LA-GRU: Building Combined Intrusion Detection Model Based on Imbalanced Learning and Gated Recurrent Unit Neural Network

Table 5

Experimental results of different oversampling algorithms on the detection performance of the model.

Method	Metrics	Category					GRU’s Training Time
		Normal	DOS	Probing	R2L	U2L
GRU+Nothing	DR(%)	99.19	99.08	98.94	49.32	37.85	3.88s
	FAR(%)	- -	0.026	0.026	0.694	0.843
GRU+Oversampling	DR(%)	99.15	99.14	99.21	88.59	83.73	5.39s
	FAR(%)	- -	0.020	0.019	0.158	0.193
GRU+SMOTE	DR(%)	99.19	99.24	99.16	98.78	98.24	5.31s
	FAR(%)	- -	0.018	0.027	0.041	0.049
GRU+LA-SMOTE	DR(%)	99.21	99.16	99.20	98.34	98.61	4.53s
	FAR(%)	- -	0.021	0.025	0.036	0.052