Creating Usage Models to Identify Misbehaving Applications on Mobile Devices

Abstract

Limited battery capacity is currently a major pain point for mobile users. The problem is made worse when poorly designed applications consume a significant amount of power in the background when they are not actively used by the user. To combat this problem, we propose an automated monitoring system that can detect misbehaving applications running on mobile devices. Our system does not require any prior knowledge about the monitored applications. Instead, it collects the user’s usage records and builds models to encapsulate the contexts when the user is likely to use each application. From those models, our system can identify misbehaving applications that are consuming system resources while providing no useful service to the end user. In this dissertation, we demonstrate the overall design for our system. This design allows us to collect detailed usage records while keeping our system’s power consumption at a minimum. We also introduce the steps we take to construct our usage models and the rationale behind each key decisions. In the end, we evaluate the effectiveness of our system by running it on a real Android device during a two month period. From the experiment, we show the misbehaving applications identified by our system have a significant impact on the battery life, and misbehaving applications with high network usage is the main cause of fast battery drain.