Base modified mesoporous silica adsorbent for heavy metal adsorption: Optimization of adsorption efficiency with machine learning algorithms

Abstract

In this study, a thoroughly characterized amine-modified mesoporous silica adsorbent was synthesized and used for the extraction of toxic metals such as Ce(III), Hg(II), and Cu(II). The adsorption efficiency was evaluated by optimizing adsorbent dosage, adsorption time, pH, and NaCl concentration. The highest adsorption efficiencies achieved by the amine-modified material were 98% for Hg(II), 97% for Ce(III), and 90% for Cu(II) within 180 min of experimental time. The prepared hybrid materials demonstrated effective adsorption efficiencies for heavy metals. Accurately predicting the adsorption efficiency of heavy metals is crucial for enhancing the efficiency of heavy metal removal techniques in environmental and industrial applications. The adsorption efficiencies of three heavy metals were predicted using a small dataset of 87 samples and fourteen different machine learning algorithms, including linear models, ensemble methods, and support vector machine. The prediction performance was evaluated using various metrics considering both nominal and derived features. SHAP analysis was employed to understand feature dependence and significance about prediction performance. A novel stacking regressor was developed that demonstrated superior performance compared to other methods, achieving a better fit and higher accuracy. Furthermore, our findings underscored the significance of time in optimizing adsorption processes, which was consistently reflected across all feature sets.