Machine Learning Driven Modelling and Optimization of Sonolysis and Electro-Fenton Processes for the Treatment of Pharmaceuticals- Atenolol and Amoxicillin

This study investigates the removal of the pharmaceutical contaminants atenolol and amoxicillin from wastewater using advanced oxidation processes (AOPs), specifically sonolysis and the electro-Fenton process. Furthermore, machine learning (ML) models were developed to predict and optimize the degradation efficiency of these treatment technologies under varying operational conditions. Four ML algorithms- artificial neural network (ANN), linear regression (LR), support vector machine (SVM), and gaussian process regression (GPR) were trained and validated using experimental data from our previous study. The results demonstrated that GPR was the superior predictive model across all scenarios, significantly outperforming the other algorithms. The GPR model achieved high predictive accuracy, with coefficients of determination (R2) of 0.999 for the sonolysis of both pharmaceuticals, 0.999 for the electro-Fenton treatment of atenolol, and 0.9854 for the electro-Fenton treatment of amoxicillin. To optimize the electro-Fenton process, the GPR model was integrated with a Genetic Algorithm (GA). This optimization framework successfully identified the optimal operational parameters for maximal degradation. For amoxicillin, the optimal hydrogen peroxide (H2O2) dosage was determined to be 3.5mM, achieving a predicted degradation of 91.45%. For atenolol, the optimal H2O2 dosage was 3.0mM, resulting in a 94.18% degradation. This study validates the use of a GPR-GA framework as a robust tool for accurately modeling and optimizing complex AOPs for pharmaceutical remediation.