Machine Learning Enabled Analysis of On-the-Road EV Charging Infrastructure: Predicting Accessibility and Optimizing Deployment

The United States is rapidly transitioning to electric vehicles (EVs) these days, and the effectiveness of this move will depend on how well charging stations are installed and made available on the road. A significant discrepancy remains in the availability of dedicated EV chargers, particularly on highways and inter-city travel corridors, despite increasing investment in charging infrastructure. In this paper, we investigate a machine learning–based approach to assessing the accessibility of existing infrastructure and projecting future demands. The objective of this research was to utilize machine learning approaches to perform a more detailed, national-scale assessment of the accessibility of EV charging infrastructure, both as it exists today and how it may best be deployed in the future. A comprehensive, multi-source dataset was compiled, encompassing several key variables. Data sources for charging station characteristics—such as number of ports, charger types (Level 2, DC fast), operational status and uptime history—were provided by the U.S. Department of Energy’s Alternative Fuels Data Center (AFDC). We implemented a multi-model framework to classify accessibility and predict infrastructure essential for facilitating optimal charging infrastructure accessibility prediction, with enhanced interpretation and baseline performance capabilities. XG-Boost and Random Forest models showed the same accuracy, getting the highest accuracy in the tested case scenario. This means both of these ensemble methods helped to classify instances in the dataset correctly. On the other hand, Logistic Regression, a more basic linear model, had a slightly lower accuracy. Deployment strategy – A multi-faceted deployment strategy is recommended based on the insights from the Model. Accessibility prediction models based on machine learning provide revolutionary capabilities for US transportation policy, particularly in enabling federal and state agencies to target infrastructure funds with data-driven tools. Using publicly available datasets and ML-enhanced planning tools, this work also helps to get closer to more equity in charger deployment as these areas in the latter sentences, namely rural, suburban, and disadvantaged communities, have had worse access to clean transportation infrastructure and are service and policy-challenged, where societal scarcities for EV chargers persist. Furthermore, predictive infrastructure modeling is crucial for alleviating EV range anxiety, a primary reason for reluctance to purchase EVs, according to surveys conducted by AAA and the Edison Electric Institute (EEI). Future Directions for Research: Given the exploratory nature of this study and the limitations discussed in the previous section, we envision several fruitful research avenues that could further enhance the predictive capacity and practical relevance of ML-enabled EV charger planning.