Table 3.

Difference (± standard deviation) of ML utility and statistical similarity for Classification between original and synthetic data, averaged on five datasets.

Method	ML utility difference			Statistical similarity difference
	Accuracy (%)	F1-score	AUC	Avg JSD	Avg WD	Diff. corr.
CTAB-GAN+	5.23 ± 1.493	0.090 ± 0.009	0.041 ± 0.003	0.039 ± 0.002	0.010 ± 0.001	2.03 ± 0.039
CTAB-GAN	8.90 ± 1.841	0.107 ± 0.008	0.094 ± 0.004	0.062 ± 0.002	0.013 ± 0.002	2.09 ± 0.031
IT-GAN	8.95 ± 1.911	0.229 ± 0.007	0.183 ± 0.002	0.078 ± 0.001	0.026 ± 0.002	2.63 ± 0.053
TVAE	7.86 ± 2.034	0.181 ± 0.010	0.140 ± 0.004	0.097 ± 0.002	0.017 ± 0.002	2.41 ± 0.055
CTGAN	21.51 ± 3.525	0.274 ± 0.012	0.253 ± 0.006	0.070 ± 0.002	0.025 ± 0.002	2.73 ± 0.097
TableGAN	11.40 ± 2.381	0.130 ± 0.009	0.169 ± 0.004	0.080 ± 0.001	0.055 ± 0.004	2.30 ± 0.078
MedGAN	14.11 ± 4.431	0.282 ± 0.017	0.285 ± 0.006	0.110 ± 0.004	0.159 ± 0.003	2.77 ± 0.181
CWGAN	20.06 ± 4.014	0.354 ± 0.022	0.299 ± 0.006	0.132 ± 0.002	0.136 ± 0.002	2.82 ± 0.167

A lower value indicates a better result. Best result in each column is highlighted in bold.