CRND: An Unsupervised Learning Method to Detect Network Anomaly

Table 5

The evaluation results with CRND as a pretrained model.

Number of five-tuple	Accuracy (%)	Precision (%)	Recall (%)	False-alarm (%)	F1-score
10	90.07	98.76	79.01	0.81	0.88
50	92.06	99.24	83.06	0.53	0.90
100	95.49	97.73	92.14	1.76	0.95
200	96.13	99.45	91.94	0.42	0.96
500	96.49	99.37	92.83	0.49	0.96
10,000	97.18	99.66	94.08	0.26	0.98
100,000	99.59	99.83	99.26	0.14	0.99