GTF: An Adaptive Network Anomaly Detection Method at the Network Edge

Table 5

Comparison of metrics obtained by different methods for UNSW-NB15. The results are expressed in %, and means in class .

Model	Overall		Individual
	Baseline	91.11	70.27	84.83	84.95	5.78	8.79	69.19	0	40.33	32.78	41.18	98.18
GTF	92.02	69.87	82.44	84.13	11.57	24.91	69.42	3.14	38.37	36.36	47.78	98.35
GT(F)	90.34	66.42	82.12	82.81	8.62	23.32	68.07	2.18	37.13	28.87	39.31	98.18
TabTransformer	84.59	69.59	84.92	84.13	0	26.35	69.97	0	38.61	0	0	98.14
GBDT2NN	70.36	55.42	77.87	25.93	3.25	22.30	58.38	5.50	26.47	0	0.67	92.63
CatBoost	91.22	70.13	85.16	84.02	0.34	1.11	68.28	0	40.39	0	31.19	98.05
FLAGB	91.04	70.17	84.87	85.01	5.24	8.61	68.93	0	40.14	43.75	39.44	98.15
SPE	90.95	56.55	73.76	61.46	10.66	24.00	60.79	8.84	33.31	7.07	11.57	97.80

Bold values represent top‐2 results.