FLOWGAN:Unbalanced Network Encrypted Traffic Identification Method Based on GAN

It is crucial to accurately identify the type of traffic and application so that it can enable various policy-driven network management and security monitoring. However, with the increasing adoption of Internet applications use encryption protocols to transmit data, traffic classification is becoming more difficult. Although existing machine learning methods and novel deep learning methods have many advantages, which can solve the drawbacks of port and payload based methods, but there are still some shortcomings, one of which is the imbalanced property of network traffic data. In this paper, we proposed a GAN based method called FlowGAN to tackle with the problem of class imbalance for traffic classification. As an instance of Generative Adversarial Network (GAN), FlowGAN leverages the superiority of GAN's data augmentation to produce synthetic traffic data for classes with few samples. Furthermore, we trained a classical deep learning model, Multilayer perceptron (MLP) based network traffic classifier to evaluate the performance of FlowGAN. Based on the public dataset 'ISCX', our experimental results show that our proposed FlowGAN can outperform an unbalanced dataset and balancing dataset by the oversampling method in terms of data augmentation. Based on the public dataset ISCX, our experimental results show that the recognition performance of FlowGAN on small samples, compared with the unbalanced dataset, Precision, Recall, and F1-score increased by 13.2%, 17.0%, and 15.6% on average, compared with the balanced dataset Precision, Recall, F1-score increased by 2.15%, 2.06%, 2.12% on average.