Network Traffic Classification Using Deep Learning Neural Networks

Document Type

Presentation Abstract

Presentation Date

10-30-2023

Abstract

Network traffic can be classified into various types, i.e., web browsing, email, chat, streaming, file transfer, VoIP, TraP2P, etc. This technology has been extensively used in Quality-of-Service control, billing, malware detection, etc. The current network traffic classification methods basically have an accuracy of about 87%, and most of them rely on human knowledge to define specific patterns for classification. Our goal of this research includes 1) to increase the classification accuracy and 2) to increase the classification efficiency.

To classify network traffic accurately and automatedly, we are using neural networks, especially Deep Learning (DL) networks, in our research. Specifically, we have focused on three issues. First, neural networks require huge amount of training data. There are some public databases that offer network traffic data, but they are either not suitable for traffic classification or being too small. Therefore, building a comprehensive database for DL network training is our first focus. Another challenge is to determine what network traffic information should be used. The information carried by the raw network traffic data is extremely huge. We have designed various approaches to reduce the information amount while keep the classification accuracy relatively high. The third challenge is the neural networks architecture design and implementation, including DL architecture, the algorithms of each DL layer, the number of hidden layers, the number of neurons in each layer, etc. Through these approaches, we have achieved an accuracy of 92% for feature-based classification and 99% for raw-data-based classification.

Additional Details

October 30, 2023 at 3:00 p.m. Math 103

This document is currently not available here.

Share

COinS