Machine learning based efficiency maximization of bio-Fenton SBR for enhanced antibiotic removal

The growing concern over antibiotic pollution (e.g., SMX) and its contribution to antimicrobial resistance (AMR) highlighted the urgent need for enhanced and efficient treatment technologies. The bio-Fenton SBR, with in-situ H 2 O 2 generation and micron-sized magnetite catalysis, represented an innovative and promising process for treating antibiotics. However, the performance of bio-Fenton SBR is critically dependent on operational conditions, such as magnetite dosage and hydraulic retention time (HRT). Here, we employed multiple machine learning models (ANN, SVR, XGBoost, and Decision Tree) to maximize the performance of the bio-Fenton SBR by adjusting two key parameters: magnetite/MLVSS and HRT. Under stable conditions (neutral pH, MLSS 7000-12,000 mg/L, MLVSS ~75% of MLSS, and room temperature), the models successfully captured the nonlinear interactions between the input parameters (magnetite dosages, HRT) and output parameter (overall removal efficiency of COD, SMX, TN, TP). Among the four models, XGBoost and Decision Tree achieved superior predictive accuracy, identifying an operating condition at 0.17-0.19 magnetite/MLVSS and 1.1-2.1 days HRT that could maximize the performance of bio-Fenton SBR. These results strongly demonstrated that machine learning approaches are effective for elucidating the operating conditions of bio-Fenton process for enhanced and efficient treatment of antibiotics. • Machine learning was firstly used to maximize bio-Fenton SBR performance. • XGBoost and Decision Tree showed superior predictive accuracy and stability. • Magnetite range (0.17-0.19/MLVSS) achieved ~90% overall removal efficiency. • HRT of 1.1-2.1 days was identified to maximize bio-Fenton SBR performance.