Comparative assessment of synthetic time series generation approaches in healthcare: leveraging patient metadata for accurate data synthesis

Table 4 Resemblance metric results

Generation Method	Model^a	DLA AUROC	Correlation Similarity	Autocorrelation MAE	\({\boldsymbol{F}_{\boldsymbol{\gamma}}}\)
A1	DGAN_T	0.902 ± 0.010	0.880 ± 0.010	2.601 ± 0.176	0.987 ± 0.006
	WGAN-GP_T	0.900 ± 0.004	0.979 ± 0.004	2.575 ± 0.058	0.980 ± 0.004
	DGAN_M	0.744 ± 0.012	0.929 ± 0.014	1.526 ± 0.043	0.989 ± 0.005
	WGAN-GP_M	0.814 ± 0.016	0.942 ± 0.015	1.527 ± 0.046	0.942 ± 0.005
	DGAN_P	1.000 ± 0.000	0.624 ± 0.091	1.004 ± 0.174	0.970 ± 0.015
	WGAN-GP_P	1.000 ± 0.000	0.743 ± 0.049	0.998 ± 0.074	0.893 ± 0.063
A2	DGAN_T	0.902 ± 0.008	0.823 ± 0.011	3.190 ± 0.358	0.836 ± 0.039
	WGAN-GP_T	0.903 ± 0.040	0.951 ± 0.002	4.160 ± 0.023	0.908 ± 0.010
	DGAN_M	0.890 ± 0.011	0.645 ± 0.118	1.582 ± 0.063	0.758 ± 0.040
	WGAN-GP_M	0.898 ± 0.025	0.797 ± 0.039	1.842 ± 0.045	0.656 ± 0.230
	DGAN_P	1.000 ± 0.000	0.550 ± 0.097	1.944 ± 0.005	0.699 ± 0.097
	WGAN-GP_P	1.000 ± 0.000	0.657 ± 0.050	1.573 ± 0.004	0.000 ± 0.000
A3	DGAN_T	0.901 ± 0.003	0.906 ± 0.003	2.518 ± 0.006	0.945 ± 0.006
	WGAN-GP_T	0.890 ± 0.024	0.894 ± 0.003	5.789 ± 0.074	0.981 ± 0.002
	DGAN_M	0.871 ± 0.004	0.742 ± 0.057	3.985 ± 0.036	0.742 ± 0.017
	WGAN-GP_M	0.972 ± 0.027	0.527 ± 0.010	5.840 ± 0.046	0.627 ± 0.044
	DGAN_P	0.999 ± 0.000	0.865 ± 0.060	3.249 ± 0.244	0.541 ± 0.084
	WGAN-GP_P	1.000 ± 0.000	0.654 ± 0.000	2.082 ± 0.006	0.128 ± 0.037

Values in bold highlight the best-performing approach per trained model and metric
DLA Data Labelling Analysis, AUROC Area Under the Receiver Operating Characteristic Curve, MAE Mean Absolute Error, DGAN DöppgelGANger, WGAN-GP Wasserstein Generative Adversarial Network with Gradient Penalty
^aThe subscripts on the model column refer to the dataset used for the training step

ISSN: 1472-6947