Sustainable arsenic removal using iron-oxide-coated natural minerals: Integrating adsorption, machine learning, and process optimization

Abstract

We investigated the sustainable removal of arsenite (As(III)) and arsenate (As(V)) from water using iron oxide-coated pumice (IOCP), sepiolite (IOCS), and zeolite (IOCZ) integrated with machine learning (ML) and optimization techniques. Adsorption kinetics followed a pseudo-second-order model, while equilibrium data were best represented by Langmuir and Sips isotherms, indicating chemisorption on heterogeneous surfaces. To predict and optimize performance, Artificial Neural Networks (ANN), Extreme Gradient Boosting (XGBoost), and Random Forest (RF) were applied, with cross-validated results demonstrating the superior accuracy of ANN (R² up to 0.96, RMSE 20–40 µg l^-1). Coupling ANN with Genetic Algorithm and Bayesian Optimization identified global optima for pH, contact time, and initial concentration, yielding residual concentrations of ∼8.1 µg l^-1 (IOCP-As(III)), ∼42 µg l^-1 (IOCS-As(III)), and ∼1.7 µg l^-1 (IOCZ-As(III)), and ∼1.3 µg l^-1 (IOCP-As(V)), ∼28 µg l^-1 (IOCS-As(V)), and ∼6.2 µg l^-1 (IOCZ-As(V)). Compared with trial-and-error conditions (residuals of ∼112 µg l^-1 for IOCS-As(III) and ∼27 µg l^-1 for IOCP-As(V)), the optimized systems reduced chemical usage by up to 65 %, lowered treatment costs to ∼0.004–0.007 $ mg^-1 As, and delivered positive environmental gains exceeding 80 % for IOCP-As(V) and IOCZ-As(III). These results demonstrate that natural mineral-based sorbents, when coupled with AI-driven optimization, can achieve near-complete removal of both As(III) and As(V) at low cost and with reduced environmental footprint, offering a technically robust and scalable framework for sustainable water treatment.