In this present work, polynomial and artificial neural networks models, adjusted by particle swarm optimization, were implemented in an attempt to improve process modelling and to achieve reliable reaction representation behaviour. Parameter optimization of the conversion reaction of ferrous to ferric sulphate in concentrated solutions using hydrogen peroxide as an oxidant is a process of interest due to the importance of producing this ferric salt, a widely used coagulant/flocculant in water treatment, as a chlorine‐free product. Previous work reported a process optimization attempt based on a factorial design of experiments. The obtained polynomial model based on least‐squares showed a minimally satisfactory R2 of 0.7481. The comparison of the obtained models' performance showed a significant improvement in the prediction of experimental conditions. The results indicated that artificial neural networks‐based models presented a higher predictive capability for highly non‐linear experimental data, and the best model achieved an R2 of 0.9744 for the conversion prediction. Optimal ranges for cost‐effective process conditions were investigated through the refined response surface charts obtained.