Machine learning-guided optimization of iron-based catalysts toward minimal-resource and efficient peroxymonosulfate...

Conventional trial-and-error approaches for designing iron-based oxide catalysts to activate peroxymonosulfate (PMS) are notoriously resource-intensive, time-consuming, and costly. To address this challenge, this study proposes a novel dual-task artificial intelligence framework (DualAI-MCD) that integrates machine learning with multi-objective optimization for the predictive design and mechanism-informed optimization of catalysts. This framework simultaneously predicts the organic pollutant degradation rate (regression task) and identifies the dominant reactive oxygen species pathway (classification task). Leveraging a curated dataset of 3720 experimental records spanning 13 critical features, various machine learning and deep learning models are developed and then automatically optimized. Results demonstrate the performance of machine learning models is significantly superior to that of deep learning models, with LGBM achieving state-of-the-art performance: 100 % classification accuracy and exceptional regression capability for degradation rate (R²=0.9469). Various interpretability analysis methods are employed to effectively illustrate the interrelationships among features and their impact on the prediction mechanism. Results reveal that reaction time dominates degradation rate prediction, while modified components govern the selection of radical/non-radical pathways. The optimized LGBM model is integrated with a hybrid optimization algorithm to reconcile conflicting goals: maximizing degradation rate while minimizing catalyst dosage and PMS concentration. This yields an engineered Fe₃O₄/MoS₂ catalyst that achieved 65.04 % cephalosporin degradation within 0.67 h under ultralow dosage conditions, outperforming previously reported benchmarks under similar experimental conditions. Independent validation experiment further confirms the framework's robustness in predicting degradation rate trends and ROS pathways for unseen catalysts.