Abstract:

Metal doping is a key modification strategy for MnO₂ cathodes in aqueous zinc-ion batteries, with parameter selection directly governing the resulting electrochemical performance. However, the intricate interplay among dopant type and concentration, synthesis conditions, and electrochemical performance renders the optimization of MnO₂ cathodes for high electrochemical properties still elusive. Traditional trial-and-error experimental screening is time-consuming and expensive, and developing a unified guideline on the design of metal-doped MnO₂ remains a long-standing challenge. To efficiently study the performance of metal-doped MnO₂, we proposed a machine learning (ML) model driven by data from the literature. A dataset was constructed from 36 articles covering 21 dopant elements, integrating elemental descriptors, synthesis parameters, and electrochemical testing conditions. After feature filtering and model evaluation, the extreme gradient boosting (XGB) model achieves a high predictive accuracy with an R² of 0.921 on the test set. Beyond prediction, model interpretability using Shapley additive explanations (SHAP) analysis identifies the dominant factors affecting capacity, revealing the influence of current density, dopant ratio, and molecular weight. Feature importance analysis further guided the design of a series of experiments that validated the accuracy and reliability of the model. Experiments on Fe- and Ni-doped MnO₂ cathodes confirmed the ability of metal doping to enhance specific capacity, and the model achieved a mean absolute error (MAE) below 12 mAh g⁻¹ for all cases. Density functional theory (DFT) calculations further verified the molecular-level mechanism of metal doping by demonstrating that dopant incorporation modulates the electronic structure of MnO₂ and narrows the bandgap, improving conductivity. The consistency between the ML results, experimental validation, and theoretical calculations highlights the robustness of the proposed framework. Having established the feasibility from multiple perspectives, we further deployed a performance prediction platform based on this model, providing a convenient tool for researchers to rapidly estimate the specific capacity of metal-doped MnO₂ under user-defined conditions. This work demonstrates a comprehensive data-driven paradigm that integrates ML, experimental validation, and theoretical calculations. We believe that this approach provides a new strategy and framework for the rational design of high-performance MnO₂ cathodes, and is broadly applicable for accelerating the discovery of other metal-doped energy storage materials.

Key words: Data driven, Machine learning, Zinc-ion battery, Cathode, Metal-doped manganese dioxide