Advanced oxidation processes (AOPs) based on UV/TiO2 are effective for the removal of organic pollutants in wastewater through the generation of highly reactive species. Machine learning (ML) offers a systematic approach to uncover the relationships between input features and degradation kinetics using large datasets, thereby minimizing experimental workload and supporting process optimization for water treatment. A total of 570 sets of data, including experimental parameters and molecular properties of the pollutants, were collected from 37 previous studies. Four machine learning algorithms, including artificial neural network (ANN), random forest (RF), support vector regression (SVR), and extreme gradient boosting (XGBoost) were optimized via Optuna-based hyperparameter search with five-fold cross-validation and subsequently applied to model the apparent rate constant of organic compound degradation. Results demonstrated that XGBoost achieved the highest predictive performance (R2 = 0.8366, RMSE = 0.2215 on the test set). Model interpretation was conducted using an integrated explainability framework, combining permutation importance, SHAP (SHapley Additive exPlanations) values, and partial dependence plots (PDPs). Initial pollutant concentration and light intensity were identified as the most influential predictors. PDPs revealed that -log(k) increased sharply at low initial concentrations and decreased under higher irradiance. This study demonstrates that data-driven modeling combined with explainable machine learning can accurately predict photocatalytic degradation rates and reveal statistically supported trends that are consistent with established photocatalytic mechanisms. The proposed framework can guide process optimization and pollutant prioritization in UV/TiO2-based water treatment applications.
扫码关注我们
求助内容:
应助结果提醒方式:
