Enhancing tetracycline degradation prediction with hybrid DBO-XGBoost on g-C₃N₄ photocatalysts

Antibiotic contamination from tetracyclines (TC) poses significant ecological and public health risks, necessitating effective remediation strategies. Current methods for predicting photocatalytic degradation efficiency often overlook complex, nonlinear interactions in dynamic environmental systems. To address this, we developed a Decision-Based Optimization-enhanced XGBoost (DBO-XGBoost) model tailored to predict and optimize TC degradation using graphitic carbon nitride (g-C₃N₄) photocatalysts. By integrating DBO for targeted hyperparameter tuning within the XGBoost framework, the model effectively captures nonlinear dependencies, achieving superior predictive accuracy (R² = 0.95, RMSE = 0.11). Utilizing a robust dataset of 157 experimental data points, we evaluated eight machine learning models, including traditional (SVR, ANN), advanced (CatBoost, LightGBM), and hybrid (GWO-XGBoost) algorithms. DBO-XGBoost outperformed all models in both performance and robustness. Correlation analysis identified specific surface area and light wavelength as critical factors influencing photocatalytic efficiency, aligning with kinetic principles and providing actionable insights for material and process optimization. Additionally, a theoretical analysis assessed the economic and environmental benefits of g-C₃N₄ photocatalysis, demonstrating its cost-effectiveness and reduced environmental impact compared to traditional methods. This study establishes DBO-XGBoost as a powerful predictive and optimization tool and presents a novel framework for advancing photocatalytic systems to effectively mitigate antibiotic pollution in aquatic environments. This study developed a machine learning-optimized photocatalytic system for TC degradation, advancing sustainable solutions in environmental remediation. • A novel hybrid DBO-XGBoost model yields R2 of 0.95 and RMSE of 0.11 for tetracycline degradation by g-C3N4, outperforming eight ML algorithms. • Analysis of 157 data points identifies surface area and light wavelength as drivers of photocatalytic efficiency, validated by SHAP and PD analyses. • The DBO algorithm converges faster ( 40 iterations) than GWO (50 iterations), enabling more efficient hyperparameter optimization in complex environmental systems overall. • Integrating ML predictions with photocatalytic kinetics, plus an economic-environmental framework, shows cost-effectiveness, reduced environmental impact, and actionable material design insights for g-C 3 N 4 .