Abstract: With the development of traffic encryption technology,encrypted traffic has gradually replaced non-encrypted traffic and become the most important part of the current network environment.While protecting users’ privacy,encrypted traffic is also used by malicious software to avoid the defense of traditional intrusion detection system based on the port or payload keywords of traffic,which brings serious threat to network security.In view of the limitations of conventional classification methods,resear-chers try to use artificial intelligence method to classify the application type of encrypted traffic,but the existing researches usually do not make full use of the characteristics of encrypted traffic,resulting in poor performance in the actual complex network environment.To solve the problems mentioned above,this paper proposes an encrypted traffic classification method based on Attention-CNN model.After the preliminary feature extraction of encrypted traffic,we use both BiLSTM+Attention and 1D-CNN model to compress and further extract the temporal and spatial features of encrypted traffic respectively.Finally,one fully connected neural network is used for the final classification based on the obtained mixed features.Experiments are carried out on the ISCXVPN2016 dataset which is the widely used open source dataset in encrypted traffic classification area.Experimental results show that the overall classification precision of the Attetnion-CNN could reach 98.7% and the F1 score is significantly improved compared with several existing studies.

Key words: 1D-CNN, Attention mechanism, BiLSTM, Cyber security, Encrypted traffic