Table 10. MIA result summary on Purchase100 dataset for heavy noise level.

Heavy noise ( $\epsilon =0.5,\delta ={10}^{-3}$)
Slice feature/	Attacker	AUC	Precision	Recall	F1 score	Attack type
Slice value	advantage
Entire dataset	0.06	0.51	0.48	0.48	0.48	Threshold
Entire dataset	0.17	0.51	0.53	0.47	0.50	Logistic regression
Correctly classified true	0.21	0.50	0.49	0.50	0.49	Threshold
Correctly classified true	0.08	0.48	0.49	0.45	0.47	Logistic regression
Correctly classified false	0.18	0.52	0.44	0.52	0.47	Threshold
Correctly classified false	0.13	0.52	0.49	0.55	0.52	Logistic regression