Understanding the black box: Android malware detection through Explainable AI

Abstract Detecting Android malware is crucial in safeguarding mobile devices and user data. We introduce a novel approach to android malware detection through convolutional neural networks (CNNs) and Explainable AI. First, we present a methodology for processing android applications, transforming them into images, and through a series of experiments, we demonstrate the efficacy of CNNs in identifying malware within these images. Furthermore, we employ Explainable AI techniques to analyse the decision-making processes of these models. Our approach goes beyond detection; we rigorously analyse the key aspects that distinguish malware, allowing us to improve and validate our data transformation methodology. This emphasis on understanding the decision-making processes of the models is the main aspect of our approach, as it provides insight into the mechanisms of malware, enhancing our understanding of malware detection and reinforcing the robustness of our conclusions. In essence, our work not only provides a comprehensive framework for Android malware detection but also underscores the significance of Explainable AI in cybersecurity research, integrating advanced computational techniques with a detailed understanding of malware behaviour.