DeepMal: maliciousness-Preserving adversarial instruction learning against static malware detection

Cybersecurity

Table 2 Experimental results on phd-data. In Tables 1 and 2, the best experimental results are shown in boldface

Detector	Baseline			Obfuscation-based				Feature-level				DeepMal
	Prec	Rec	F1	Prec	Rec	F1	ΔF1	Prec	Rec	F1	ΔF1	Prec	Rec	F1	ΔF1
LR	0.9941	0.9927	0.9934	0.9940	0.9710	0.9824	0.0110	0.9915	0.6811	0.8075	0.1859	0.9259	0.2083	0.3401	0.6533
SVM	0.9970	0.9710	0.9838	0.9970	0.9652	0.9808	0.0030	0.9943	0.5072	0.6717	0.3121	0.9958	0.6956	0.8191	0.1647
RF	0.9606	0.9550	0.9578	0.9595	0.9275	0.9432	0.0146	0.9354	0.5666	0.7057	0.2521	0.9467	0.6956	0.8020	0.1558
Lenet-5	0.9841	0.9884	0.9862	0.9819	0.7894	0.8752	0.1110	0.9009	0.1449	0.2496	0.7366	0.8450	0.0869	0.1576	0.8286
All-Cov	0.9849	0.9492	0.9667	0.9848	0.9470	0.9629	0.0038	0.9333	0.2028	0.3333	0.6334	0.9167	0.1594	0.2716	0.6951
NiN	0.9642	0.9768	0.9704	0.9626	0.9347	0.9485	0.0219	0.8883	0.2884	0.4354	0.5350	0.8275	0.1739	0.2874	0.6830
VGG16	0.9524	0.9421	0.9473	0.9211	0.8994	0.9103	0.0370	0.9244	0.2328	0.5786	0.3687	0.8312	0.1018	0.4665	0.4808
ResNet	0.8999	0.8732	0.8865	0.8534	0.8213	0.8373	0.0492	0.7984	0.2085	0.5035	0.3830	0.8243	0.0876	0.4559	0.4306
Standard DNN	0.8867	0.8813	0.8840	0.8673	0.8575	0.8623	0.0217	0.5000	0.6375	0.5604	0.3236	0.8889	0.1000	0.1798	0.7066