Abstract
•Predictive models of Electric Vehicle public charging units, jobs, and registrations.•Solar energy, education, and air quality have the greatest predictive impact.•Simple models explain most (51.5–62.4%) of the variance.•Advanced models capture many non-linear relationships.•The advanced models are significantly more accurate than the simple models.
In this paper, we develop several predictive models pertaining to Electric Vehicles in the United States. We set out to understand three phenomena: Public Charging Units, Electric Vehicle Jobs, and Electric Vehicle Registrations. To model them, we include variables from various categories – demographics, economics, education, environment, finance, geographics, public health, and technographics – in the timespan of 2010–2020. We integrate data from multiple data sources and use them to understand, explain, and predict the Electric Vehicle phenomena combining state and county level data. We obtain three random effects linear regression models having variance explained of 51.5–62.4%. We also fit several machine learning models to improve the accuracy of the models, highlighting nonlinearities, to provide additional insights. The overall predictive accuracy of each Gradient Boosted Tree model is far superior to that of each linear regression model and significantly better than the other machine learning models. Our core contribution is that, spanning all three phenomena, the three most important predictor variables are solar generation of electricity, high school graduation rates, and air quality. We interpret the models and discuss their implications for research and practice.
