Generation and evaluation of synthetic patient data

Table 8 Average and std of data utility metrics computed on the BREAST large-set

	PCD ↓	Log-Cluster ↓	CrCl-RS (1)	CrCl-SR (1)	Supp. Coverage ↑
Method
IM	8.94 (0.01)	-3.56 (0.12)	0.7 (0.0)	1.0 (0.0)	0.99 (0.0)
BN	4.88 (0.02)	-4.51 (0.38)	0.94 (0.0)	1.0 (0.0)	0.99 (0.0)
MPoM	2.3 (0.01)	-7.83 (0.38)	0.85 (0.0)	1.01 (0.0)	0.99 (0.0)
CLGP	0.69 (0.02)	-8.2 (0.27)	0.95 (0.0)	1.02 (0.0)	1.0 (0.0)
MC-MedGAN	– (–)	-1.39 (0.0)	0.76 (0.05)	0.58 (0.0)	0.16 (0.0)
MICE-LR	1.31 (0.04)	-5.77 (0.13)	0.87 (0.0)	1.07 (0.01)	0.99 (0.0)
MICE-LR-DESC	2.94 (0.07)	-5.09 (0.14)	0.85 (0.0)	1.04 (0.0)	1.0 (0.0)
MICE-DT	0.14 (0.02)	-11.14 (0.53)	1.01 (0.0)	1.0 (0.0)	0.99 (0.0)

CLGP presented the overall best performance. It was not possible to compute PCD metric for MC-MedGAN as the method generated at least one variable with a unique value. The symbols on the right side of metric’s name indicate: ↑ the higher the better, ↓ the lower the better, and (1) the closer to one the better

ISSN: 1471-2288