Explainable and Uncertainty-Aware Machine Learning for EV Energy Consumption Prediction

Electric vehicle consumption prediction is essential for improving route planning, charging management, and battery utilization. Although machine learning approaches have shown promising capability in prediction, the majority of the existing research mostly emphasizes point prediction, providing limited model interpretability and little or no quantification of prediction uncertainty. As a result, their practical application in actual EV energy management is still difficult, where confidence estimation and prediction transparency are crucial. This paper proposes an explainable and uncertainty-aware machine learning framework for reliable EV energy consumption prediction. A statistical analysis has been performed, and the features were engineered. Different machine learning techniques have been compared, and the best model was selected. SHAP-based explainability analysis was conducted to identify the contribution of proposed features to the model’s prediction. To increase the reliability of the prediction, split conformal-based prediction was used to obtain statistically valid prediction intervals. The proposed framework achieves an empirical coverage of 94.4% at a 95% confidence level with a mean prediction interval width of 1.961 kWh. Calibration analysis shows a good agreement between nominal and observed confidence levels. The obtained results show that the proposed framework offers accurate prediction, transparent model interpretation, and reliable uncertainty estimates that can be useful for practical EV energy management and decision-support applications.