Lightweight Traffic Classification Model Based on Deep Learning

Table 6

Ablation study results in the Bot-IoT-10.

Model	Attention mechanism	Multiscale feature extraction structure	Accuracy (%)	Parameters (M)	Floating-point operations (M)
Contrast model	SE (basic model)	0	97.60	0.15	3.26
	CA	0	98.48	0.17	3.82
	SE	1	99.08	0.20	4.07
Proposed model	CA	1	99.82	0.26	5.26