Research Article
A Novel Way to Generate Adversarial Network Traffic Samples against Network Traffic Classification
Table 8
Untargeted defense on LeNet-5 and transferability on Vgg-16.
| | Methods of crafting perturbation | Traffic application class | Deceiving rate on LeNet-5 | L2 norm | L0 norm | SSIM | Time consuming
(second) | Transferability deceiving rate on Vgg-16 |
| | L-BFGS | Geodo | 6.40% | 8.31% | 73.28% | 74.86% | 13 | 31.25% | | Neris | 53.60% | 2.58% | 65.20% | 91.04% | 13 | 62.31% | | Virut | 46.00% | 2.71% | 68.96% | 93.72% | 13 | 60.87% | | Cridex | 0.20% | 5.00% | 81.00% | 86.53% | 15 | 100.00% | | Average | 26.55% | 4.65% | 72.11% | 86.54% | 14 | 63.61% |
| | FGSM | Geodo | 100.00% | 10.83% | 51.24% | 38.28% | 2 | 38.28% | | Neris | 90.20% | 8.79% | 60.45% | 54.64% | 2 | 98.00% | | Virut | 95.40% | 7.74% | 70.86% | 77.97% | 2 | 58.91% | | Cridex | 90.80% | 8.26% | 76.53% | 76.33% | 2 | 44.05% | | Average | 94.10% | 8.91% | 64.77% | 61.81% | 2 | 59.81% |
| | C&W | Geodo | 100.00% | 1.00% | 50.00% | 99.59% | 151 | 0.00% | | Neris | 100.00% | 1.00% | 42.00% | 99.99% | 89 | 0.40% | | Virut | 100.00% | 1.00% | 58.00% | 99.98% | 110 | 1.20% | | Cridex | 100.00% | 1.00% | 75.00% | 99.88% | 134 | 0.20% | | Average | 100.00% | 1.00% | 56.25% | 99.86% | 121 | 0.45% |
| | JASM | Geodo | 99.80% | 11.38% | 4.52% | 63.30% | 135 | 61.12% | | Neris | 96.40% | 10.12% | 5.36% | 65.20% | 137 | 62.24% | | Virut | 86.20% | 8.91% | 5.86% | 71.04% | 135 | 36.19% | | Cridex | 99.80% | 7.89% | 5.26% | 72.98% | 136 | 31.66% | | Average | 95.55% | 9.56% | 5.25% | 68.10% | 136 | 47.80% |
|
|
