Prediction and optimization of antibiotic-resistant bacteria (ARB) inactivation using machine learning and...

The prevalence of antibiotic-resistant bacteria (ARB) in wastewater is growing, posing significant risks to human health and the environment, and requiring more sophisticated data-driven treatment strategies to achieve effective pathogen inactivation. Traditional treatment procedures are often inadequate for removing biological pollutants, highlighting the need for improved and proactive treatment methods. This study used machine learning (ML) models to predict the efficiency of ARB inactivation in Fenton-based advanced oxidation processes. After screening and comparing ten tree-based ML algorithms, Gradient Boosting (GB) demonstrated superior predictive accuracy and robustness for ARB inactivation across various Fenton processes. The GB model fit the training data well and showed low prediction error, as indicated by a high coefficient of determination (R² = 0.99273) and the lowest root mean squared error (RMSE = 0.18190). Using this strong predictive performance, the optimal operational conditions were determined for photo-Fenton treatment, achieving ARB log removals greater than 6.30-log in different water matrices. For real wastewater (RWW), the model-predicted optimal conditions (0.1 mM Fe³⁺, 0.2 mM EDDS, and 0.3 mM H₂O₂ under light irradiation 130 mW/cm² for 76 min) yielded a predicted inactivation of 6.37 log, closely matching the experimentally observed value of 6.30-log inactivation (limit of detection). In ultrapure water, complete inactivation was achieved under identical reagent conditions with a shorter treatment time (25 min), while simulated wastewater required higher reagent dosages (0.2 mM Fe³⁺, 0.4 mM EDDS, and 0.6 mM H₂O₂) to achieve equivalent performance within 28 min. These findings indicate that the GB-guided framework can adequately capture the influence of water matrix complexity on ARB inactivation, enabling targeted optimization of treatment conditions. The study indicates the potential integration of ML models with Fenton-based AOPs to develop effective strategies for combating antibiotic resistance in wastewater systems.