AI-Driven Modeling and Statistical Assessment of AgO/ZnO/g-C3N4 Photocatalysts on Wastewater Treatment: Impact of UV-Visible Light

Abstract

Minimizing the environmental impact of dye-laden wastewater, remains a critical challenge with high economic implications. This study focuses on the development and optimization of g-C3N4-based photocatalysts doped with AgO and ZnO at varying Zn (0%-100%) and Ag (0%-2.5%) loadings. Catalysts were applied at four dosages (0.03-0.12 g/100 mL), and their methylene blue degradation efficiencies were evaluated over time intervals up to 3 h. Photocatalysts were synthesized using both conventional and ultrasound-assisted (US-assisted) co-precipitation methods. The US-assisted synthesis yielded improved morphology and dispersion, as evidenced by SEM-EDS and XRD analyses, and enhanced photocatalytic performance. Experimental data were used to train three AI models; artificial neural network (ANN), support vector regression (SVR), and Random forest (RF). SVR exhibited the highest predictive accuracy (R-2 = 0.9854, RMSE = 0.0401), while ANN and RF also showed strong performance (R-2 approximate to 0.980). Model robustness was validated through residual analysis and statistical tests. To assess the influence of input variables, one-way and multi-factor Type II ANOVA were conducted. Zn and Ag content, catalyst dosage, and reaction time were all statistically significant (p < 0.05), with US treatment and Zn loading having the most dominant effects. Ag's contribution was also significant but more composition-dependent.