With the rapid development of network technology, network traffic data shows an explosive growth. How to accurately identify abnormal traffic from these massive data is of great significance to ensure network security and improve network service quality. This paper proposes a new network traffic anomaly detection method, namely, GANs-T(Generative Adversarial Networks and Transformers). This approach combines the strengths of GAN in data generation and feature learning with the power of Transformer in processing sequence data. The GANS-T network model is constructed, which is composed of generator, discriminator and Transformer encoder. The experimental results show that the GANs-T network is superior to the traditional FGAN, N-GAN, RNN and CNN network in detection accuracy and F1 score. Especially when dealing with traffic data of medium scale and high data volume, GANS-T network shows higher stability and robustness. In addition, we found that at very low data volumes, while the detection performance of all methods decreased, GANS-T networks still performed relatively well, thanks to their strong feature learning ability and generalization performance. The research in this paper not only provides a new idea and method for network traffic anomaly detection, but also provides a useful reference for the subsequent research work.
Yin et al. (Fri,) studied this question.