Deep Learning for Network Traffic Classification: Methods, Datasets, and Future Directions

Abstract

With the rapid advancement of network technology and the growing reliance on internet services across various sectors, effective network traffic management has become imperative. Network traffic classification plays a crucial role in this management by categorizing and predicting network traffic, thereby aiding in quality control, custom pricing, resource allocation, and testing. Traditional methods such as port-based and signature-based classification have become less effective due to evolving application techniques. Deep learning, a subset of machine learning, offers a robust solution by training classification models without explicit programming and handling complex patterns efficiently.This paper delves into network traffic identification using deep learning techniques, providing an overview of existing methods and analyzing both machine learning and deep learning approaches. A comprehensive examination of relevant literature and datasets is conducted, highlighting key representative datasets used in deep learning-based traffic identification. The study categorizes and evaluates deep learning methods, including multi-layer perceptron, convolutional neural network (CNN), and recurrent neural network (RNN), discussing their respective advantages and limitations. Special attention is given to the shortcomings of CNNs in traffic identification, proposing the use of attention mechanism-based CNNs for improved context recognition. Finally, the paper outlines potential future applications and advancements in deep learning methods for network traffic identification.