Research Article
GTF: An Adaptive Network Anomaly Detection Method at the Network Edge
Table 4
Comparison of metrics obtained by different methods for KDD’99. The results are expressed in %, and
means
in class
.
| Model | Overall | Individual | | | | | | | |
| Baseline | 76.79 | 69.59 | 97.65 | 74.86 | 21.34 | 2.03 | 0 |
| GTF | 87.71 | 82.46 | 98.44 | 84.19 | 77.47 | 11.11 | 45.07 | GT(F) | 86.59 | 82.80 | 98.54 | 84.53 | 76.18 | 9.51 | 20.37 |
| TabTransformer | 79.58 | 82.33 | 98.39 | 84.45 | 81.31 | 0 | 0 | GBDT2NN | 80.19 | 82.04 | 98.41 | 83.70 | 76.53 | 9.60 | 20.90 | CatBoost | 82.23 | 61.86 | 91.29 | 71.84 | 76.50 | 0 | 0 | FLAGB | 85.31 | 69.98 | 95.96 | 78.64 | 39.87 | 3.40 | 21.13 | SPE | 81.04 | 64.92 | 97.70 | 66.67 | 19.34 | 14.22 | 1.36 |
|
|
Bold values represent top‐2 results.
|