Machine learning-driven SDN for reliable emergency response in industrial IoT networks

Abstract

Industrial Internet of Things (IIoT) applications operating in safety-critical environments—such as chemical plants, oil refineries, and disaster response systems—demand ultra-low latency, deterministic communication, and absolute message reliability, particularly during emergency scenarios. Traditional Software-Defined Networking (SDN) controllers often fall short under such dynamic conditions due to static routing mechanisms, fixed flow timeout policies, and lack of real-time congestion awareness. This paper proposes an enhanced SDN architecture integrating a lightweight XGBoost-based machine learning model into the POX I2_learning controller to proactively predict and mitigate congestion in IIoT environments. The proposed system dynamically adjusts flow routing, optimizes timeouts, and enables congestion-aware path selection while maintaining compatibility with industry protocols like OPC UA and GOOSE. Designed for edge deployment, the controller achieves sub-5 ms latency, 0% packet loss, and up to 35% fewer control messages across simulated emergency scenarios involving bursty and multi-zone traffic. Extensive evaluations using a real-time Mininet-based emulation framework reveal notable gains in latency stability, throughput, scalability, and processing efficiency. This work offers a practical and deployable solution for resilient, intelligent, and standards-compliant IIoT networking in mission-critical environments.