Optimized XGBoost framework for RUL prediction of lithium-ion batteries using multi health indicators

Abstract

Lithium-ion batteries are central to the growth of electric vehicles (EVs), providing high energy density, long life, and efficiency for sustainable transportation. Nevertheless, information on the Remaining Useful Life (RUL) of battery remains a prominent concern as it is vital for preventing failures, optimizing battery life, reducing costs, and ensuring reliable performance. In order to increase the accuracy in battery RUL prediction, this work proposes a hybrid framework that integrates RRelief-based feature selection with Bayesian-optimized Extreme Gradient Boosting (XGBoost). Only five dominant features have been used while applying Bayesian hyperparameter optimization, thereby lowering computational cost, mitigating overfitting, and enhancing robustness. The proposed approach is benchmarked against multiple machine learning models, including Support Vector Machine (SVM), Gaussian Process Regression (GPR), Neural Networks, and Ensemble techniques. The outcomes indicate that the optimized framework achieved a substantial reduction in Root Mean Squared Error (RMSE) from 6.84 to 2.61 and execution time by 2.36 s as compared to the untuned baseline model. Dual validation strategies—5-fold cross-validation and a 25% hold-out split—confirmed its generalizability.