VTGAN: hybrid generative adversarial networks for cloud workload prediction

Journal of Cloud Computing

Advances, Systems and Applications

Table 7 Comparison of regression results

Model	Window	Training	Time	RMSE	MAPE	MAE	Theil	ARV	POCID	\({R^2}\)
	size	epochs	(Sec.)
VTGAN (LSTM-based)	3	3000	74.2	1.256±0.014	3.075±0.077	1.013±0.021	0.383±0.033	0.039±0.001	79.235±0.444	0.963±0.001
CNN-LSTM	15	235	35	1.776±0.013	4.186±0.058	1.404±0.02	1.310±0.13	0.078±0.005	75.247±0.452	0.927±0.001
Stacked LSTM	20	381	62.7	1.449±0.012	3.444±0.045	1.151±0.009	0.699±0.025	0.053±0.002	74.975±0.685	0.951±0.001
VTGAN (GRU-based)	3	3000	72	1.096±0.013	2.669±0.044	0.887±0.015	0.242±0.037	0.029±0.001	80.490±0.74	0.972±0.001
CNN-GRU	15	209	29.3	1.685±0.04	3.958±0.221	1.314±0.053	1.213±0.025	0.069±0.007	5.345±0.745	0.934±0.003
Stacked GRU	20	255	48.9	1.492±0.012	3.490±0.011	1.155±0.002	0.788±0.032	0.053±0.001	70.524±0.747	0.948±0.001