PET-Droid: Android Malware Detection Using Static Analysis

Abstract

Android is the most popular mobile operating environment with the highest share in the market. This makes Android vulnerable to attacks from cybercriminals. Cybercriminals develop malware to attack Android applications. Malware detection models rely on anti-virus vendors to acquire signatures of malware. These signatures are used to train models in a supervised machine learning paradigm. However, a significant number of data is mislabeled, which affects the detection of malware as the model is trained with inaccurate data. To address this issue, a malware detection model, PET-Droid is developed in this literature which uses unsupervised machine learning to find commonalities in the features possessed by malware and goodware samples. PET-Droid detects Android malware with an accuracy of 96.8481%.