Encrypted Traffic Classification with a Convolutional Long Short-Term Memory Neural Network
Классификация зашифрованного трафика с использованием сверточной нейронной сети с долговременной краткосрочной памятью
2018-06-01
SCID: 54.1/gtfmqcgy
Discuss with AI
convolutional long short-term memorydeep learningencrypted traffic classificationflow feature extractionpacket feature extraction
Figures from the paper
Abstract (AI)
With the rapidly emerging encryption techniques for network traffic, the classification of encrypted traffic has increasingly become significantly important in network management and security. In this paper, we propose a novel deep neural network that combines both the convolutional network and the recurrent network to improve the accuracy of the classification results. The convolutional network is used to extract the packet features for a single packet. The recurrent network is trained to pick out the flow features based on the inputs of the packet features of any three consecutive packets in a flow. The proposed model surpasses the existing studies which ask for the first packets of a flow, and it provides more flexibility in real practice. We compare our model with the existing work under deep learning for encrypted traffic classification, based on the public dataset. The experimental results show that our model outperforms the state-of-the-art work in terms of both higher efficiency and effectiveness.
Key Findings
1
A hybrid convolutional long short-term memory network combines single-packet feature extraction with recurrent flow-level feature learning for encrypted traffic classification.
2
On a public dataset, the proposed model outperformed existing deep-learning approaches in encrypted traffic classification in both efficiency and effectiveness.
3
The recurrent component learns flow features from any three consecutive packet features rather than requiring the initial packets of a flow.
4
Using any three consecutive packets provides greater flexibility for practical encrypted-traffic classification than methods dependent on a flow’s first packets.
Research Object
encrypted network traffic flows
Research Subject
packet- and flow-level feature extraction and classification performance using three consecutive packets, including efficiency and effectiveness
Publication Details
Publication Date
2018-06-01
Journal
Publisher
ISSN
Cited by
150
Access Type
Author Information
Download PDF
Subscribe to digest
References available in scid.ai5
ImageNet classification with deep convolutional neural networks2017
Very Deep Convolutional Networks for Large-Scale Image Recognition2014
Adam: A Method for Stochastic Optimization2014
End-to-end encrypted traffic classification with one-dimensional convolution neural networks2017
A survey of methods for encrypted traffic classification and analysis2015