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

Table 7

Comparison of detection rate among different IDMs for five-category classification.

Method		Normal	DOS	Probe	R2L	U2R
Imbalanced Learning	CANN+SMOTE[9]	N/A	N/A	N/A	92.97	55.91
	MHCVF[10]	94.29	99.99	99.39	80.00	84.05
	DENDRON[11]	98.98	95.94	97.17	83.75	76.92
	I-NGSA[12]	99.58	99.59	96.47	84.61	88.95
Shallow Learning	SVM[2]	96.16	98.06	57.36	22.24	14.29
	OS-ELM[3]	99.07	99.14	90.35	56.75	78.10
	TLMD[4]	99.24	98.57	93.77	56.20	75.71
	GA-LR[32]	99.97	99.98	98.44	95.48	52.17
Deep Learning	CNN+LSTM[13]	N/A	99.10	83.35	74.19	64.25
	S-NADE[14]	99.49	99.79	98.74	9.31	0.00
	DNN[15]	97.43	99.5	99.00	91.00	91.00
	SCDNN[16]	98.42	97.23	80.23	11.4	6.88
Proposed Method	LA-GRU	99.21	99.16	99.20	98.34	98.61