A Low-Latency Fog-based Framework to secure IoT Applications using Collaborative Federated Learning

Attacks against the IoT network are increasing rapidly, leading to an exponential growth in the number of unsecured IoT devices. Existing security mechanisms are facing several issues due to the lack of real-time decisions, high energy consumption, and high time delays. In this context, we propose a novel Low-Latency Fog-based Framework, called FogFed, to secure IoT applications using Fog computing and Federated Learning (FL). The fog brings security mechanisms near IoT devices reducing delays in communication, while FL enables a privacy-aware collaborative learning between IoT while preserving their privacy. FogFed combines two levels of detection, Fog-based IoT attack detection using a binary FL classifier and cloud-based IoT attack detection using a Multiclass FL classifier. The in-depth experiments results with well-known IoT attack/malware using, the UNSW-NB15 datastet, show the significant accuracy (99%) and detection rate (99%), which outperforms centralized ML/DL models, while significantly reducing delays and preserving the privacy.