This study introduces a dual strategy that combines kinetic modeling and advanced machine learning (ML) models to enhance the prediction and optimization of Rhodamine B (RhB) degradation using an H₂O₂/UV system. As a novel scientific contribution, we report the first-time synthesis of an iron-based catalyst, integrating α-Fe₂O₃ with structural ions (SO₄²⁻, OH⁻, Cl⁻), synthesized via a microwave-assisted hydrothermal method, a rapid and energy-efficient approach conducive to scalability. The synthesized catalyst was thoroughly characterized by XRD, FTIR, SEM-EDX, XPS and BET, confirming its crystalline integrity, surface richness, and robust textural properties.

In the machine learning analysis, CatBoost regression outperformed XGBoost, ERT, and GPR, delivering the highest predictive accuracy for RhB degradation under varying operational conditions of H₂O₂/UV. SHAP (SHapley Additive exPlanations) interpretation revealed that reaction time held the greatest predictive importance, followed by the initial concentrations of H₂O₂ and RhB. Experimental results showed that the α-Fe₂O₃ catalyst consistently achieved complete RhB discoloration within 60 min under illuminated conditions, demonstrating exceptional photocatalytic activity. Most interestingly, the integration of the catalyst with H₂O₂, under both dark and illumination conditions (heterogeneous Fenton and photo-Fenton processes), resulted in complete RhB discoloration in as little as 1 min.

Overall, this work highlights the transformative potential of ML-assisted process design in environmental catalysis and introduces a robust, scalable iron-based material, for water treatment applications, particularly in scenarios with variable light exposure or energy constraints.