Machine Learning-Based Prediction of Electric Vehicle Charging Demand for Sustainable Transportation

Abstract

The increasing adoption of Electric Vehicles (EV) presents challenges in managing uncertain charging demand, optimizing infrastructure utilization, and maintaining energy system stability. This study aims to develop and evaluate a machine learning-based prediction framework for estimating EV charging energy demand and identifying the most effective predictive algorithm to support sustainable transportation planning. The proposed framework employs historical EV charging observations and contextual variables, including charging duration, charging time, day of the week, ambient temperature, charging frequency, and historical energy consumption. Three supervised machine learning algorithms, namely Random Forest (RF), Support Vector Regression (SVR), and Extreme Gradient Boosting (XGBoost), are comparatively evaluated. Predictive performance is assessed using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and the coefficient of determination (R²). The reported experimental findings indicate that XGBoost achieves the highest predictive accuracy, with an RMSE of 2.84 kWh, MAE of 1.97 kWh, MAPE of 6.21%, and R² of 0.941. However, the comparative performance metrics for RF and SVR must be reported in the main Results section to substantiate the claimed superiority of XGBoost. The proposed framework demonstrates the potential of machine learning to identify charging demand patterns using historical charging, temporal, and environmental information. Subject to further experimental verification, the findings may support data-driven charging infrastructure management, electricity resource allocation, and sustainable transportation planning.