Toward High‐Performance Electrochemical Energy Storage Systems: A Case Study on Predicting Electrochemical Properties and Inverse Material Design of MXene‐Based Electrode Materials with Automated Machine Learning (AutoML)

Abstract

This study highlights the potential of Automated Machine Learning (AutoML) to improve and accelerate the optimization and synthesis processes and facilitate the discovery of materials. Using a Density Functional Theory (DFT)‐simulated dataset of monolayer MXene‐based electrodes, AutoML assesses 20 regression models to predict key electrochemical and structural properties, including intercalation voltage, theoretical capacity, and lattice parameters. The CatBoost regressor achieves R2 values of 0.81 for intercalation voltage, 0.995 for theoretical capacity as well as 0.807 and 0.997 for intercalated and non‐intercalated in‐plane lattice constants, respectively. Feature importance analyses reveal essential structure‐property relationships, improving model interpretability. AutoML's classification module also bolsters inverse material design, effectively identifying promising compositions, such as Mg2+‐intercalated and oxygen‐terminated ScC2 MXenes, for high‐capacity and high‐voltage energy storage applications. This approach diminishes reliance on computational expertise by automating model selection, hyperparameter tuning, and performance evaluation. While MXene‐based electrodes serve as a demonstrative system, the methodology and workflow can extend to other material systems, including perovskites and conductive polymers. Future efforts should prioritize integrating AutoML with real‐time experimental feedback and hybrid simulation frameworks to create adaptive systems. These systems can iteratively refine predictions and optimize trade‐offs among critical metrics like capacity, stability, and charge/discharge rates, driving advancements in energy storage and other material applications.