Statistical modeling and application of machine learning for antibiotic degradation using UV/persulfate-peroxide based advanced oxidation process

For designing an efficient homogeneous advanced oxidation process (AOP) based wastewater treatment, understanding the performance of multiple oxidants in multiple contaminant matrices is imperative, with appropriate modeling and prediction of the desired performance. In this study, sixteen different combinations of Persulfate (PS) and Peroxide (PO) concentrations were taken to examine the degradation performance of ciprofloxacin (CIP) and metronidazole (MNZ) separately and combined matrix with the UV-assisted AOPs. The 2 mM PS concentration alone degraded to 97.5 % and 99.8 % of CIP and MNZ, and the 2/2 mM PS/PO combination effectively removed 100 % of both antibiotics under 420 min of UV irradiation. The degradation of the combined antibiotic matrix was the same as that of individual degradation. The final pH was found to become acidic after treatment. First-order rate kinetics best fit the degradation kinetic data, and the UV irradiation, PO, and PS contributed 9.9 %, 15.3 %, and 75.7 % of the total removal of the antibiotics. Pearson correlation and statistical multivariate linear regression (MLR) were applied to model the removal% and pH final of both antibiotics, along with the three machine learning algorithms, Artificial neural network (ANN), support vector machine (SVM), and Random Forest (RF), to make the same predictions. Based on RMSE, R 2 , MAE, and MSE values, the optimized ANN and RF (LSBoost) outperformed the SVM and all of the MLR models to predict the % removal and pH final for both antibiotics. The Shapley values of ANN and estimated predictor importance of LSBoost indicated that PS conc and time have the highest impact on % removal and initial pH, and PS conc on final pH modeling. Initiating the ML modeling approach for the homogeneous AOPs with this study, further research is needed to come out with more accurate ML prediction models with larger data sets, higher numbers of variables, and real wastewater samples.