Mining Identification Rules for Classifying Mobile Application Traffic

Classifying mobile application traffics is important in many network management tasks. Existing works rely on human expertise and reverse engineering to build classification rules. The huge number of mobile applications make it ineffective and even infeasible to do reverse engineering on every mobile application. In this thesis, we design a novel structure of app identification rules. Two algorithms are developed to mine the rules from HTTP header fields without any other external input. In addition, we also explore the function and effects of different HTTP header fields in the identification task. An extensive empirical study on real data verifies the effectiveness of our algorithms.