Revealing key parameters and mechanisms governing membrane selectivity for dye/NaCl mixtures by explainable machine learning

Artikel i vetenskaplig tidskrift, 2026

Membrane selectivity is crucial for the separation of dye/NaCl mixtures. However, because separation performance is governed by the interplay of multiple factors, including membrane properties, operating conditions, and feed characteristics , identification of the key parameters and underlying mechanisms remains challenging. To address this, this study employed several machine learning methods to predict membrane selectivity and investigatethe mechanisms governing separation. Data were extracted from 157 peer-reviewed papers. Among the four models employed (namely, Random Forest (RF), Support Vector Regression (SVR), Artificial Neural Network (ANN), and Extreme Gradient Boosting (XGB)), XGB gave the best predictive performance with R2 values of approximately 0.7. SHapley Additive exPlanations (SHAP) analysis was applied to evaluate the effect of dye and membrane properties as well as operation-related parameters. The maximum projection radius of the dye, membrane pore radius, membrane water contact angle, and effective membrane area were identified as the four predominant features influencing the separation factors of dye/NaCl mixtures. Model results indicated that high separation factors were achieved under specific conditions, namely, membrane pore radius below 4 nm, membrane water contact angle below 25 degrees, maximum dye projection radius above 1.2 nm, and effective membrane area below 20 cm2. Overall, this study demonstrates the potential of machine learning for predicting membrane selectivity, and providing new insights for the optimization of membrane properties and operating conditions for the treatment of textile wastewater.