Few-Shot Learning-Based Network Intrusion Detection through an Enhanced Parallelized Triplet Network

Table 3

Refinement training for detection performance improvement.

Dataset	Method	TP	FP	TN	FN	Precision	Recall	F-measure
CICIDS-2017	Initial training	10,055	655	9,466	534	93.88	94.96	94.42
	Refinement training	10,301	409	9,409	591	96.18	94.57	95.37
UNSW_NB15	Initial training	6,444	218	9,081	719	96.73	89.96	93.22
	Refinement training	6,575	187	9,671	329	97.23	95.23	96.22

The bold values represent the optimal values for the different evaluation criteria in the comparison experiment. For example, Refinement training achieved the best value under the evaluation criterion TP.