A GAN and Feature Selection-Based Oversampling Technique for Intrusion Detection

Table 7

Performance of different methods on the UNSW-NB15 dataset.

		SMOTE	ADASYN	K-SMOTE	G-SMOTE	ACGAN-SVM	GAN	Proposed
NB	Acc	0.7241	0.7378	0.7265	0.7228	0.7246	0.7243	0.7121
	F1	0.7627	0.7719	0.7643	0.7619	0.7628	0.7629	0.7550
DT	Acc	0.8904	0.8837	0.6534	0.8854	0.8779	0.7269	0.8877
	F1	0.9156	0.9110	0.6670	0.9112	0.9040	0.7569	0.9162
RF	Acc	0.9077	0.8993	0.8820	0.9097	0.8872	0.8629	0.8857
	F1	0.9295	0.9229	0.9090	0.9313	0.9109	0.8896	0.9137
GBDT	Acc	0.9086	0.9200	0.8870	0.8997	0.8862	0.8847	0.9278
	F1	0.9308	0.9408	0.9146	0.9232	0.9101	0.9088	0.9490
SVM	Acc	0.8937	0.9098	0.8679	0.8993	0.8302	0.8309	0.9205
	F1	0.9171	0.9314	0.8954	0.9220	0.8682	0.8688	0.9433
K-NN	Acc	0.8723	0.8738	0.8744	0.8694	0.8666	0.8666	0.8824
	F1	0.8983	0.8997	0.9002	0.8956	0.8931	0.8931	0.9080
ANN	Acc	0.8630	0.7583	0.7397	0.8780	0.8325	0.8712	0.9032
	F1	0.8893	0.7848	0.7649	0.9028	0.8616	0.8970	0.9270