Explainable XGBoost modelling of removal efficiency in PMS- and PDS-based advanced oxidation processes
Chemical Engineering Science, cilt.333, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 333
- Basım Tarihi: 2026
- Doi Numarası: 10.1016/j.ces.2026.124220
- Dergi Adı: Chemical Engineering Science
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Chemical Abstracts Core, Chimica, Compendex, INSPEC, zbMATH, Academic Search Ultimate (EBSCO), Engineering Source (EBSCO)
- Anahtar Kelimeler: Advanced oxidation processes (AOPs), Explainable machine learning (XAI), Peroxydisulfate (PDS), Peroxymonosulfate (PMS), Pollutant removal efficiency prediction, XGBoost regression model
- Ondokuz Mayıs Üniversitesi Adresli: Evet
Özet
The performance of peroxymonosulfate (PMS) and peroxydisulfate (PDS) based Advanced Oxidation Processes (AOPs) depends on interrelated factors such as radical generation pathways, solution chemistry, and catalyst–oxidant interactions. This inherent complexity limits the reliability of conventional kinetic and mechanistic models, creating a need for data-driven approaches capable of capturing nonlinear relationships. In this study, a transparent machine learning (ML) framework was developed using a comprehensive literature-based database to predict the removal rates of complex organic pollutants in PMS/PDS-based AOPs. A total of thirty-four input features were compiled, covering pollutant type and concentration, oxidant type and dosage, catalyst type and dosage, and three external energy inputs: UV irradiation, visible light, and ultrasound. Several ML algorithms—Random Forest, XGBoost, Gradient Boosting, CatBoost, Ridge regression, and Extra Trees—were systematically compared. XGBoost delivered the strongest baseline performance, and subsequent hyperparameter tuning further enhanced its accuracy, yielding a test R2 of 0.847 and an RRMSE of 14.6%. The optimized model successfully represented the nonlinear dynamics of sulfate radical-driven oxidation and generalized well across diverse experimental conditions. To strengthen scientific interpretability, explainable AI (XAI) methods were applied, including SHAP analysis, partial dependence plots, interaction studies, and permutation feature importance. Reaction time, initial pollutant concentration, oxidant dose, and catalyst concentration emerged as the dominant predictors. A pronounced interaction between time and oxidant dose aligned with established AOP chemistry. Overall, the proposed XGBoost–XAI framework combines predictive accuracy with mechanistic transparency, offering a practical decision-support tool for designing and optimizing PMS/PDS-based AOPs.