Predictive Modeling of Bacterial Inactivation With Hydrogen Peroxide Over a Cobalt Ferrite Catalyst

Hydrogen peroxide activated by a cobalt ferrite catalyst generates hydroxyl radicals that effectively inactivate bacteria. A continuous‐flow packed‐bed reactor containing cobalt ferrite was developed and tested for bacterial disinfection. The inactivation kinetics were highly nonlinear. The classical Weibull model accurately described the initial decline in bacterial counts at low H 2 O 2 concentration (3 mM) but lost accuracy at longer contact times and higher concentrations (6–10 mM). To improve prediction, 10 machine learning algorithms were evaluated. Gradient boosting (GB) and random forest (RF) showed the best performance and were assessed using cross‐validation, error metrics, and independent validation with previously unseen data at 4 mM H 2 O 2 . GB demonstrated superior generalization, whereas RF was more accurate under moderate oxidative stress. A combined GB + RF model, based on averaging predictions from both algorithms, provided the most balanced results, with average errors below ±10% at low‐to‐moderate peroxide concentrations. The findings highlight the limitations of conventional kinetic models for cobalt ferrite‐catalyzed peroxide disinfection and provide a predictive framework for selecting peroxide dose and contact time in continuous‐flow treatment. Independent validation confirmed high predictive accuracy, with larger relative errors observed only at very low residual bacterial counts.