MOBDroid: An Intelligent Malware Detection System for Improved Data Security in Mobile Cloud Computing Environments

Abstract

We propose an intelligent malware detection system (MOBDroid) that aims to protect the end-user’s mobile device (MD) in mobile cloud computing (MCC) environment. MOBDroid utilizes the Android Operating System (OS) permission-based security system. The APK files of 28,306 benign and malicious applications (apps) collected from the AndroZoo and RmvDroid malware repositories were used in the system development process. The apps were decompiled in order to extract their manifest files and construct a dataset comprising the permissions requested by each of the apps. We identified some unique permissions that could be used to distinguish between malicious and benign apps and performed a series of experiments using a machine learning (ML) model; the model drew on the ML.net library and was implemented in C#.net. In the experiments conducted, we obtained classification accuracy of 96.89%, a detection rate of 98.65%, and false negative rate of 1.35%. The results indicate that our model compares very favorably to other models reported in the extant literature.