Machine learning based encrypted traffic classification using estimated application layer statistics

To provide the best Quality of Experience (QoE) to a user, prioritization of traffic flows is required to satisfy different Quality of Service (QoS) requirements of different applications. Traffic flows need to be classified into different QoS classes to prioritize traffic. Since encrypted traffic now comprises most of the Internet traffic, classification of encrypted traffic is necessary for a satellite broadband system. We propose a novel approach to classify encrypted traffic using machine learning (ML). Our approach uses a unique set of features which are based on actual measured application layer characteristics. We show that our technique can achieve an accuracy of over 96%. The approach needs only a few packets to classify a traffic flow and consequently, it can be used as a real-time classifier. The classification type corresponds to a QoS traffic class which is readily amenable to queuing and servicing at a customer premise equipment (CPE). Due to its low complexity, our ML approach can be easily implemented in a constrained embedded device such as a CPE.