Machine learning-Powered estimation of simultaneous removal of sulfamethoxazole, 17-β Estradiol, and carbamazepine via photocatalytic degradation with M-Al@ZnO

Abstract

The recalcitrant nature of emerging contaminants in water has raised serious concerns, and addressing their removal aligns with the aims of SDG 6. This has necessitated research on photocatalysis, but its cost-intensive nature requires thorough optimization, which is a tedious manual process. Hence, in this study, different machine learning (ML) models (Elastic-Net, Lasso, XGBoost, Gradient Boosting (GB), Random Forest (RF), and Artificial Neural Network (ANN)) have been used to model the photocatalytic degradation of sulfamethoxazole, 17β-Estradiol, and carbamazepine in the presence of M-Al@ZnO. The training dataset included removal of the 3 contaminants, with pH varied from 2 to 10 (M-Al@ZnO dose = 0.5 g/L and contaminant concentration = 1000 μg/L), M-Al@ZnO dose varied from 0.1 to 1 g/L (pH = 8, and contaminant concentration = 1000 μg/L), and contaminant concentrations varied from 500 to 2000 μg/L (pH = 8 and M-Al@ZnO dose = 0.5 g/L). Across all the models, GB exhibited the most promising results (R²: 0.9648 and RMSE: 3.9581). SHAP analysis revealed that irradiation time (∼60–85 %) was the dominant factor, followed by pH (∼20–45 %) and dose (∼5–15 %), with pollutant concentrations having minimal or negative impact on removal. The models were then used to predict data at non-experimented points (pH varied between 2 and 10, dose varied between 0.1 and 1/g/L, and time varied between 20 and 120 min). Contour plots generated using GB-predicted data explained the interactive effects of dependent variables and hinted that at optimum conditions (pH = 8, Dose = 0.7 g/L), the system can effectively remove 90 % of contaminants within 90 min.