Multi-Model Machine Learning and SHAP Analysis of Fe2+/PMS Systems for Textile Wastewater Treatment

Abstract

This study presents a comprehensive and innovative comparison of homogeneous and heterogeneous peroxymonosulfate (PMS) activation processes for the advanced treatment of real textile wastewater. While conventional methods such as biological treatment and chemical coagulation often have disadvantages such as low pollutant removal efficiency, long processing times, and formation of secondary pollution, the Fe2+/PMS/AC system offers a fast, effective, and sustainable alternative that overcomes the fundamental limitations of conventional wastewater treatment approaches. In homogeneous PMS activation, Fe2+ ions were used as catalysts, while in heterogeneous activation, biomass-based reed activated carbon (RAC) and commercial activated carbon (CAC) were employed, enabling the simultaneous evaluation of sustainable and economical alternatives. Under optimal conditions (pH 7.0, 2 mM Fe2+, 5 mM PMS), the Fe2+/PMS system achieved a high efficiency of 92.1 % in colour removal, while COD and TOC removals reached 29.4 % and 28.2 %, respectively. The addition of activated carbon at a dose of 2 g/L significantly improved process performance, achieving almost complete colour removal, and exceeding 90 % COD and TOC removals in just 10 min. This represents a significant improvement over conventional AOPs in terms of treatment speed and efficiency. High-accuracy models (R-2 > 0.97), including Random Forest, XGBoost, and ANN, were developed for process prediction and optimization. SHAP analysis revealed key parameters influencing pollutant removal and improved model interpretability. The findings demonstrate the strong potential of Fe2+/PMS/AC systems for sustainable textile wastewater treatment and mark a step forward in applying explainable machine learning to advanced oxidation processes.