Feature driven machine learning framework for accurate and computationally efficient remaining useful life prediction of lithium-ion batteries

Accurate estimation of the Remaining Useful Life (RUL) of lithium-ion batteries is essential for improving the reliability, safety, and cost-effectiveness of electric vehicles (EVs). Although several machine learning models have been reported for RUL prediction, many require extensive feature engineering or high computational effort, which limits their practical use in battery management systems. This work presents a predictive machine learning framework that balances prediction accuracy with computational efficiency. The proposed framework uses sensor-derived features from the NASA battery degradation dataset. The data are first preprocessed through exploratory data analysis and normalization to improve their quality. Feature analysis identifies the maximum discharge voltage and the minimum charge voltage as the two most influential indicators of battery degradation. These features are then used to train three regression models, namely k-Nearest Neighbors (kNN), Support Vector Regression (SVR), and Random Forest (RF). The models are evaluated using multiple statistical performance metrics. Among them, the RF model delivers the best overall performance with a coefficient of determination ( \(R²\) ) of 0.9421 and a root mean square error (RMSE) of 77.47. The kNN model achieves comparable prediction accuracy while requiring less training time. However, the SVR model exhibits lower prediction accuracy incurring higher computational cost. The proposed framework is further validated using independent battery datasets, where the predicted RUL closely follows the observed degradation trend. These results demonstrate that the proposed approach provides accurate, reliable, and computationally efficient RUL estimation, making it suitable for real-time battery health monitoring and predictive battery management in electric vehicles. In the future, this framework has the potential to contribute to effective and scalable battery health management systems. Furthermore, it could be extended to deep learning and adaptive models to be used under diverse operating conditions.