The electrocatalytic two-electron oxygen reduction reaction (2e– ORR) synthesis of hydrogen peroxide (H2O2) is an eco-friendly process. In this study, density functional theory (DFT) and machine learning (ML) were utilized to systematically explore various single metal atoms anchored on mono-vacancy fullerene C60 (TM@C60) for H2O2 production. Among 19 single-atom catalysts (SACs), Sc@C60 and Y@C60 were identified for high the thermodynamic feasibility of H2O2 formation, exhibiting overpotentials of 0.01 V and 0.02 V, respectively. Moreover, competitive four-electron oxygen reduction reaction (4e– ORR) and hydrogen evolution reaction (HER) were significantly suppressed on these two catalysts. The formation of peroxide intermediates on the two screened SACs was identified as key to achieving high activity and selectivity for H2O2 formation. Analysis using ML methods also revealed that the balance between bond strengths and adsorption energies for O2 is crucial in regulating the performance of 2e– ORR. These results offer important insights for electrocatalytic H2O2 production.