Machine Learning Approach for Predicting Perfluorooctanesulfonate Rejection in Efficient Nanofiltration Treatment and Removal

Abstract

Perfluorooctanesulfonic acid (PFOS) is a persistent environmental contaminant posing significant health risks, requiring efficient remediation methods. This study explores the use of advanced nanofiltration techniques, combined with machine learning (ML) optimization, to enhance PFOS removal from water. Key parameters such as membrane type, temperature, PFOS concentration, pH, pressure, and cation presence were analyzed for their influence on PFOS rejection efficiency. Five ML models─multiple linear regression (MLR), lasso regression, ridge regression, random forest (RF), and artificial neural networks (ANN)─were applied to improve predictive accuracy and optimize the filtration process. Data from various studies were analyzed, revealing that PFOS rejection was highly sensitive to trivalent cations and pH changes. The ANN model achieved the highest accuracy (R² = 0.89) in predicting PFOS rejection, followed by RF, ridge, lasso, and MLR, in that order. The study highlights the importance of optimizing operational conditions to improve nanofiltration efficiency. ML integration provided valuable insights into treatment processes, offering practical solutions for more effective water purification. This study provides novel insights into PFOS rejection mechanisms, focusing on operational parameters and their interactions to optimize nanofiltration. It provides practical guidance for improving water treatment efficiency and protecting public health and the environment.