EURASIP Journal on Information Security

Table 2 Accuracy on the WINE dataset (in boldface best results for each k)

From: Feature partitioning for robust tree ensembles and their certification in adversarial scenarios

Model	Parameters					Accuracy
	b	r	p	ml	\(\|{\mathcal {T}}\|\)	\(\phantom {\dot {i}\!}ACC_{A_{0}}\)	k	\(\phantom {\dot {i}\!}ACC_{A_{k}}\)	Δ f-FPF
f-FPF	2	60		8	300	0.972	1	0.861	0.000
	5	27		8	297	0.889	2	0.750	0.000
	5	27		8	297	0.889	3	0.611	0.000
h-FPF	2	60		8	300	0.972	1	0.861	0.000
	5	27		8	297	0.917	2	0.722	−0.028
	5	27		8	297	0.917	3	0.611	0.000
RSM			0.2	8	300	0.944	1	0.833	−0.028
			0.2	4	300	0.972	2	0.722	−0.028
			0.2	8	300	0.944	3	0.444	−0.167
RF				8	300	0.944	1	0.833	−0.028
				4	300	1.0	2	0.389	−0.361
				4	300	1.0	3	0.000	−0.611
RT	1			8	300	0.738	1	0.738	−0.123
	2			8	300	0.738	2	0.738	−0.012
	3			8	300	0.738	3	0.738	+0.127

Back to article page