Predictive Modeling of Membrane Fouling in Domestic Wastewater Membrane Bioreactor Using Machine Learning

ABSTRACT Membrane fouling is widely recognized as a significant drawback of membrane technology, as it reduces filtration flux and impairs the overall efficiency of wastewater treatment systems. Accurate prediction of membrane fouling, therefore, offers a crucial pathway to optimizing system operation and developing proactive mitigation strategies. This study developed machine learning models—including linear regression, support vector regression, and decision tree regression—to predict transmembrane pressure, a key indicator of fouling severity. Input descriptors such as pH, ammonium, nitrate, and alkalinity, measured at multiple stages of the anoxic–aerobic membrane bioreactor system, were used to train and evaluate the models. Among the tested approaches, nonlinear models—particularly decision tree regression—demonstrated superior performance, achieving high prediction accuracy ( R 2 = 0.99). Moreover, machine learning helped identify the most influential input descriptors and uncover hidden patterns within the collected data. This study presents a promising alternative approach for predicting membrane fouling in wastewater treatment systems.