Encrypted Traffic Classification Based on a Convolutional Neural Network

Abstract

To resolve the issues of low accuracy, weak universality, and easy invasion of privacy in traditional encryption traffic classification methods, an encryption traffic classification method based on a convolutional neural network is offered. Firstly, according to the packet size and time message of the net traffic, the original traffic is transformed into a two-dimensional picture to avoid relying on the packet payload to violate privacy, and then the model is embedded. The Inception module performs feature fusion to improve the classification accuracy. Finally, the average pooling layer and the convolution layer are used to replace the fully connected layer, increasing the calculation speed and avoiding overfitting. Experimental results show that the algorithm achieves an accuracy of more than 95% for application traffic classification tasks.