Table 9.
Performance of anomaly detection models by varying the size of the training set in HAI.
| Model | Training Set Size | F1-Score | Precision | Recall | FNR | FPR | TP | FN | FP |
|---|---|---|---|---|---|---|---|---|---|
| 20% | 75.9% | 67.6% | 86.5% | 13.5% | 0.9% | 7738 | 1209 | 3712 | |
| 40% | 77.1% | 66.4% | 91.9% | 8.1% | 1.1% | 8226 | 721 | 4158 | |
| InterFusion | 60% | 75.8% | 69.4% | 83.6% | 16.4% | 0.8% | 7476 | 1471 | 3301 |
| 80% | 80.2% | 74.8% | 86.4% | 13.5% | 0.7% | 7734 | 1213 | 2610 | |
| 100% | 78.9% | 74.4% | 83.9% | 16.1% | 0.6% | 7504 | 1443 | 2579 | |
| 20% | 69.1% | 86.7% | 57.4% | 42.6% | 0.2% | 4775 | 3543 | 731 | |
| 40% | 88.5% | 89.2% | 87.8% | 12.9% | 0.3% | 7305 | 1013 | 882 | |
| RANSynCoder | 60% | 71.3% | 89.7% | 59.1% | 40.9% | 0.2% | 4918 | 3400 | 563 |
| 80% | 70.8% | 77.5% | 65.1% | 34.9% | 0.5% | 5417 | 2901 | 1572 | |
| 100% | 82.9% | 89.1% | 77.6% | 22.4% | 0.2% | 6452 | 1866 | 793 | |
| 20% | 31.2% | 85.0% | 19.1% | 80.9% | 0.1% | 1708 | 7239 | 301 | |
| 40% | 45.5% | 63.3% | 35.5% | 64.5% | 0.5% | 3178 | 5769 | 1846 | |
| GDN | 60% | 53.1% | 65.4% | 44.4% | 55.6% | 0.5% | 3975 | 4972 | 2054 |
| 80% | 55.9% | 73.3% | 45.3% | 54.7% | 0.4% | 4055 | 4893 | 1472 | |
| 100% | 59.7% | 78.5% | 48.3% | 54.0% | 0.2% | 4323 | 4624 | 1054 | |
| 20% | 15.9% | 9.0% | 71.3% | 28.6% | 16.4% | 6383 | 2564 | 64,573 | |
| 40% | 72.2% | 79.0% | 66.4% | 33.6% | 0.4% | 5944 | 3003 | 1581 | |
| LSTM-ED | 60% | 71.8% | 80.3% | 64.9% | 35.1% | 0.4% | 5807 | 3140 | 1421 |
| 80% | 72.3% | 80.0% | 65.9% | 34.1% | 0.4% | 5895 | 3052 | 1476 | |
| 100% | 71.7% | 79.1% | 65.5% | 34.5% | 0.4% | 5864 | 3083 | 1547 | |
| 20% | 60.5% | 92.5% | 44.9% | 73.0% | 0.1% | 2229 | 6244 | 383 | |
| 40% | 58.6% | 94.8% | 42.4% | 73.8% | 0.1% | 2231 | 6312 | 354 | |
| USAD | 60% | 59.7% | 81.5% | 47.1% | 70.9% | 0.1% | 2485 | 6058 | 608 |
| 80% | 61.1% | 88.4% | 46.7% | 71.8% | 0.1% | 2407 | 6136 | 467 | |
| 100% | 58.8% | 76.0% | 48.0% | 71.3% | 0.2% | 2447 | 6096 | 821 |