Design of MnO₂-based catalysts with activity approaching Pt/C via machine learning

MnO₂ is the preferred oxygen reduction reaction (ORR) catalyst for commercial metal-air batteries due to its diverse structures, high tunability, abundant reserves, and low cost. However, its catalytic activity remains far inferior to that of Pt/C. This work employs a machine learning strategy to develop a bimetallic Ce/Ni co-doped δ-MnO₂ ORR catalyst with activity close to Pt/C. Firstly, a multi-output regression model framework based on various ensemble learning algorithms is constructed. Using 64 sets of experimental data for training and comparison, Gradient Boosting was ultimately identified as the optimal prediction model. Based on this model, the predicted optimal composition is Ce/Ni co-doped δ-MnO₂ (7.5Ce/15Ni-MnO₂, the atomic ratios of Ce/Mn and Ni/Mn are 0.075 and 0.15, respectively), with predicted ORR performance including an onset potential (E_onset) of 0.893 V vs. RHE, a half-wave potential (E_1/2) of 0.830 V, and a limiting current density (J_L) of 4.504 mA cm^-2. Experimental validation shows that the measured E_onset, E_1/2, and J_L for this catalyst are 0.890 V, 0.828 V, and 4.56 mA cm^-2, respectively, which are in excellent agreement with the machine learning predictions. This study provides a machine learning-assisted design method for developing highly active MnO₂-based ORR catalysts.