Interpretable Machine Learning Accelerates Rational Design of Dual-Site Catalysts for Synergistic Catalytic Aerobic Oxidative Synthesis of Benzimidazole

The aerobic oxidative synthesis of benzimidazole requires heterogeneous catalysts that combine high activity with practical stability. Because conventional catalyst optimization depends heavily on laborious experimental screening, we established an interpretable machine learning (ML) workflow to accelerate catalyst selection and identify the variables governing performance. An XGBoost model was trained to relate catalyst descriptors and reaction conditions to benzimidazole synthesis, and partial-dependence-plot (PDP) analysis was used to determine favorable catalyst compositions and operating windows. Guided by the model, we prepared the cobalt-based dual-site catalyst CoNP-CoSA@NC-800. Under mild conditions (90 °C, 9 h) with atmospheric air as the oxidant, this material delivered 100% conversion and 96% benzimidazole selectivity. It also tolerated a broad range of substrates and retained its activity during recycling. Control experiments and theoretical calculations indicate that the Co nanoparticle (CoNP) and Co single atom (CoSA) sites act cooperatively: the CoNP component modifies the d-band center and electronic structure of CoNP-CoSA@NC-800, thereby enhancing O2 adsorption and activation to facilitate benzimidazole formation.