RSM-ANN-GA framework for predictive modeling and optimization of sonocatalytic eosin dye degradation using ZnO@SiO2 nanocomposites

Abstract

This study investigates the removal of Eosin Yellow, a xanthene-based synthetic dye with low biodegradability and high aquatic toxicity, from industrial wastewater using a ZnO@SiO₂ sonocatalyst. The effects of ultrasonic frequency, pH, catalyst dosage, initial dye concentration, and electrolytes on dye decolorization and Chemical Oxygen Demand (COD) reduction were examined. A hybrid modeling framework combining Artificial Neural Networks (ANN) and Response Surface Methodology (RSM) was developed to optimize the process. ANN architectures with 2, 4, 10, 16, and 20 hidden layers were evaluated, with hyperparameters tuned via Bayesian optimization. Model performance was assessed using MAE, RMSE, and R² with 95% confidence intervals, and parity plots with prediction intervals were generated to ensure predictive reliability. Comparative analysis demonstrated the superior predictive accuracy and generalization ability of the 10-layer ANN over RSM. Electrolyte addition influenced reaction kinetics, while optimization of process parameters enabled efficient dye removal and COD reduction. This work establishes a reproducible framework integrating sonocatalysis with computational intelligence, providing a robust approach for modeling, optimization, and mechanistic investigation of complex dye wastewater treatment systems.