Research Article
Few-Shot Learning-Based Network Intrusion Detection through an Enhanced Parallelized Triplet Network
Table 5
Detection results for different attack types on the CICIDS-2017 dataset.
| Type | CNN | ResNet | XGBoost | Proposed |
| FTP-Patator | 99.80 | 88.17 | 81.50 | 99.50 | SSH-Patator | 97.90 | 98.63 | 98.30 | 98.70 | DoS Hulk | 95.10 | 96.87 | 99.70 | 95.50 | DoS GoldenEye | 94.90 | 79.30 | 98.90 | 95.20 | DoS Slowloris | 95.80 | 97.30 | 99.00 | 96.10 | DoS Slowhttptest | 93.90 | 95.10 | 96.80 | 97.90 | Web Attack-Brute Force | 99.60 | 96.90 | 97.40 | 99.70 | Web Attack-XSS | 94.23 | 93.30 | 99.07 | 97.35 | Bot | 88.10 | 97.66 | 92.30 | 97.70 | DDoS | 93.40 | 92.20 | 99.60 | 91.30 | Port Scan | 97.60 | 82.40 | 83.60 | 91.20 | Heartbleed | 18.18 | 100.00 | 27.27 | 100.00 | Infiltration | 11.11 | 8.00 | 8.33 | 47.22 | Web Attack-Sql Injection | 80.95 | 57.14 | 85.71 | 90.48 |
|
|
The values in bold represent the optimal values for the different methods in the comparison experiments.
|