Research on DUAL-ADGAN Model for Anomaly Detection Method in Time-Series Data

Table 2

Comparison of the performance of each anomaly detection model.

Model	RealAdExchange-CPC				RealTraffic-SPEED				RealTraffic-TravelTime				Avg
	Acc	Pre	Recall	F1	Acc	Pre	Recall	F1	Acc	Pre	Recall	F1	F1
K-means [17]	0.737	0.407	0.73	0.523	0.857	0.474	0.6	0.529	0.863	0.64	0.666	0.653	0.568
OC-SVM [20]	0.876	0.571	0.834	0.678	0.892	0.588	0.667	0.625	0.887	0.611	0.785	0.687	0.663
LOF [16]	0.884	0.625	0.769	0.689	0.911	0.692	0.6	0.642	0.9	0.653	0.8	0.719	0.683
IF [19]	0.908	0.785	0.733	0.758	0.937	0.666	0.8	0.727	0.939	0.744	0.842	0.790	0.758
LSTM-AE [22]	0.906	0.75	0.774	0.761	0.938	0.785	0.734	0.758	0.883	0.633	0.679	0.655	0.725
NSIBF [24]	0.96	0.92	0.851	0.885	0.927	0.88	0.815	0.846	0.96	0.892	0.846	0.868	0.866
Mad-GAN [29]	0.915	0.72	0.75	0.735	0.946	0.846	0.734	0.786	0.932	0.783	0.763	0.773	0.765
Fence-GAN [34]	0.914	0.71	0.73	0.72	0.929	0.7857	0.688	0.734	0.94	0.848	0.737	0.789	0.748
Tad-GAN [30]	0.942	0.8	0.889	0.842	0.938	0.789	0.834	0.81	0.938	0.828	0.763	0.795	0.816
DUAL-ADGAN	0.962	0.848	0.967	0.903	0.949	0.818	0.9	0.857	0.96	0.845	0.926	0.883	0.881