Optimization of Fe3O4/TiO2-based biopolymer nanocomposite for heavy metal removal from aqueous solution using artificial neural network modelling and response surface methodology

This study applied Response Surface Methodology (RSM) and Artificial Neural Network (ANN) modeling to optimize and predict the removal efficiencies of chromium (VI), cadmium (II), and lead (II) ions from aqueous solutions using a Fe 3 O 4/ TiO 2 banana peel biopolymer nanocomposite. Batch adsorption experiments were conducted using a fixed solution volume of 100 mL and designed according to a Box–Behnken experimental design to investigate the effects of pH, initial metal ion concentration, adsorbent dosage, and contact time. The RSM quadratic model for Cr(VI) (F = 5.91, p = 0.0010; R 2 = 0.86) and the linear models for Cd(II) (F = 6.70, p = 0.0005; R² = 0.87) and Pb(II) (F = 2.95, p = 0.0406; R 2 = 0.33) were statistically significant, identifying the dominant factors influencing adsorption efficiency. Optimization using the desirability function predicted maximum removal efficiencies of 95.95% for Cr(VI) at pH 8.4, initial concentration 47.30 mg/L, adsorbent dosage 16.7 g/L, and contact time 56.77 min; and complete removal (100%) for Cd(II) and Pb(II) at pH 6.5, initial concentration 30 mg/L, adsorbent dosage 15 g/L, and contact time 65 min. The ANN models demonstrated superior predictive performance compared to RSM, with high coefficients of determination (R 2 = 0.97, 0.98, and 0.95 for Cr(VI), Cd(II), and Pb(II), respectively) and low prediction errors (RMSE = 0.464–8.88; MAE = 0.375–5.45), effectively capturing nonlinear interactions among process variables. The Fe 3 O 4/ TiO 2 /banana peel nanocomposite exhibited high adsorption efficiency, magnetic recoverability, and strong potential for simultaneous multi-metal removal, while a hybrid RSM–ANN modeling framework provided robust process optimization and accurate predictive capability.