Supervised Learning Based Detection of Malware on Android

Abstract This chapter aims to present a new approach for detecting Android malware by relying permissions and supervised learning techniques. For that, we present security and its flaws in the Android system. Then we present concepts around machine learning and how they can be used for malware detection in general. We discuss works using permissions as key feature for the characterization of applications to detect malicious behavior. We present a detection system combining the proportion of requested permissions and risks induced on resources. This system requires the user to specify resources to protect and inform in an understandable way, activities performed in background with those permissions. We pass through some graphical interfaces of the implementation, then elucidate results concerning detection and prediction performance with the support of learning algorithms. We compare these results against well-known antiviruses and related solutions on the same collected datasets of malicious and benign applications. It is revealed that our system outperforms most of them and it is able to detect zero-day malware. Therefore it constitutes an interesting step forward to help users understanding the risks induced on resources and to help them detecting malware.