Android Malware Detection Using Convolutional Neural Networks

Abstract

The usage of mobile devices has been increased in recent years. Android is the most widely used operating system on mobile devices. As millions of applications with varied veracity are made available for download in the play store, it is essential to automatically identify malicious mobile applications. Most of the existing works rely on detecting Android malware based on either static (permissions, intents, application program interface calls, etc.) or dynamic features (API calls and system call traces, etc.) of the APIs. In this paper, Android malware and benign apks are converted into images based on their sizes, and on top of it, convolutional neural network architecture for deep learning is proposed for detecting Android malware. Malware apks are collected from virusshare.com, and benign apks are collected from play store, apkpure.com. Extensive experimentation conducted in two different settings has shown that only classes.dex files of apks are sufficient for Android malware detection. The proposed deep learning framework with convolutional neural networks could achieve 97.76% accuracy in detecting Android malware with minimal information requirement.