Comparative machine learning models for predicting reactive blue 19 degradation in ultraviolet light emitting diode/chlorine process

UV-LED/chlorine AOP effectively removes persistent pollutants, but its nonlinear behavior complicates accurate degradation prediction. This study develops an integrated experimental–machine learning (ML) framework to predict the degradation efficiency of the textile dye Reactive Blue 19 (RB19) during UV-LED/chlorine oxidation. A dataset of 467 experimental samples covering a broad range of operational conditions, interfering anions, and radical scavengers was used to train and evaluate five ML models: Multiple Linear Regression (MLR), K-Nearest Neighbors (KNN), Random Forest (RF), Extreme Gradient Boosting (XGB), and Artificial Neural Network (ANN). Model performance was assessed using R², RMSE, MAE, and Index of Agreement (IA), while feature importance and SHapley Additive exPlanations (SHAP) were employed to interpret model predictions. Among the tested models, XGB achieved the best performance, with a test R² of 0.9899, RMSE of 3.24%, MAE of 2.17%, and IA of 0.9974, indicating excellent predictive capability without overfitting. UV-LED fluence (23.6%) and reaction time (20.8%) had the highest XGB feature-importance values, followed by sodium sulfite (13.4%) and pH (13.2%). SHAP analysis indicated that higher values of irradiation-related parameters were generally associated with higher predicted degradation efficiency, whereas higher initial dye concentration and sulfite concentration were associated with lower predicted degradation efficiency. Overall, XGB provides a reliable tool for process optimization, control, and reducing experimental effort in textile wastewater treatment.