Table 50 Ablation study.
From: A hybrid deep learning model for detection and mitigation of DDoS attacks in VANETs
Model Variant | Accuracy | Precision | Recall | F1-Score | FPR | FNR | MCC | ROC-AUC |
|---|---|---|---|---|---|---|---|---|
Full Model (CNN + BiLSTM + Attention + Res/Dense) | 0.9918 | 0.9915 | 0.9915 | 0.9915 | 0.0077 | 0.0085 | 0.9917 | 0.9983 |
w/o CNN (only LSTM + Attention) | 0.9587 | 0.9573 | 0.9569 | 0.9571 | 0.0301 | 0.0431 | 0.9424 | 0.9762 |
w/o BiLSTM (CNN + Attention) | 0.9619 | 0.9605 | 0.9582 | 0.9593 | 0.0293 | 0.0418 | 0.9445 | 0.9784 |
w/o Attention Module (CNN + LSTM only) | 0.9601 | 0.9582 | 0.9566 | 0.9574 | 0.0305 | 0.0434 | 0.9429 | 0.9771 |
w/o Residual/Dense Connections | 0.9633 | 0.9612 | 0.9601 | 0.9606 | 0.0287 | 0.0399 | 0.9481 | 0.9796 |
Basic CNN + LSTM (no Attention or Res/Dense) | 0.9495 | 0.9472 | 0.9443 | 0.9457 | 0.0351 | 0.0557 | 0.931 | 0.9687 |
