Integrating Plant‐Based Coagulation and Explainable Machine Learning for Sustainable River Water Treatment

Abstract

Industrial wastewater discharge has severely degraded freshwater resources, necessitating the development of sustainable and environmentally benign treatment technologies. This study investigates the performance of three plant‐based natural coagulants, namely, Moringa leaf powder (MLP), banana peel powder (BPP), and neem leaf powder (NLP), for the treatment of Sirsa River water contaminated with organic pollutants and heavy metals. Coagulation–flocculation experiments were conducted under optimized operating conditions, and treatment efficiency was evaluated based on turbidity, chemical oxygen demand (COD), biological oxygen demand (BOD5), and the removal of Fe, Zn, and Hg. Among the investigated biosorbents, MLP exhibited the highest treatment performance, achieving removal efficiencies of 85.4% turbidity, 76.2% COD, 73.7% BOD5, 83.1% Fe, 83.0% Zn, and 81.8% Hg. SEM and FTIR analyses confirmed that hydroxyl, amino, carbonyl, and carboxyl functional groups played a key role in contaminant removal through coagulation–flocculation and surface interactions. To complement the experimental investigation, advanced multi‐output machine learning models, including CatBoost, FT‐Transformer, and TabPFN, were developed to predict water quality under varying operating conditions. Among the evaluated models, CatBoost demonstrated the highest predictive accuracy, achieving R2 values of 0.80–0.86 across the investigated output parameters. SHAP analysis identified biosorbent dosage, solution pH, and contact time as the most influential operational variables governing treatment performance. The integration of experimental biosorption with explainable machine learning provides an effective framework for predicting and optimizing river water treatment processes, highlighting the potential of sustainable plant‐based coagulants for intelligent water quality management.