From data to catalysis: A hybrid experimental verification-machine learning framework for predicting and enhancing visible-light-driven RhB degradation by template-synthesized g-C3N4

Organic dye pollutants have posed a significant threat to aquatic ecosystems. Graphitic carbon nitride (g-C 3 N 4 ), as a metal-free photocatalyst with broad application potential, has been widely explored for the degradation of environmental pollutants. However, its practical efficiency is severely limited by a low specific surface area and a high recombination rate of photogenerated electron-hole pairs. Furthermore, its performance is highly sensitive to synthesis parameters such as method, temperature, and precursor ratio. To address this, this study developed a multilayer perceptron-artificial neural network (MLP-ANN) model to quantify parameter impacts on Rhodamine B (RhB) degradation rate, integrating four synthesis methods (template-free, soft/hard/self-template), material characterization, and mechanistic analysis. MLP-ANN (R 2 =0.9774, outperforming random forest R 2 =0.563) identified melamine (FI=0.33), urea (FI=0.19), and silica template (FI=0.11) as key factors (for porous structure) and catalyst concentration (FI=0.17, increasing active sites) as dominant in reactions. Guided by the model, self-template g-C 3 N 4 (g-C 3 N 4 -SeT) showed optimal performance: k= 0.0613 min −1 , ∼100 % RhB degradation (43.8 times better than template-free g-C 3 N 4 -TF). SEM/TEM confirmed uniform porosity in g-C 3 N 4 -SeT, while UV-Vis DRS and PL showed the enhanced visible-light absorption and suppressed carrier recombination. Superoxide radicals (•O 2 - ) were proved as the main active species. Efficient separation of electron-hole pairs by g-C 3 N 4 -SeT promotes the generation of •O 2 - , thereby accelerating RhB degradation. This work demonstrates the successful integration of machine learning with experimental characterization to optimize both the synthesis strategy and photocatalytic performance of g-C 3 N 4 , establishing a data-driven framework for the rational design of advanced functional catalysts.