. 2022 Mar 10;12:4209. doi: 10.1038/s41598-022-07939-1

Table 2.

Results of experiments. Normalized Mutual Information U, fairness (DP-ratio, Flip) and performance (ROC AUC) metrics for different algorithmic approaches described in section “Experiments” on 3 datasets: the Adult public dataset and 2 synthetic datasets generated by samples of model (19). All metrics are expressed in percentage values.

	Adult					Synthetic #1					Synthetic #2
	$U (Y, A) = 4.5$					$U (Y, A) = 93.1$					$U (Y, A) = 20.9$
	FTU	Supp $_{l}$	Supp $_{h}$	CDP	DP	FTU	Supp $_{l}$	Supp $_{h}$	CDP	DP	FTU	Supp $_{l}$	Supp $_{h}$	CDP	DP
$U (\hat{Y}, A)$	4.3	0.5	1.0	3.5	0.3	68.3	0.0	9.6	32.9	0.0	23.0	0.0	3.9	18.5	0.4
ROC AUC	88.4	73.4	76.7	74.9	73.4	98.4	50.4	75.9	75.7	50.5	95.2	52.5	81.7	70.1	53.6
Flip	100.0	100.0	100.0	89.0	77.8	100.0	100.0	100.0	96.8	92.2	100.0	100.0	100.0	79.1	55.8
DP-ratio	32.8	73.5	63.3	37.9	83.7	5.8	99.4	47.5	48.8	99.8	28.5	99.4	62.7	53.4	96.7