Analysis and Predictive Modelling of EV Charging Patterns and User Behaviour

As more Electric Vehicles (EVs) are released, the ability to predict energy consumption and charging duration becomes crucial in building optimal infrastructure and a proper system of managing energy. This paper proposes a machine learning model that predicts these two important parameters using a real-world dataset. The dataset consists of 1320 EV charging sessions made between January and February 2024 on Kaggle. The data set includes vehicle specifications, time stamps of sessions, environmental conditions, and user behaviour. Feature engineering tasks followed a thorough preprocessing procedure where missing values were imputed, outliers were removed, and the type of data was converted, and included time-based transformations, interaction terms, station popularity measures. Three regression models were developed: Light Gradient Boosting Machine (LGBM), Random Forest (RF), and Support Vector Regression (SVR) to evaluate different modelling approaches and test the predictive efficacy of ensemble against kernel-based methods. These models were trained and tuned using GridSearchCV combined with TimeSeriesSplit cross-validation to maintain temporal integrity. Evaluation metrics included Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and R-squared (R²). Results showed that RF achieved the highest accuracy in predicting energy consumption with an R² of 0.6620, while LGBM performed best in predicting charging duration with an R² of 0.9152. Final testing on unseen data validated the generalization capabilities of these models. The findings support practical infrastructure recommendations and demonstrate the potential of machine learning in enhancing EV charging operations.