Looking deeper: Using deep learning to identify internet communications traffic
Kyle Millar, Dorothé Smit, Christina Page, Aijie Cheng, Hong Gunn Chew, Camille Velasco Lim · Adelaide Research & Scholarship (AR&S) (University of Adelaide) · 2017
Recent years have shown an unprecedented reliance on the internet to provide services essential for business, education, and personal use. Due to this reliance, coupled with the exponential growth of the internet traffic being generated, there has never been a greater necessity for effective network management techniques. Network traffic classification is one key component of this network management which aims to identify the types and quantity of traffic flowing through a network. Previous traffic classification techniques are limited by the use of non-standardised port numbers and the encryption of traffic contents. To tackle these challenges, we propose using deep learning techniques for network traffic classification. This paper investigates the viability of using deep learning for traffic classification with a focus on both network management applications and detecting malicious traffic. Our preliminary results thus far show that a highly accurate classifier can be created using the first 50 bytes of a traffic flow.