A Machine Learning-Based Approach for Modeling Leaching Conditions in the Recycling of NMC Lithium-Ion Batteries

Abstract

In this study, the effects of leaching parameters on the hydrometallurgical recycling of NMC-type (LiNiₓMnᵧCo_zO₂) lithium-ion batteries were investigated using machine learning-based models. The dataset, compiled from the literature, consists of 269 experimental data points and includes key variables such as time, temperature, solid-to-liquid (S/L) ratio, H₂O₂ concentration, and citric acid concentration. The dissolution rates of lithium (Li) and transition metals (Co, Ni, Mn) were used as target variables. Linear Regression, Decision Tree, and Random Forest algorithms were compared during the modeling process, with the Random Forest model achieving the highest accuracy (R² > 0.88). Model outputs were interpreted using SHAP (SHapley Additive Explanations) analysis to determine the relative importance of the variables, and Partial Dependence Plots (PDPs) were used to visualize the effects of process parameters on leaching behavior. The findings revealed that operational conditions such as temperature and time were particularly decisive for Li dissolution, while H₂O₂ and citric acid concentrations had a greater influence on the dissolution of transition metals. This study demonstrates that re-evaluating literature data through artificial intelligence can effectively guide experimental optimization efforts.