Deep learning-Guided optimization of cobalt catalysts for antibiotic degradation

Confronting the escalating issue of antibiotic pollution in global aquatic environments, our research integrates machine learning with optimization algorithms to improve antibiotic removal in advanced oxidation processes and accelerate inorganic catalyst development. We collected and preprocessed data from 207 papers, focusing on cleaning, standardizing, and encoding. Using exploratory data analysis, we examined the data's intrinsic properties. The TabNet model, applied in the deep learning stage, achieved an 82.02 % accuracy in classification and an R² value of 0.96 in regression. Additionally, we introduced a novel Sparrow Search Algorithm (SSA) to identify optimal experimental conditions. To validate our model, we synthesized single-atom cobalt catalyst Co-CuO, cobalt oxide Co₃O₄, and cobalt ferrite CoFe₂O₄, finding the model's degradation rate predictions within a 2 % error margin. The model, interpreted through SHapley Additive exPlanations (SHAP) analysis, effectively identified key contributors to antibiotic degradation, differentiating between free and non-free radical mechanisms. This research shows theoretical innovation and practical significance in materials science and environmental engineering.