VTGAN: hybrid generative adversarial networks for cloud workload prediction

Journal of Cloud Computing

Advances, Systems and Applications

Table 5 Comparison of regression results

Model	Window	Training	Time	RMS E	MAPE	MAE	Theil	ARV	POCID	\(\varvec{R^2}\)
	size	epochs	(Sec.)
ARIMA	5		21.1	8.434	28.175	6.548	0.882	0.867	77.177	0.544
SVR	5		18.9	1.43	4.51	1.12	0.353	0.022	89.493	0.981
VTGAN (LSTM-based)	3	3000	74.4	0.569±0.16	1.401±0.364	0.464±0.126	0.089±0.078	0.008±0.004	88.726±1.62	0.992±0.005
CNN-LSTM	10	576	74.8	1.043±0.042	2.407±0.091	0.818±0.042	0.364±0.051	0.028±0.004	79.941±2.622	0.975±0.002
Stacked LSTM	20	626	98.7	1.287±0.006	3.003±0.01	1.018±0.006	0.897±0.016	0.046±0.002	84.768±0.453	0.962±0.0003
VTGAN (GRU-based)	3	3000	68.7	0.755±0.118	1.868±0.279	0.611±0.094	0.125±0.068	0.014±0.004	86.961±0.189	0.987±0.004
CNN-GRU	15	244	34.1	0.94±0.033	2.187±0.094	0.747±0.025	0.282±0.023	0.023±0.001	82.446±2.685	0.979±0.001
Stacked GRU	20	394	65	0.934±0.021	2.145±0.043	0.761±0.016	0.269±0.009	0.024±0.001	88.032±0.747	0.980±0.001