Interpretable machine learning for green design of iron-based catalysts for efficient antibiotic degradation

The conventional development of high-performance catalysts for antibiotic degradation often relies on resource-intensive trial-and-error methods, highlighting a critical bottleneck in green process design. Therefore, this study develops an interpretable machine-learning framework to enable rapid, low-cost, and efficient antibiotic degradation via PMS activation with iron-based oxide catalysts. A high-quality dataset of 1,022 experimental records is constructed to investigate the structure-activity relationships of catalyst properties. Six machine learning models are automatically optimized using the Optuna framework, with CatBoost (CAB) identified as the optimal predictor (test-set R 2 = 0.9595). The interpretability of the CAB model is deciphered through Shapley additive explanations and partial dependence plots, revealing reaction time, drug, pH, PMS concentration, and catalyst dosage as the most critical features governing degradation efficiency. Finally, the preferred model integrated with the NSGA-III algorithm is employed to resolve the efficiency-cost-time trilemma, simultaneously maximizing degradation rate while minimizing reaction time and PMS consumption. The optimized MnFe 2 O 4 /CuS and Fe 3 O 4 /MoS 2 catalysts achieve degradation rates of 96.15% and 94.51% for lomefloxacin hydrochloride and ofloxacin, within a 25.01-minute reaction time and at a PMS concentration of 0.14 mM, which are significantly superior to the reported values. This work establishes a data-driven paradigm for green chemical engineering, offering a novel toolkit for the rational and sustainable design of advanced oxidation processes with minimized resource footprint. • An interpretable ML framework is proposed to achieve rapid and low-cost antibiotic degradation via PMS activation by Fe-based oxide catalysts. • CatBoost is identified as the optimal predictor with the highest accuracy. • Interpretability analysis is conducted to elaborate the importance and interaction mechanism of features. • Interpretable ML is coupled with NSGA-III algorithm to resolve the critical tradeoff among antibiotic degradation process. • Optimized catalysts achieve high degradation rates and low dosage for the degradation of lomefloxacin hydrochloride and ofloxacin.