Anomaly Detection in QAR Data Using VAE-LSTM with Multihead Self-Attention Mechanism

Table 5

Anomaly detection performance on four public benchmark datasets.

Methods	KPI1			KPI2			NAB1			NAB2
		R	F1		R	F1		R	F1		R	F1
LSTMS	0.7639	0.6544	0.7049	0.5850	0.9997	0.7382	0.4536	1.0	0.6241	0.8604	0.7915	0.8779
LSTM-AE	0.7261	0.8521	0.7841	0.6773	0.8230	0.7430	0.7611	0.6807	0.6870	0.7627	0.8733	0.8142
LSTM-VAE	0.7815	0.9545	0.8594	0.8734	0.9271	0.8995	0.7468	1.0	0.8550	0.9090	0.6563	0.7623
VAE-based MHSA-LSTM	0.8221	1.0	0.9023	0.8786	1.0	0.9354	0.8731	1.0	0.9322	0.9547	0.8146	0.8791