Enhancing Mobile Threat Detection through TinyML and Auditable ML

Abstract

The proliferation of mobile devices has significantly increased the risk of exposure to advanced cyberattacks, and classical detection methods increasingly became difficult to apply. Cloud-Based solutions, while being highly effective, introduce high latency, privacy concerns, and resource expensiveness. Existing studies focus on either TinyML or explainability; Our framework is the first to combine the two for mobile security. The current work introduces a novel paradigm that intertwines Auditable Machine Learning (Auditable ML) with Tiny Machine Learning (TinyML) to overcome the above challenges for general purpose detection. TinyML supports real-time, low-energy on-device threat detection with low cloud infrastructure dependency and greater privacy from local data processing. In parallel, Auditable ML facilitates transparency and accountability in the system by combining explainable AI and secure logging features so detection operations are explainable and auditable. The efficacy of the proposed solution is validated through large-scale simulations, which show substantial detection accuracy improvements, resource utilization, and cross-platform portability improvements. With the combination of TinyML and Auditable ML, this framework provides a scalable, privacy-preserving mobile security solution. It gives a good basis on which to build future innovation in the mobile ecosystem, addressing both nascent threats and the requirement for robust, user-centric security.