Optimized adsorption removal and capacity prediction of anionic pollutants using a hybrid strategy of machine learning algorithms

Abstract

Accurate prediction of adsorption performance is crucial for optimizing wastewater treatment systems; however, the complex interactions among operational variables and adsorbent properties often limit conventional modeling approaches. In this study, a machine learning framework was developed to predict the adsorption removal efficiency and kinetic capacity of anionic pollutants in aqueous systems. A comprehensive experimental dataset was generated using four representative pollutants, i.e., bovine serum albumin, methyl orange, sulfate, and nitrate, and four adsorbent materials, including powdered activated carbon (PAC), thermally modified PAC, chemically modified PAC, and ion-exchange chitosan beads. Key operational parameters, including pH, contact time, adsorbent dosage, BET surface area, solution volume, and pollutant concentration, were used as input features. Four ML algorithms, i.e., Decision Tree (DT), Gaussian Process Regression (GPR), Support Vector Machine (SVM), and Ensemble Learning Tree (ELT), were developed and further optimized using Bayesian optimization to improve predictive performance. Among the evaluated models, the optimized ELT model demonstrated the highest predictive accuracy with a coefficient of determination (R2) of 0.78, indicating its strong capability in capturing nonlinear adsorption behavior. Model interpretation through partial dependence plots revealed significant influences of pH, adsorbent dosage, BET surface area, and initial pollutant concentration on adsorption performance, while Sobol sensitivity analysis confirmed the dominant role of initial concentration. Experimental validation using jojoba-derived biochar for the removal of methyl orange and Eriochrome Black T dyes showed strong agreement with model predictions. The developed ML models provide a reliable tool for predicting adsorption performance and designing efficient adsorbent-based wastewater treatment systems.