Network Traffic Classification Based on SD Sampling and Hierarchical Ensemble Learning

Table 3

Related works of balanced datasets.

Detailed method	Literature	Description	Dataset	Best accuracy (%)
Data	[26]	The variational autoencoder generates samples	UNSW-NB15	96.13
	[27]	The wGAN-GP method generates samples	NSL-KDD/UNSW-NB15/CICIDS2017	86.69/94.90/99.84
	[28]	A three-point domain sample generation method based on the SMOTE algorithm	NSL-KDD	99.00
	[29]	The SVR model predicts SMOTE sampling proportion	KDD Cup 1999	98.10
	[30]	The SMOTE algorithm combining clustering and instance hardness	DoHBrw2020/CIC_Bot/CIC_Inf/DOS2017/UNSW/Botnet2014	AUC = 89.60/90.21/92.09/75.92/93.19/73.81
	[31]	A technique for sampling samples based on the difficulty of sample classification	NSL-KDD/CICIDS2018	82.84/96.99
	[32]	A method combining TGAN and slow start	KDD cCup 1999	93.98
	[33]	An encrypted traffic generation method based on GAN	ISCX	99.10
Algorithm	[34]	A cost-sensitive deep neural network	NSL-KDD/CIDDS-001/CICIDS2017	92.00/99.00/92.00
	[35]	A method of weighted extreme learning machine	UNSW-NB15/KDD cup 1999	96.12/99.71
	[36]	The HM-loss cost method	Personal real data	F1 = 87.00
	[37]	A method of batch balancing datasets based on deep learning	CHB-MIT/BonnEEG/FAHXJU	95.96/100.00/87.93