<p>The prediction of the remaining useful life (RUL) of lithium-ion batteries (LIBs) is essential for effective battery management systems (BMS), safety assurance, and timely maintenance, particularly in ensuring the reliable operation of electric vehicles (EVs). This study aims to develop an accurate prediction model using Optuna for hyper-tuning with the Kolmogorov Arnold Network (KAN) to assess battery health and estimate its lifespan. KAN, an advanced neural network, introduces a novel approach to machine learning by replacing traditional linear weights with univariate functions parameterised by splines. This enables the model to flexibly capture complex activation patterns, significantly enhancing its predictive capabilities. In this study, we propose Opt-KAN-XAI as an effective method to accurately estimate the RUL in energy storage devices. To further enhance interpretability, we integrate explainable artificial intelligence (XAI) techniques using Shapley additive explanations (SHAP) values. This analysis examines the influence of key features, such as temperature, cycle index, voltage, and current, on RUL predictions, highlighting the substantial impact of temperature on discharge capacity. The findings of this research underscore the potential of machine learning models in LIBs management within the XAI framework, demonstrating their strategic role in optimising energy storage systems. To validate the Opt-KAN-XAI method, we conducted experiments on the NASA and CALCE datasets and compared the results with other existing approaches. The experimental results confirm the model’s high accuracy and robustness, achieving a minimum test loss, root mean square error (RMSE) and a minimum mean absolute error (MAE), demonstrating its effectiveness in the precise estimation of RUL.</p>