YarowskyDroid: Semi-supervised based Android malware detection using federation learning

Abstract

In this paper, a novel approach is proposed entitled as “YarowskyDroid”, that works on the principle of semisupervised machine learning approach and federation learning to detect malware-infected apps. In order to protect user privacy, apps are exclusively installed locally on the user’s smartphone. This prevents service providers or developers from learning which apps a user has downloaded. Meanwhile, information from smartphone users is gathered in order to improve the malware detection algorithm. The primary issue in this study is that users cannot tell whether an app they have loaded is contaminated with malware or not. In order to overcome this problem, a semi-supervised machine learning technique is proposed in this study that improves classification accuracy on comparison to the base model set up in the cloud. A experiment was carried out using 50,000 malware-free and 25,000 malicious app downloads from different repositories. The empirical finding shows that the suggested framework, with 210 users and 40 rounds of the federation, has a detection rate of 97.9%.