Machine learning integration with response surface methodology to enhance the removal efficacy of arsenate (V)...

Arsenic, a poisonous and carcinogenic heavy metal in drinking water, presents severe health risks to humans, including skin lesions, neurological damage, and circulatory disorders. Despite extensive research efforts have been carried out on removing arsenate (As(V)) using membrane technologies, however, there remains a critical need to further enhance membrane removal efficacy to achieve maximum reusability. Thus, to minimize this challenge, herein we explore the impact of S-functionalized Mxene coating over the surface of quaternary ammonium poly (2,6-dimethyl-1,4-phenylene oxide)/polyvinyl alcohol (QPPO/PVA) anion exchange AEM membrane against As(V) removal. To optimize and validate adsorption efficacy, response surface methodology (RSM) study was carried out using a central composite design (CCD) with R² = 0.995. The significance of these variables was also confirmed by CCD matrix, yielding statistically significant results (p < 0.0001). Adsorption efficacy was further enhanced by employing different machine learning (ML) regression models to finely tune experimental parameters. Among ML models, Random Forest Regression (RFR) has shown highest predictive accuracy (R² = 0.929, RMSE = 4.57 mg L^-1) and identified As(V) concentration as the most influential factor affecting adsorption efficacy. ML-optimization has shown maximum adsorption efficacy at pH (3), contact time (5 min), and As(V) concentration (50 mg L^-1). Freundlich isotherm model has shown adsorption capacity of 413 mg/g (R² = 0.997), while pseudo-second-order kinetic model reflected R² of 0.989. Thermodynamic assessments (ΔG^ᵒ, ΔH^°, ΔS^°) confirmed the process as spontaneous. To the best of our knowledge, this is the first study in which ML is employed to design highly efficient and reliable membranes, providing a novel approach to enhance membrane-based remediation strategies.