Interpretable prediction of sulfonamide degradation performance during UV/peracetic acid treatment by integrating reaction variables and DFT-derived molecular descriptors

This study investigated the degradation of three sulfonamide antibiotics, sulfamethizole (SMT), sulfapyridine (SPD), and sulfadiazine (SDZ), in a UV/peracetic acid (UV/PAA) system and developed an interpretable machine learning framework to predict their normalized concentration (C/C 0 ) evolution. UV/PAA markedly accelerated degradation compared with UV or PAA alone, with pseudo-first-order rate constants following the order SPD (0.331 min −1 ) > SMT (0.273 min −1 ) > SDZ (0 .1 53 min −1 ). UPLC-QTOF-MS and DFT analyses indicated common transformation pathways but distinct electronic characteristics among the three sulfonamides. The prediction framework integrated a full feature set comprising six reaction variables and three DFT-derived molecular descriptors. The reaction variables included UV intensity, PAA concentration, reaction time, UV fluence, PAA exposure, and UV-PAA coupling. Among six regression models evaluated using curve-level partitioning, support vector regression (SVR) achieved the best performance and maintained high predictive accuracy in the independent hold-out validation set (R 2 = 0.963). Ablation analysis showed that the combined reaction variables provided marginal improvement, whereas the DFT-derived molecular descriptors substantially enhanced prediction. SHAP analysis identified reaction time, UV fluence, and PAA exposure as the dominant contributors, while the molecular descriptors captured electronic differences. The approach provided an interpretable basis for predicting sulfonamide degradation performance and could support water treatment optimization.