MFSI: Multi-flow based service identification for encrypted network traffic

Abstract

Encrypted traffic identification plays a crucial role in improving service quality, optimizing network management, and maintaining network security. Various machine learning and deep learning based methods have been proposed to address the challenge of identifying encrypted traffic. However, existing methods face two main challenges. First, they are easily affected by interfering traffic, which reduces the accuracy of identifying target traffic. Second, they rely on expert annotations to identify unknown applications. In this paper, we propose a multi-flow based method, namely MFSI, for identifying the service of encrypted network traffic. MFSI treats multiple flows as the classification unit to reduce the impact of interfering flows and constructs a robust graph structure Multi-Flow Multi-Relational Graph (MMRG), based on three types of relationships. Then, it introduces Relational Graph Convolutional Networks to update vertex features in MMRG and generates global graph-level representations for multi-flow classification. We conduct experiments on raw network traffic. The results show that MFSI can achieve a classification accuracy of 98.58% without filtering or deleting any traffic, surpassing state-of-the-art schemes. It also performs well in identifying the type of services of unknown encrypted traffic.