ENCRYPTED TRAFFIC RECOGNITION WITH ATTENTION MECHANISM

For the problem of poor feature extraction and selection in encrypted traffic identification techniques, this paper proposes to adopt the idea of end-to-end structure, and to build an Attention-CNN encrypted traffic detection model with the attention mechanism. The convolutional neural networks were used to automatically learn features directly from traffic data. The information dynamically captured by the attention layer of the Softmax activation function was used to dynamically weighting the output of the convolutional layer. The fully connected neural network was used for recognition. The ISCX VPN2016 public dataset was used for experiments, and the model was verified by ten-fold cross validation method. The attention feature map and the corresponding session byte information were analyzed in detail. Locations with characteristic information were found and explained the content of encrypted traffic of attention. The experiment results show that this method has a significant improvement over the existing methods. At the same time the evaluation indexes for each type of traffic classification achieve better results.