Table 8 Configuration and evaluation metrics of the best results obtained for the three lines tested in the experiments
From: A novel botnet attack detection for IoT networks based on communication graphs
Test | Hyperparameters | Accuracy | Precision | Recall | F1-score |
|---|---|---|---|---|---|
CGN | Units L1: 7 L2: 5 | Train: 94.29% | |||
Bottleneck: 3 | Test Benign: | 98.09% | 94.81% | 96,42% | |
IQR factor: 0.01 | Test Malicious: | 98.16% | 99.59% | 98.87% | |
Epochs: 10 | Test: 98.16% | 98.13% | 97.20% | 97.65% | |
Batch size: 32 | |||||
CGN w/o BCM, LCM, CCM | Units L1: 7 L2: 5 | Train: 92.8% | |||
Bottleneck: 3 | Test Benign: | 99.76% | 89.05% | 94.10% | |
IQR factor: 0.01 | Test Malicious: | 95.52% | 99.86% | 97.64% | |
Epochs: 20 | Test: 96.7% | 97.64% | 94.46% | 95.87% | |
Batch size: 32 | |||||
Flows | Units L1: 8 L2: 7 L3: 6 L4: 5 | Train: 73.50% | |||
Bottleneck: 3 | Test Benign: | 27.76% | 66.09% | 39.24% | |
IQR factor: 0.01 | Test Malicious: | 97.52% | 88.02% | 92.53% | |
Epochs: 20 | Test: 86.73% | 62.64% | 77.46% | 65.88% | |
Batch size: 32 |