AI-driven optimization of Congo red photo degradation using the spinel CdCr₂O₄ photocatalyst: From sol-gel synthesis to DT_LSBOOST predictive modeling coupled with the dragonfly algorithm

Abstract

In this study, a nanostructure CdCr₂O₄ spinel photocatalyst was successfully synthesized via a low-cost sol–gel combustion route and thoroughly characterized by XRD, TGA-DTA, SEM-EDS, FTIR, and UV–Vis spectroscopy. The catalyst exhibited a well-defined spinel structure, high crystallinity, and nanometric grain size (∼29 nm), with strong visible-light absorption (band gap ≈ 1.97 eV). Photocatalytic performance was evaluated using Congo red (CR) as a model pollutant under visible LED light. Optimal degradation conditions (pH 6, [CR] ₀ = 10 mg/L, 1 g/L catalyst, 150 min) led to an outstanding removal efficiency of 98.45 %, with a kinetic constant of 2.11 × 10−2 min−1. Mechanistic studies revealed that hydroxyl (•OH) and superoxide (•O₂−) radicals played dominant roles in the degradation process. To model and optimize the system, a hybrid machine learning approach combining Decision Tree with Least Squares Boosting (DT_LSBOOST), optimized using the Dragonfly algorithm, was implemented. The model demonstrated excellent prediction accuracy (R = 0.9998, RMSE = 0.66) and successfully identified optimal operating conditions with <1 % deviation from experimental results. Stability and reusability tests confirmed the photocatalyst retained >90 % efficiency after five successive cycles, with no significant structural degradation. Compared to state-of-the-art materials, CdCr₂O₄ proved highly competitive in visible-light-driven photocatalysis, establishing its suitability for advanced wastewater treatment applications