Optimized lightweight ensemble deep learning for Intrusion Detection System in IoMT environment

In smart cities, the Internet of Medical Things (IoMT) systems can interact with various urban services, including healthcare, emergency response, environmental monitoring, and transportation networks. This integration provides comprehensive, context-aware health services, which align with the goal of creating secure, responsive, and citizen-centric health management. Ultimately, this approach enhances the resilience and sustainability of modern urban living. However, IoMT communications depend on wireless technology, which has various vulnerabilities that could endanger patient safety. Intelligent Intrusion Detection Systems (IDS), particularly those utilizing machine or deep learning at the edge as gateways, offer significant advantages over traditional systems. The emerging challenges include addressing computational limitations, accommodating the diversity of edge devices, and minimizing false detection rates. Our methods optimize individual models by incorporating focal loss, achieving lightweight models that are under 1 MB in size, while ensemble learning models remain under 2.5 MB. The final results from ensemble learning can reduce false detection rates by up to 28% compared to individual models.