Smart Feature Selection-Based Machine Learning Framework for Calendar Loss Prediction of Li-Ion Electric Vehicle Battery

Li-ion batteries are widely used in electric vehicles (EVs), but they suffer from battery degradation, particularly in terms of calendar loss. Machine learning has been used to predict battery health deterioration due to cyclic loss, but the accuracy depends on input feature selection. This study introduces an improved feature selection method, enhancing battery calendar loss prediction. Eight machine learning algorithms are applied to an EV battery dataset, and results show improved prediction accuracy and reduced mean absolute error (MAE). Notably, Gaussian process regression (GPR), random forest (RF) regression, and XGBoost methods combined with the proposed feature selection method show the most significant accuracy improvement.