A data-driven chemometric model was applied to sulfonated MXene-based TFN membranes to predict desalination performance. Using models like k-SVM-PSO, RF, and DT, the framework accurately predicted flux, salt rejection, fouling dynamics, and chlorinated membrane FTIR spectra. Membrane surface and electrolyte properties influenced output prediction. The hybrid model (k-SVM-PSO) achieved superior accuracy, showing promise for real-time membrane design and process optimization.

Abstract

Machine learning (ML) has emerged as a valuable tool in advancing thin-film nanocomposite (TFN) membranes for sustainable water treatment. In this study, a novel data-driven chemometric framework integrated with ML was developed to predict the performance of sulfonated MXene-incorporated TFN polyamide membranes for desalination applications. Six ML algorithms namely kernel support vector machine (k-SVM), decision tree (DT), long short-term memory (LSTM), recurrent neural network (RNN), random forest (RF), and a hybrid k-SVM model optimized using particle swarm optimization (k-SVM-PSO) were systematically evaluated. Feature engineering of sulfonated MXene concentration, membrane characteristics such as contact angle, surface roughness, surface charge, and physiochemical properties of electrolytes showed prominent control in electrolyte flux and rejection. The metaheuristic-driven k-SVM-PSO model showed outstanding predicted accuracy for electrolyte flux with R² = 0.983, based on physicochemical properties parameters of TFN membrane and sulfonated MXene and electrolytes. Predictive ML algorithms also strongly agreed with the experimental dataset in determining non-linear flux dynamics related to organic foulants fouling patterns. Furthermore, the spectral intensity of chlorinated TFN membranes was successfully predicted, with DT and RF models achieving the highest performance (R² = 0.999 and minimal error metrics). This chemometric approach enables advanced prediction of membrane performance for desalination.