Machine learning-driven design of nonradical persulfate activation process for water decontamination: a case of Cu-based catalysts

The rational design of heterogeneous persulfate activation systems with predictable mechanism remains a critical challenge for water decontamination. This study proposes a tripartite framework integrating “data-driven discovery, mechanistic decoding, and inverse design”, leveraging machine learning (ML) to simultaneously address the critical challenges of mechanism elucidation and performance prediction in persulfate-based catalysis. A comprehensive database encompassing 1,035 Cu-based catalyst datasets was constructed, enabling systematic evaluation of diverse ML algorithms. The developed Random Forest multi-classifier and CatBoost regressor achieved good performance in predicting dominant reaction mechanisms (radical/nonradical) and contaminant removal efficiency, respectively. To achieve genuine mechanistic decoding, we employed SHapley Additive exPlanations (SHAP) analysis. Moving beyond simple feature importance ranking, this approach quantifies the magnitude and direction of each feature’s contribution to individual predictions and provides nuanced insights into the underlying drivers of different catalytic pathways. Experimental validation demonstrated that their BPA removal rates in persulfate activation systems aligned remarkably with ML predictions (<12.5 %). Importantly, quenching experiments conclusively validated the model’s mechanistic predictions, confirming nonradical mechanism as the predominant contributor to the BPA removal in the designed Cu-based catalytic system. This work establishes a pioneering ML-aided catalyst design paradigm that bridges data science and environmental nanotechnology, advancing the intelligent, mechanism-aware water decontamination process.