The Advancements of Deep Learning Approaches for Abnormal Network Traffic Detection

Abstract

For network technology, abnormal traffic detection is crucial to ensuring network security. In order to address abnormal traffic, remarkable results have been achieved when deep learning methods are applied to network abnormal traffic detection. This paper first introduces the significance of analyzing network traffic and the impact of abnormal traffic and then focuses on discussing abnormal traffic detection methods based on deep learning, which are divided into three major kinds: Deep Neural Network (DNN), Convolutional Neural Network (CNN), and Recurrent Neural Network (RNN). For DNN, it covers Multilayer Perceptron Based DNN, Multi-layer DNN with Adaptive Parameter, and Two-Level Detection System. CNN-related methods include Wavelet-Enhanced CNN and hybrid architectures with RNN or autoencoder. RNN-based approaches involve Hierarchical Hybrid RNN, Session Gate RNN, etc. The discussion section is about the challenges associated with deep learning, including real-time requirements, low interpretability, and the need to address diverse network attacks. It also proposes future directions for deep learning, such as online learning, improving model interpretability, and adversarial training. Finally, the conclusion section states that this article provides a comprehensive review, helping readers better understand the development and future directions of deep learning in abnormal traffic detection.