Modeling and optimization of electro-Peroxone degradation of methyl orange using machine learning algorithms

This study examines the electroperoxone process for degrading methyl orange in water and applies a data-driven strategy for predicting and optimizing system performance. Six operating variables were investigated: pH, reaction time, current intensity, ozone dosage, electrolyte concentration, and initial dye concentration, using a combined OFAT and DOE dataset. Under the optimal conditions, pH 3, current 500 mA, ozone 0.36 g/h, Na₂SO₄ 0.05 M, and 50 mg/L dye, the process achieved 99.5% removal within 30 min. Four machine learning models, Extreme Gradient Boosting, Genetic Programming, k Nearest Neighbors, and Support Vector Regression, were compared with a DOE-based regression model. Extreme Gradient Boosting and Genetic Programming showed the strongest predictive accuracy, with R 2 values of 0.78 and 0.71, respectively, whereas the regression model reached 0.52, indicating moderate-to-good predictive performance for the present laboratory dataset. A multi-criteria assessment using TOPSIS identified Extreme Gradient Boosting as the most reliable predictor and highlighted ozone dosage and reaction time as the dominant variables. The integrated DOE ML TOPSIS framework offers a practical and transferable approach for optimizing electroperoxone systems and supports the development of advanced oxidation processes for real wastewater treatment. • A novel integrated framework (DOE–ML–MCDM) for optimizing the electro-peroxone process in wastewater treatment. • Benchmarks advanced ML models against regression approaches to enhance predictive optimization. • Highlights the critical influence of operational parameters on pollutant removal efficiency. • Applies multi-criteria decision analysis (TOPSIS) to rank and select the most robust predictive models. • Provides a scalable and sustainable strategy for improving water and wastewater treatment systems.