A Stacked Regression Framework for Remaining Driving Range Estimation in Electric Two-Wheelers

Accurate range prediction is crucial for reducing range anxiety and optimizing the energy management of electric two-wheelers (E2Ws). In this study, we propose a stacked ensemble model to predict the remaining driving range (RDR) of E2Ws with enhanced accuracy. Grid Search Cross-Validation (Grid Search CV) was employed for model selection and hyperparameter tuning to ensure the most effective regression models were chosen as base learners. The final ensemble consists of AdaBoost, CatBoost, Gradient Boosting, and Lasso Regression as base learners, while Ridge Regression serves as the meta-learner to refine predictions. To develop and validate the model, real-world driving data was collected from three different E2W models. The dataset underwent preprocessing and was evaluated using 10-fold cross-validation to ensure robustness. Experimental results demonstrate that the proposed stacked model achieves a Mean Absolute Error (MAE) of 0.13, corresponding to an average prediction error of 130 m per trip.