Predicting micropollutant removal through nanopore-sized membranes using several machine-learning approaches based on...

Area of Science:

• Environmental Engineering
• Water Treatment Technologies
• Membrane Science

Background:

• Predicting micropollutant separation efficiency using functionalized membranes is challenging due to complex interactions.
• Physicochemical properties of micropollutants and membrane characteristics influence separation efficacy.

Purpose of the Study:

• To compare the predictive performance of various machine learning (ML) tools on a modest dataset for micropollutant removal.
• To evaluate ML models for functionalized reverse osmosis (RO) and nanofiltration (NF) membranes using micropollutant and membrane attributes.

Main Methods:

• Utilized supervised algorithms: adaptive network-based fuzzy inference system (ANFIS), linear regression (LR), stepwise linear regression (SLR), and multivariate linear regression (MVR).
• Employed unsupervised algorithms: support vector machine (SVM) and ensemble boosted tree (BT).
• Performed feature engineering and parametric dependency analysis to identify key input variables.

Main Results:

• Micropollutant characteristics, including maximum projection diameter (MaxP), minimal projection diameter (MinP), molecular weight (MW), and compound size (CS), positively correlated with removal efficiency.
• Model combinations with these key variables showed high prediction accuracy for both supervised and unsupervised ML.
• An ANFIS-grid partitioning (NF-GP) model achieved the highest accuracy (R² = 0.965) with low error metrics (RMSE = 3.65, MAE = 3.65).

Conclusions:

• Key micropollutant properties (MaxP, MinP, MW, CS) are critical for efficient micropollutant rejection in real-time filtration.
• The ANFIS-GP model effectively handles complex data aspects, improving prediction of rejection efficiency.
• Findings can guide the design of self-prepared membranes with optimized pore sizes for enhanced micropollutant separation.