Energy-Aware TinyML for Intrusion Detection in IoT Networks

The rapid expansion of the Internet of Things (IoT), accelerated by the capabilities of emerging 5G/6G networks, introduces new challenges in terms of security and privacy. The heterogeneity of devices and the prevalence of cyberthreats demand automated intrusion detection systems (IDS) that can operate efficiently under severe resource constraints. This paper presents an IDS based on Tiny Machine Learning (TinyML) that supports both on-board training and inference directly on MicroController Unit (MCU). The proposed approach leverages the AIfES framework to enable lightweight yet effective neural network models capable of monitoring network traffic without relying on external servers, thereby enhancing privacy and reducing communication overhead. Experimental evaluation on a representative intrusion detection dataset using constrained IoT hardware demonstrates that the proposed IDS achieves high detection accuracy while maintaining a minimal memory and energy footprint. These results confirm the feasibility of TinyML-based IDS as an effective and sustainable security solution for distributed IoT in the 5G/6G era.