Table 3.

Comparison of synthetic dataset vs. real-world dataset (NSL-KDD/CICIDS2017).

Metric	Synthetic dataset	Real-world dataset (NSL-KDD/CICIDS2017)
Accuracy (%)	85.00%	95.00%
Precision (%)	84.20%	94.50%
Recall (%)	83.50%	95.20%
F1-Score (%)	83.80%	94.80%
AUC-ROC Score	0.89	0.97
Training Loss	0.31	0.1832
Validation Loss	0.29	0.1541
Time Complexity (Training Time)	Moderate	Lower
Epochs for Convergence	20 epochs	10 epochs
Overfitting Risk	Moderate	Lower (with Dropout)
Robustness to Noise/Attacks	Moderate	High