Machine learning-guided optimization of UV/chlorine process for sustainable micropollutant abatement

UV-based advanced oxidation processes (AOPs) are increasingly adopted for micropollutant removal in drinking water treatment and potable reuse, yet challenges persist due to the proliferation of diverse micropollutants and their high carbon footprint. To overcome these limitations, we developed a machine learning (ML) framework to predict micropollutant degradation rate constants (k) by the UV/chlorine process at 254 nm across varied environmental and operational conditions. External validation with six micropollutants yielded predicted k values comparable to literature-reported results (mean error < 50%), and experimental validation with clarithromycin and simazine also showed close agreement (mean error < 38%), demonstrating the model's generalizability. Feature importance analysis identified aromatic carbons, hydroxyl groups, and C-I bonds as enhancers of k, while nitro groups and heterocyclic nitrogen atoms inhibited degradation kinetics. Applying the validated model, we predicted the degradation of five emerging micropollutants (excluded from the training dataset) in the UV/chlorine process under realistic conditions-tap, surface, and reclaimed water matrices. The model further provided optimal chlorine dosage recommendations for specific micropollutant abatement across different water quality scenarios. The framework was further extended to enable preliminary prediction of halogenated byproduct formation. To facilitate practical use, we created an open-access web interface for water engineers and practitioners. This study highlights the potential of ML to optimize UV-based AOPs, promoting sustainable and efficient micropollutant control in water treatment.