Applying stacked machine learning models to guide electrochemical oxidation of antibiotics: Key parameter identification and process optimization insights

This is the authors' abstract. We don't add key points for this paper.

The continuous accumulation of antibiotics in the environment has become an increasingly concerned global environmental problem. Electrochemical advanced oxidation processes (EAOPs) have been attracted much attention in antibiotic degradation due to their unique advantages, but their effectiveness is influenced by various factors, and how to pinpoint the crucial factors remains unclear. In this study, six machine learning algorithms (i.e., GBDT, XGBoost, SVM, KNN, RF and BPNN) were employed to simulate and predict antibiotic degradation based on a dataset incorporating key features: (i) electrode properties (anode material, cathode type and oxygen evolution potential); (ii) degradation conditions (initial pH, electrode distance, temperature, current density, electrolysis time, electrolyte type and concentration, and antibiotic concentration); (iii) antibiotic properties (pKa and logKow). The optimized GBDT model achieved excellent prediction performance (R² = 0.91, RMSE = 2.21). Feature importance analysis revealed that the degradation conditions, antibiotic properties and electrode properties contributed 69.96 %, 15.4 %, and 14.6 % to the removal efficiency, respectively. SHapley Additive exPlanations (SHAP) further highlighted current density, antibiotic concentration and pKa as critical factors. Additionally, an open-source web application based on stacked models was developed. This work can offer guidance for optimizing experimental design and provide insights into effective strategies for antibiotic pollution control.