors, such as vehicle price and household income. By contrast, hydrogen vehicle selection is heavily influenced by the availability of hydrogen refueling infrastructure and demographic factors, such as age. Additionally, partial dependence plots illustrate the influence of recharging or refueling convenience on preferences, providing insights for strategic investments in zero-emission infrastructure. This study provides valuable insights for policymakers and infrastructure planners seeking to promote the adoption of zero-emission vehicles by demonstrating the variation in factor importance across fuel types.