Large-Scale Network-Traffic-Identification Method with Domain Adaptation

With the continuous evolution of the Internet, a variety of web applications, such as video streaming and social network services, are widely used, significantly increasing the amount of traffic. Classifying the types of network traffic in detail is becoming more important from the perspectives of resource management and security analysis. In an Internet service provider (ISP) level large-scale network, details of packet payload cannot be collected due to the huge amount of traffic. Alternatively, network flow data representing the statistics of a sequence of packets are collected from devices. However, due to the lack of packet details, the granularity of classification is limited to the protocol level. For better understanding of the types of traffic in detail, we propose an application-level traffic identification method combining both packet data of small-scale networks and flow data of large-scale networks. With our proposed method, we adapt an identification model trained from small-scale network data to large-scale network data with domain adaptation.