Research Article
Anomaly Detection in QAR Data Using VAE-LSTM with Multihead Self-Attention Mechanism
Table 1
Comparison of anomaly detection performance based on precision, recall, and F1-score.
| Methods | Climb | Cruise | Descent | | R | F1 | | R | F1 | | R | F1 |
| IF | 0.5833 | 0.3293 | 0.4211 | 0.6333 | 0.7039 | 0.6667 | 0.5108 | 0.7055 | 0.5926 | LSTMS | 0.6944 | 0.9801 | 0.8195 | 0.4425 | 0.9440 | 0.6136 | 0.8240 | 0.6352 | 0.7769 | LSTM-AE | 0.8885 | 0.9426 | 0.9147 | 0.8768 | 0.9417 | 0.9134 | 0.7284 | 0.8534 | 0.7860 | LSTM-VAE | 0.7722 | 0.9443 | 0.8496 | 0.8119 | 0.9716 | 0.8961 | 0.8902 | 1.0 | 0.9419 | VAE-based MHSA-LSTM | 0.9145 | 0.9833 | 0.9503 | 0.8840 | 1.0 | 0.9384 | 0.9453 | 1.0 | 0.9718 |
|
|