A Novel Federated Machine Learning-Based Approach for Explainable and Privacy-Preserving Android Malware Detection through Dynamic Analysis
Abstract
Over the years, the number of attacks has increased drastically due to the rise of mobile devices. Due to this enormous number of attacks, protecting users' personal information has become essential. In this paper, we propose an explainable method for Android malware detection by exploiting Federated Machine Learning. In detail, we propose a dynamic analysis that collects system call traces and converts them into images to train several models for dynamic Android malware detection. Moreover, we used non-independent and identically distributed data to better represent a real-world scenario. To provide explainability, we consider adopting Class Activation Mapping to highlight the area of the image under analysis that is symptomatic of a given prediction (i.e., malware or trusted), aiming to provide insights into the model's decision-making process.