EdgeGuardIA: an adaptive in-network machine learning framework for real-time IoT gateway security
摘要
The proliferation of Internet of Things (IoT) devices across latency-sensitive applications has exposed critical vulnerabilities in traditional cloud-centric security architectures. This paper proposes EdgeGuardIA, an intelligent, in-network threat detection and mitigation framework operating within programmable IoT gateways. By embedding lightweight machine learning inference into the data plane, EdgeGuardIA enables real-time packet analysis and rapid threat response. The system features dynamic model retraining and seamless updates via shadow table reconfiguration. Experimental evaluations show that EdgeGuardIA achieves a detection accuracy of up to 97.2%, processes up to 8,50,000 packets per second at a traffic rate of 1 Gbps, and maintains an inference latency below 65 microseconds. These results demonstrate EdgeGuardIA's ability to provide high-performance, scalable, and adaptive security for diverse IoT scenarios.