Online Incremental Learning for High Bandwidth Network Traffic Classification

Table 6

Overall performance comparison.

Dataset	Cambridge		UNIBS
Distance measure	Euclidean	Manhattan	Euclidean	Manhattan
Average accuracy (%)	94.15	94.38	86.71	88.79
Running time (ms/1,000 flow instances)	5.14	1.75	6.93	2.39
Classification time (ms/1,000 flow instances)	4.45	1.49	6.03	2.02
Injecting labeled instance time (s/labeled flow instances)	4.68	1.57	6.47	2.32
Reconstruction time (ms/chunk)	0.18	0.09	0.19	0.11
Memory required (KiB)	140	140	140	140