Research Article
A Novel Way to Generate Adversarial Network Traffic Samples against Network Traffic Classification
Table 9
Targeted defense (MySql as the targeted class) on LeNet-5 and transferability on Vgg-16.
| Methods of crafting perturbation | Traffic application class | Matching rate on LeNet-5 | L2 norm | L0 norm | SSIM | Time consuming (second) | Transferability deceiving rate on Vgg-16 |
| L-BFGS | Geodo | 100.00% | 1.07% | 66.19% | 98.65% | 68 | 0.00% | Neris | 100.00% | 1.24% | 72.76% | 97.71% | 69 | 0.00% | Virut | 94.60% | 1.15% | 75.26% | 99.21% | 68 | 0.00% | Cridex | 100.00% | 1.11% | 78.59% | 99.66% | 66 | 0.00% | Average | 98.65% | 1.14% | 73.20% | 98.80% | 68 | 0.00% |
| FGSM | Geodo | 10.60% | 10.54% | 46.00% | 36.26% | 2 | 0.00% | Neris | 0.20% | 8.00% | 72.00% | 78.37% | 2 | 0.00% | Virut | 1.80% | 8.33% | 67.22% | 76.67% | 2 | 0.00% | Cridex | 2.20% | 8.27% | 76.27% | 80.75% | 2 | 0.00% | Average | 3.70% | 8.79% | 67.37% | 68.01% | 2 | 0.00% |
| C&W | Geodo | 100.00% | 1.00% | 51.00% | 99.71% | 174 | 0.00% | Neris | 100.00% | 1.00% | 55.00% | 99.86% | 135 | 0.00% | Virut | 100.00% | 1.00% | 66.00% | 99.89% | 139 | 0.00% | Cridex | 100.00% | 1.00% | 76.00% | 99.73% | 167 | 0.00% | Average | 100.00% | 1.00% | 62.00% | 99.80% | 154 | 0.00% |
| JASM | Geodo | 93.60% | 7.43% | 2.62% | 75.24% | 136 | 0.00% | Neris | 61.40% | 11.10% | 5.24% | 59.74% | 135 | 0.33% | Virut | 28.80% | 10.27% | 6.06% | 64.65% | 135 | 0.69% | Cridex | 69.80% | 8.62% | 5.65% | 70.47% | 135 | 0.00% | Average | 63.40% | 9.36% | 4.87% | 67.53% | 135 | 0.26% |
|
|