Ensemble learning-driven optimization of coagulant dosing for drinking water treatment plants using a scalable framework for smart and sustainable process control

Abstract

Coagulation and flocculation remain foundational in drinking water treatment plants (DWTPs) worldwide. Determining optimal coagulant dosage remains a persistent challenge due to raw water variability and reliance on conventional jar tests, labor-intensive, time-consuming methods, and being prone to operational inefficiencies. This study introduces a novel machine learning (ML) based framework leveraging tree-based ensemble models to predict coagulant dosing with high precision, offering a transformative alternative to empirical approaches. Seven algorithms, Random Forest (RF), ExtraTree, REPTree, and M5P (M5 Prime) Tree, RF-Extra Tree, RF-REPTree, RF-M5P Tree, were systematically evaluated using real-world operational data, incorporating five key water quality parameters: turbidity (Tb), Electric conductivity (EC), pH, temperature, and dissolved oxygen (DO). The performance of the proposed models was assessed based on Root Mean Square Error (RMSE), Nash-Sutcliffe Efficiency (NSE), Kling-Gupta Efficiency (KGE), Willmott’s index of agreement (WI), and the coefficient of determination (R 2 ) values. Among the models, the RF-Extra Tree model exhibited superior predictability (RMSE = 0.515, MAE=0.329, NSE = 0.9850, WI=0.996, KGE = 0.969 and R 2 = 0.985), significantly outperforming traditional heuristics and other models. The RF algorithm also demonstrated robust results (RMSE = 0.807, MAE=0.586, NSE = 0.963, WI=0.99, KGE=0.899, and R 2 = 0.963), highlighting its potential for deployment in dynamic, data-driven treatment environments. This research not only underscores the capacity of ensemble learning to model the complex, non-linear relationships inherent in water treatment but also provides a scalable decision-support tool capable of enhancing treatment consistency, reducing chemical use, and optimizing operational efficiency across diverse geographic and climatic conditions. The proposed methodology holds global relevance for advancing smart water treatment infrastructure, offering both environmental and economic benefits to utilities and stakeholders. • Jar tests are inefficient for optimal coagulant dose prediction • Novel ML-based framework with ensemble models predicts coagulant dosing accurately • Hybrid RF-ExtraTree ensemble achieved highest accuracy, R 2 = 0.985 with lowest RMSE = 0.515 • Standalone RF showed robust predictive capability, enabling real-time dose control • Ensemble models offer scalable, data-driven, & adaptive dosing solutions