Closed-Loop Machine Learning-Guided Optimization of Fe- and Mo-Doped Ni Nanocatalysts for Efficient Hydrogen Evolution Reaction

Efficient hydrogen evolution reaction (HER) electrocatalysts are essential for sustainable hydrogen production, yet the optimization of nickel-based nanomaterials remains challenging due to complex, non-linear relationships between synthesis parameters and catalytic performance. Dopant composition, thermal treatment, and precursor chemistry strongly influence active site formation, electronic structure, and interfacial kinetics in alkaline media, limiting rational design using conventional trial-and-error methods. Herein, a machine learning (ML)-guided experimental framework is developed to optimize Fe- and Mo-doped Ni nanocatalysts for HER. A dataset of 32 catalysts was constructed using systematically varied synthesis conditions, including calcination temperature (300-600 °C), dopant type (Fe/Mo), dopant concentration (0-20 wt%), precursor ratio, and reaction time. Electrochemical performance was evaluated using linear sweep voltammetry and electrochemical impedance spectroscopy (EIS), extracting key outputs such as overpotential at 10 mA cm⁻², Tafel slope, and charge transfer resistance (Rct). Machine learning models were implemented in Python (scikit-learn), using pandas for data preprocessing, NumPy for numerical operations, and StandardScaler for feature normalization. Regression algorithms including linear regression, support vector regression, and random forest were evaluated, with hyperparameter tuning performed via GridSearchCV and 5-fold cross-validation. The random forest model achieved the best performance (R² > 0.92) and was used for feature importance analysis. Results indicate that calcination temperature and dopant concentration are the dominant factors governing HER activity, controlling Ni–M alloy formation, defect density, and NiOOH/Ni(OH)₂ surface redox behavior. Guided by ML predictions, an optimized Ni–Mo catalyst achieved an overpotential of ~120 mV at 10 mA cm⁻² and a Tafel slope of ~78 mV dec⁻¹, indicating enhanced Volmer-Heyrovsky kinetics. This work demonstrates that integrating Python-based ML with experimental validation provides a robust closed-loop strategy for rational catalyst optimization and accelerated discovery of high-performance, non-precious HER electrocatalysts.