Abstract
The proliferation of IoT devices presents significant security challenges because traditional intrusion detection systems (IDS) are insufficient in handling resource constraints, latency sensitivity, and dynamic network conditions. To address these issues, this paper proposes an adaptive edge–cloud collaborative IDS framework that dynamically divides threat detection tasks between edge and cloud resources. The framework features key innovations, including a context-aware decision engine that adjusts offloading thresholds based on real-time metrics like threat confidence, CPU and memory usage, and network latency. It incorporates a lightweight edge-based Random Forest classifier for quick inference and a cloud-based deep neural network for comprehensive analysis. This design ensures efficient resource use while maintaining high detection accuracy and low latency. Experimental evaluations using the IDSIoT2024 dataset demonstrate that the hybrid model achieves 99.48% accuracy with a latency of 0.034ms, outperforming both edge-only and cloud-only solutions. The system maintains 98.6% accuracy under various network conditions and adapts to adversarial noise by intelligently increasing offloading rates from 42.8% to 83.0%, thereby enhancing robustness.