An approach for privacy-preserving mobile malware detection through federated machine learning

Considering the diffusion of smart devices and IoT devices, mobile malware detection represents a task of fundamental importance, considering the inefficacy of signature-based antimalware free and commercial software, which can detect a threat only if its signature is present in the antimalware repository. In the last few years, many methods have been proposed by academia to identify so-called zero-day malware through machine learning: these techniques typically extract a series of features from the mobile device to send to a server where the detection model is located. Typically, these features include network traces or installed applications, among other information that may compromise user privacy. In this context, Federated learning is emerging with privacy advantages because the raw data never leaves the local device. In this paper, we propose a method to integrate federated machine learning in malware detection.Malicious software typically aims to extract sensitive and private data, and mobile devices emerge as particularly enticing targets from the perspective of attackers. In the experimental analysis, comprising a pool of 10 clients from which 7 are uniformly sampled at each round, we demonstrate the efficacy of the proposed method by achieving an accuracy of 0.940.