Network Traffic Identification with Convolutional Neural Networks

Abstract

Network traffic identification plays a major role in modern-day network monitoring systems. Most network systems identify traffic based on features such as, flow statistics, static signatures and port numbers. Identifying network traffic is essential, because plenty of information regarding a network flow can be learned by knowing the application protocol associated with it. However, the challenge for traffic classification is to identify features in the network flow data. This paper explores the issue of network traffic identification with neural network and deep learning. A convolutional neural network (CNN) with different optimization algorithms is trained to identify application protocols based on network flow data. The image and text processing hypothesis of the CNN model is extended to naturally fit to the curated dataset. Protocol labels with a high frequency distribution are easily detected by the model, and the results show that the CNN model works equally well on network flow data. A discussion on the performance of different optimization algorithms used with the CNN model is presented.