Video Traffic Detection in VPN Network Using Lightweight Machine Learning Models

The rapid growth in internet traffic and encryption has made the traditional payload-based classification technique ineffective as it would require big, expensive probes. In this work, we address this issue by exploring the binary classification of network flow into multimedia and non-multimedia categories. The goal is to isolate one part of the traffic using only limited computational resources, thus reducing the workload of the probes.Namely, we want to isolate the multimedia part of the traffic using only a small set of temporal features extracted in real-time. Experimental results show that our lightweight approach rivals more complex models, giving the possibility of deploying our solution on low-cost FPGAs.