Analysis and Detection of Android App Privilege Escalation Vulnerability Based on Machine Learning

Abstract

Android is currently playing an important role in smartphone operating systems. However, there are potential risks hidden in Android applications. One of them is application vulnerability, which can put users’ information in danger. In this paper, we investigate the machine learning techniques in identify one of the most common application vulnerabilities—privilege escalation vulnerability (referred as PE vulnerability). We propose a machine learning system for detecting PE vulnerability. Our method is component-based detection, which means a more granular method. In this way, we can not only identify App with problems, but also locate component with loopholes. We first analyze the principle of PE vulnerability’s formation process, and then propose an EMPC model according to key elements in the process. The model is used to select features in the following steps. Second, we apply machine learning techniques to build a classification model to classify samples. Seven classification algorithms are applied. The experiment results show that our method is feasible in detecting PE vulnerability in component-degree.