Data-driven modeling of lithium-ion battery degradation using XGBoost with extended Kalman filter-based internal resistance correction

Accurate prediction of battery degradation is essential for the reliability and cost-effectiveness of lithium-ion batteries in energy storage. Previous degradation prediction methods such as model-based methods provide interpretability but rely on simplifications, whereas data-driven approaches capture nonlinearities yet lack robustness to noise and extrapolation. To address these challenges, this paper proposes a hybrid degradation modeling framework that integrates extreme gradient boosting (XGBoost) with an extended Kalman filter (EKF) correction. Experimental data were collected from 2000 continuous charge–discharge cycles of lithium iron phosphate (LFP) cells. Statistical features were extracted from current and voltage profiles to train the XGBoost model for capacity fade prediction. While the XGBoost model effectively learns complex degradation patterns, its predictions may be affected by operational uncertainties. To enhance robustness, the EKF was employed to estimate internal resistance in real time based on a Thevenin equivalent circuit model, and this estimation was used as a correction term for the XGBoost outputs. This hybrid design combines the accuracy of machine learning with the physical interpretability and stability of model-based estimation. Validation results show that the proposed method achieves a root mean square error (RMSE) of 0.041, mean absolute percentage error (MAPE) of 3.47%, and coefficient of determination ( R 2 ) of 0.98. In addition, the framework maintains low computational cost, making it suitable for real-time embedded applications. These results highlight the potential of the proposed method for reliable state-of-health monitoring and predictive maintenance of advanced battery energy storage systems. • A hybrid XGBoost–EKF model is proposed for Li-ion battery degradation. • XGBoost learns nonlinear patterns from historical operational data. • EKF enables real-time correction via internal resistance estimation. • The model achieves high accuracy: RMSE 0.041, MAPE 3.47%, and R 2 of 0.98. • Primary lab-tested dataset ensures reliability and experimental validity.