Efficient and sustainable water decontamination is a critical global challenge, driving the need for advanced approaches to optimize photocatalytic processes. In this study, over 1000 experiments were conducted to evaluate the degradation efficiency of various photocatalytic water decontamination approaches, resulting in a significant dataset. Furthermore, the capabilities of several machine learning models, including Polynomial Regression, Support Vector Regression (SVR), XGBoost, Random Forest, Gradient Boosting, and Artificial Neural Networks (ANN), for predicting degradation efficiency were investigated. Among these, the ANN model demonstrated superior performance, achieving a Coefficient of Determination (R²) of 0.970, a Mean Absolute Error (MAE) of 2.70, and a Mean Squared Error (MSE) of 5.88. According to sensitivity analyses by Shapley Additive exPlanations (SHAP) method, the top three important features were the type of pollutant, its concentration, and the photocatalyst dosage. Notably, SHAP analysis indicated that CuO/TiO₂ was the most efficient photocatalyst among those studied. To further understand the underlying interactions, Molecular Dynamics (MD) simulations were conducted, focusing on parameters such as radius of gyration (Rg), Root Mean Square Deviation (RMSD), Solvent Accessible Surface Area (SASA), and energy profiles. The simulations revealed a high electrostatic interaction energy of -22 kJ/mol and a low RMSD of 3.95 nm, confirming a strong affinity between CuO and TiO₂, consistent with experimental observations. This integration of machine learning models with atomistic level MD analysis establishes a novel framework for optimizing photocatalytic water decontamination and culminates in a predictive tool for forecasting degradation efficiency and enhancing sustainable remediation strategies.
