Improving the Performance of Machine Learning-Based Network Intrusion Detection Systems on the UNSW-NB15 Dataset

Table 3

Comparing the proposed system with related works on accuracy.

Method
Class	GV-SVM [7]	RF [8]	CNN [9]	ELM [12]	LCC-MI-SVM-FS [17]	RF [18]	The proposed system
Normal	97.45	99.50	99.7	91.26	75.8	—	98.16
Analysis	—	2.00	0	98.96	99.1	84.67	99.44
Backdoors	—	5.00	0	99.11	99.2	83.53	99.06
DoS	91.24	20.00	0	94.75	94.9	92.12	98.14
Exploits	79.19	99.50	61.8	89.13	84.2	79.21	93.91
Fuzzers	96.39	—	6.8	91.30	91.6	93.43	98.92
Generic	97.51	97.00	97.7	98.16	91.5	96.37	98.34
Reconnaissance	91.51	86.00	0	94.60	95.7	89.45	98.74
Shellcode	99.45	80.00	0	99.40	99.5	92.79	99.92
Worms	—	70.00	0	99.92	99.9	65.31	97.28