Research on Network Traffic Anomaly Detection for Class Imbalance

Abstract

It is actually a classification that the anomalistic detection on traffic of network belongs to. The imbalance of data about network traffic, leading to the disability of anomaly model on learning the characteristics of the minority class samples, makes the detection accuracy in the minority samples poor. Aiming at the imbalance of network flow data, a network traffic anomaly detection method combining Deep Convolutional Generative Adversarial Networks (DCGAN) and Vision Transformer is proposed to make the detection accuracy of network traffic data more perfect. The network traffic dataset is sampled, and the sampled imbalanced dataset is class balanced by DCGAN. The class balanced network traffic data is fed into Vision Transformer for prediction. To simplify the model structure, the vision Transformer model is proposed, and only N encoders are included in the vision Transformer, and the decoder structure is removed. Using residual network in encoder to solve the problem of model degradation. The dataset uses CIC-IDS-2017 network intrusion detection data. After experiments, it was enough to hold that the model which was proposed can positively perform class balancing of data and increase the prediction precision of data about network traffic.