Machine learning prediction of permeate and retentate water quality using operational and environmental parameters

ABSTRACT Graph showing MLP outperforming other ML models in predicting water quality, with 98% accuracy. Permeate and retentate water are essential streams from the water filtration process for evaluating the efficiency of a water treatment system. The permeate quality is a crucial indicator of a system's efficiency in eliminating water contaminants and generating water that meets regulatory criteria for potable or industrial use. At the same time, retentate water quality regulation is vital for system efficiency, environmental compliance, and optimizing waste treatment procedures. In this study, we explored the potential of machine learning (ML) models for the prediction of permeate and retentate water quality in water treatment systems based on operational and environmental variables in the dataset from the National Alliance for Water Innovation seedling project, Oak Ridge National Laboratory. ML models such as logistic regression, support vector machine, multilayer perceptron (MLP), AdaBoost, Gaussian naive Bayes (GNB), and random forest were trained and evaluated for accuracy, precision, recall, and F-1 score. The result showed that MLP ranked highest based on 0.98, 0.97, 0.98, 0.98, and 1.00 as accuracy, precision, recall, F-1 score, and Area Under the Curve (AUC), respectively. In conclusion, ML models are proficient in the prediction of water quality, which enhances the efficacy of water treatment plants and guarantees adherence to regulatory standards.