Prediction of chlorination degradation rate of emerging contaminants based on machine learning models

Assessing the degradation of emerging contaminants in water through chlorination is crucial for regulatory monitoring of these contaminants. In this study, we developed a machine learning model to predict the apparent second-order reaction rate constants for organic pollutants undergoing chlorination. The model was trained using second-order reaction rate constants for 587 organic pollutants, with 314 data points obtained from actual experiments, the other data points 273 came from previous studies. We evaluated ten machine learning algorithms with Modred molecular descriptors and MACCS molecular fingerprints, optimizing the hyperparameters through Bayesian optimization to enhance the predictive capability of the model. The optimized model GPR algorithm combined with molecular fingerprint model achieved R²_train = 0.866 and R²_test = 0.801. Subsequently, the model was fed with chemical features of four organic pollutants, and the predicted results were compared with experimentally obtained values, the deviations between predicted and experimental values were found to be 2.12%, 0.37%, 0.15%, and 14.8%, respectively, further validating the accuracy of the predictive model. SHAP analysis showed that the amino-methyl group CN(C)C had the highest feature value, demonstrating the interpretability of the model in predicting chlorine-degraded pollutants The model established in this study is more representative of real chlorination environments, providing preliminary guidance for chlorination plants on the degradation of numerous emerging contaminants lacking treatment standards and facilitating the refinement of strategies for the prevention and control of emerging contaminants.