Evaluating degradation efficiency of pesticides by persulfate, Fenton, and ozonation oxidation processes with machine learning.

Quantifying organic properties is pivotal for enhancing the precision and interpretability of degradation predictive machine learning (ML) models. This study used Binary Morgan Fingerprints (B-MF) and Count-Based Morgan Fingerprints (C-MF) to quantify pesticide structure, and built the ML model to forecast degradation rates of pesticides by persulfate (PS), Fenton (FT) and ozone oxidation (OZ). The result demonstrated that the C-MF-XGBoost model excelled, achieving R2 of 0.914, 0.934, and 0.971 on test-sets for the above three processes, respectively. The model accurately linked molecular structural variations to degradation rates, demonstrating that impact of molecular structure on the degradation rate was observed to be 12.4 %, 15.2 %, and 21.6 % respectively, across a broader range of SHAP values. Additionally, optimal pH ranges were identified for PS (3.5-5.5) and FT (2.5-4.0), while OZ showed a positive correlation with pH. The model identified electron gain/loss groups' promoting/inhibiting effects on degradation rates and highlighted the significance of N atomic structures in PS. Then, Tanimoto coefficient was used to evaluate the applicability of the model. This study lays a groundwork for quantifying organic compound structures and predicting their degradation impacts, presenting a novel framework to assess future organic pollutants' degradation performance.