Online Encrypted Mobile Application Traffic Classification at the Early Stage: Challenges, Evaluation Criteria, Comparison Methods

Junzheng Shi, Mengyan Liu, Chengshang Hou, Minghao Jiang, Gang Xiong

Abstract: Encrypted mobile application traffic classification at the early stage refers to the identification of the application to which the mobile network traffic belongs in a limited time or within limited network packets and feeding back the results to the policy routing control or QoS module. Early traffic classification is the basis for mobile ISPs to implement tasks such as QoS, network billing, and network security management. Current studies generally improve the performance of classification in encrypted traffic by exploring more complex model or constructing more representative features of off-line traffic. Few studies are focusing on the classification of mobile application encrypted traffic in the early stage. What's more, encrypted traffic classification still faces the constraints of limited computing and memory resources when applied to the high-speed network. Therefore, in addition to meeting real-time requirements, early encrypted traffic identification methods must also make the constraints of limited resources into consideration. In this paper, we analyze the challenges of early classification of encrypted mobile application traffic in online settings and proposes evaluation criteria for early traffic classification based on different scenarios. Several typical encrypted traffic classification methods in recent years were selected to make comparisons over our mobile application traffic datasets under the standards of this paper.