A Deep Learning based Traffic Flow Classification with Just a Few Packets

Recently, traffic flow classification has received unprecedented attention due to the introduction of a variety of network applications. Classification plays a crucial role in cyber-security and network management such as resource allocation. The previous studies have shown quite a good performance, but they require a large number of packets in the flow to identify an associated application. In this paper, we propose a deep learning based model for traffic flow classification with just a few packets. We compute five meaningful statistics from the flow and use them as hand-crafted features in the model. Such features when combined with deep learning features, improve the classification accuracy significantly. We evaluate the effectiveness of the model on a real-world traffic dataset that we collected by using a tcpdump utility of Linux. The initial experimental results show that the model can distinguish the traffic types quite accurately with only 15 packets of the flow by carefully extracting the features from the data.