Evaluating Machine Learning Models for Sustainability to Improve Electric Vehicle Range Prediction

Sustainability in machine learning considers energy consumption, computational demand, model performance, and deployment at ease. The rapid growth and implementation of electric vehicles (EVs) presents two challenges: significant opportunities, largely associated with resolving "range anxiety" about an EV battery’s life, where consumers have little to no faith in the range prediction technology employed, EV adoption needs an accurate estimate as does user willingness. This study evaluates four machine learning models with a multivariate dataset-Decision Trees, Random Forests, Support Vector Regression (SVR), and K-nearest neighbors (KNN). For this experiment, the goal is restricted to deciding which model is better at predicting EV range. The dataset has features like Acceleration, Efficiency, Powertrain Manufacture Cost Quick Charge Capacity Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> Score (coefficient of determination) are the evaluation metrics used to measure model correctness. The results reveal that Decision Trees and SVR distantly performed the worst, as Random Forests outperformed all models with a top R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> score and lowest error metrics. Based on our results, Random Forests lead to the highest accuracy in predicting EV range; however, model complexity and interpretability needs should be considered. In addition, this study sheds light on the suitability of different machine learning models for EV range prediction by highlighting pathways that could be used to conduct additional research and refine existing predictive models.