Encrypted Network Traffic Classification Method Based on Deep Capsule Network

Inhalt:
With the increase of users' privacy awareness and the rapid development of network traffic encryption technology, encrypted traffic identification becomes a key task to maintain the security of cyberspace. In this paper, we propose a deep capsule network (DeepCapNet) based encrypted traffic classification method, which replaces the traditional dynamic routing algorithm with a three-dimensional convolutional routing algorithm to build a deep network, reduce the time complexity of the model, enhance the extraction of deep features of network traffic, and improve the classification accuracy. It improves the accuracy of classification. The model introduces a residual capsule layer to effectively solve the gradient disappearance problem of the deep neural network model. Experimental results on the ISCXVPN2016 dataset show that DeepCapNet outperforms existing encrypted traffic classification methods in terms of accuracy and has higher stability when dealing with small datasets.