Table 9 Runtime vs. performance trade-off across configurations.
Configuration | AUC | F1-score | Training time (approx.) | p-value | Remarks |
|---|---|---|---|---|---|
Deep-CTGAN + ResNet + TabNet(proposed) | 0.85 | 0.81 | ~ 42 min (50 epochs) | 0.005 | Best performance, higher compute cost |
Deep-CTGAN + TabNet | 0.82 | 0.79 | ~ 32 min | 0.009 | Lower cost, slight drop in accuracy |
SMOTE + TabNet | 0.79 | 0.76 | < 10 min | 0.015 | Lightweight, moderate gain |
ADASYN + TabNet | 0.78 | 0.75 | < 10 min | 0.020 | Similar to SMOTE |
