AppHunter: Mobile Application Traffic Classification

Abstract

Traffic classification finds its application in the implementation of various services like Quality of Service (QoS) and security monitoring. In today’s networks, a significant portion of traffic is generated from mobile applications. Thus, a robust and accurate mobile application traffic classification technique is needed. In this paper, we propose AppHunter, a mobile application classification technique to classify Android applications using Deep Packet Inspection (DPI). Unlike previously known mobile application classification techniques, AppHunter is an unsupervised approach and does not require training with flows explicitly collected for each application. AppHunter extracts required fields from HTTP/HTTPS header of a flow and compares them with application details extracted from Google Playstore. We test the classification performance of AppHunter with two publicly available datasets and one dataset generated by simulating more than thousand applications in our testbed setup and report the results. We also show an application of AppHunter by using its rules for network traffic filtering and shaping.