Heteroatom-doped transition metal oxides (H-TMOs) are regarded as a promising class of electrocatalysts for the oxygen evolution reaction (OER). However, it is challenging and time-consuming to optimize the proper structure and elemental composition of H-TMOs. Herein, we develop an effective strategy that integrates a machine learning model with a genetic algorithm (GA) to forecast the overpotentials of NiO-based OER electrocatalysts with different heteroatom doping. Compared with other traditional machine learning and deep learning models, the Random Forest Regression (RFR) model exhibits the highest accuracy, achieving a root-mean-square error (RMSE) of only 4.73 mV on the test set. The prediction showed that the NiFeCeO electrocatalysts with mole fractions of Ce and Fe in the ranges of 0-0.25 and 0.15-0.65, respectively, exhibit lower overpotentials. Furthermore, the RFR predictions and GA optimization pinpointed Ni0.62Fe0.23Ce0.15O as the most promising OER electrocatalyst. Experimental validation shows that Ni0.62Fe0.23Ce0.15O exhibits an overpotential of 260 mV at a current density of 10 mA/cm2, positioning it near the apex of the activity volcano plot. Density functional theory (DFT) calculations illustrate that Fe and Ce doping into NiO can effectively reduce/eliminate the bandgap of NiO, leading to greatly improved electronic conductivity and electron transfer kinetics. Notably, the OER energy barrier of NiFeCeO (1.63 eV) is lower as compared to that of NiCeO (1.87 eV), NiFeO (1.72 eV), and NiO (1.96 eV). This study established a synergistic strategy combining machine learning-guided screening, experimental validation, and mechanistic DFT analysis for the rational design of heteroatom-doped TMOs, offering a paradigm for accelerating catalyst discovery.
Xue et al. (2026) studied this question.