Mobile malware is growing and affecting more and more mobile users around the world. Malicious developers and organisations are disguising their malware payloads on apparently benign applications and pushing them to large app stores, such as Google Play Store, and from there to final users. App stores are currently losing the battle against malicious applications proliferation and existing malware. Detection methods based on signatures, such as those of an antivirus, are limited, new approaches based on machine learning start to be explored to surpass the limitations of traditional mobile malware detection methods, analysing not only static characteristics of the app but also its behaviour. This paper contains an overview of the existing machine learning mobile malware detection approaches based on static, dynamic and hybrid analysis, presenting the advantages and limitations of each, and a comparison between the reviewed methods.

This work is part of the AppSentinel project, co-funded by Lisboa2020/Portugal2020/EU in the context of the Portuguese Sistema de Incentivos à I&DT - Projetos em Copromoção (project 33953).

Android,Machine learning,Malware,Mobile,Security

• Physical Sciences - Natural Sciences