EFS-DNN: An Ensemble Feature Selection-Based Deep Learning Approach to Network Intrusion Detection System

Table 7

Binary-class classification performance on UNSW-NB15 in terms of TPR (%), FPR (%), Acc (%), and F1 score (%). EFS-DNN outperforms all baselines when the number of selected features is 21.

Method	TPR	FPR	Acc	F1 score	# Features
Rule-based [43]	88.91	12.52	88.28	87.01	13
GWO [11]	93.80	20.95	85.68	85.48	20
PSO and GWO [11]	80.83	23.27	78.58	77.25	12
FFA and GA [11]	94.15	24.13	83.99	83.93	15
FFDNN [1]	n/a	n/a	87.10	n/a	21
EFS-DNN	87.54	12.46	88.34	88.14	15