An Automated Classifier Generation System for Application-Level Mobile Traffic Identification

Abstract

As mobile devices and smartphones have become increasingly popular in recent years, there are now a growing number of mobile applications that require access to the Internet. Network administrators need to be able to identify application-level mobile traffic in order to deal with this mobile ‘big-bang’, but doing so is a challenge because of the rapid increase in the number of mobile applications and the volume of mobile application traffic. To deal with diverse mobile applications with minimal effort, we need a highly accurate automated method to identify mobile applications.This thesis investigates the architecture of an automated system to generate classifiers. These classifiers represent key features of each mobile application’s traffic. This architecture includes mobile traffic measurement agents (mTMAs) which are installed on mobile devices and monitor their application-level traffic. The proposed mTMAs are designed to work well even on mobile devices that have low computation processing power and minimal memory. The proposed flow matching shows how the information collected by mTMAs can be converted into effective input for a classifier generation system.Finally, this thesis proposes an algorithm to automatically find each mobile application’s classifiers. Using this proposed algorithm, we extract several mobile traffic classifiers from our campus network traffic. The experiment results show the feasibility of our proposed method with acceptable accuracy.