Classification of Markov Encrypted Traffic on Gaussian Mixture Model Constrained Clustering

Table 1

The comparison of existing recent traffic classification algorithms.

Algorithm	Advantages	Disadvantages
HEDGE [5]	High accuracy	Difficult in feature extraction
Tree-RNN [6]	Small classification	Difficult in feature extraction and model training
DISTILLER [7]	Overcome performance limitations of single-modality DL-based TC proposals	Difficult in feature extraction
SOM [9]	Enrich the state transitions in the Markov chain and construct more distinctive application fingerprints	Ignore the state transition of network communication and have duplicate fingerprints
MAAF [10]	Can accurately classify the applications of the same developer	Poor classification to applications with different developers but similar certificates
MaMPF [11]	High accuracy in real networks	Difficult in feature extraction
LS-CapsNet [12]	Solve difficult in feature extraction	High computation overhead
MGHMM [4]	Need fewer features and computational overhead	Constrained by encryption methods