Interpretable Machine Learning-Guided Optimization of Laser-Driven Photocatalytic Degradation of Tetracycline Hydrochloride over ZnO/Cu1.35O/g-C3N4 Composites

Laser-driven photocatalysis provides a wavelength-tunable approach for antibiotic removal, but its performance depends on complex interactions among catalyst properties, irradiation parameters, and operating conditions. This study developed an interpretable machine learning-guided framework integrating entropy weight-based evaluation, predictive modeling, and SHAP analysis to investigate the apparent removal efficiency of tetracycline hydrochloride (TCH) over ZnO/Cu1.35O and ZnO/Cu1.35O/g-C3N4 composites under red, blue, and mixed laser irradiation. At an initial TCH concentration of 20 mg L−1, entropy weight analysis identified 0.02 g ZnO/Cu1.35O under red laser irradiation at 50 μmol m−2 s−1 as the highest-scoring scheme (0.73). At 30 mg L−1, ZCG-3 under mixed red/blue laser irradiation at 200 μmol m−2 s−1 achieved the highest score (0.90). Among five machine learning models, CCO-XGBoost exhibited the best average predictive performance under five-fold group-based cross-validation, with a mean R2 of 0.92 ± 0.06 and a mean RMSE of 8.27 ± 2.79%. SHAP analysis identified reaction time, laser intensity, and initial TCH concentration as the most influential predictive features, reflecting model-based associations rather than physicochemical causality. These findings provide a data-driven framework for treatment-scheme evaluation and UV-Vis-derived apparent TCH removal prediction within the investigated experimental domain