Probing Photocatalytic Degradation of Organic Dyes by ZnO and ZnO/Graphitic Carbon Nitride Composite though Machine Learning

Abstract Zinc oxide (ZnO) and graphitic carbon nitride (g-C₃N₄) are important photocatalysts in pollutant degradation, and ZnO/g-C₃N₄ composite is a novel photocatalyst to mitigate the environmental hazard caused by the release of organic dyes into natural ecosystems. Optimizing this process through developing and integrating high-accuracy models will provide deep insights and enhance photodegradation efficiency by reducing time and labour demands. To this end, a comprehensive experimental dataset was compiled from the scientific literature to train machine learning (ML) models aimed at accurately estimating pollutant photodegradation rate. Several ML models were evaluated on this dataset using a range of accuracy metrics to assess their effectiveness in estimating pollutant photodegradation based on variables such as reaction time, initial pollutant concentration, pH, catalyst dosage, pollutant type, light source (UV or visible), and ZnO weight percentage in the composite photocatalyst. The results demonstrated the superiority of the extreme gradient boosting machine (XGBM), achieving the highest coefficient of determination (R² = 0.97) and lowest root mean squared error (RMSE = 4.9%). Additionally, SHapley Additive exPlanations (SHAP) analysis was conducted to elucidate the contribution of each factor to performance in dye photodegradation, identifying the reaction time, initial dye concentration, and pH as the most significant factors.