A data-driven interpretable method to predict capacities of metal ion doped TiO2 anode materials for lithium-ion batteries using machine learning classifiers

Abstract

Metal ion doping is an effective method to improve the electrochemical performance of metal oxide anode materials for lithium-ion batteries (LIBs). In the past many years, screening for dopant oxides has largely relied on trial and error. Benefiting from the data that researchers have provided with great experimental efforts and improvements of computer science, machine learning (ML) provides a chance to predict the performance of doped oxide anode materials before synthesizing them, which can save amounts of time and resources. Herein we built four ML classifier models with the datasets of 14 different ions doped into TiO₂ materials to predict their charging and discharging performance. Among them, the gradient boosting (GB) model achieved an accuracy of 0.79 and a specificity (true negative rate, TNR) of 0.90. Furthermore, a Pearson correlation study and SHapley Additive exPlanations (SHAP) were introduced to establish the initial correlation between the properties of the dopant and the capacity performance of the material, indicating that the higher the electronegativity of the doping element, the more likely it is the material possesses higher capacity. This study is expected to provide an effective method for exploring high-performance doped metal oxide anode materials.