A MEC-based architecture to secure IoT applications using federated deep learning

Internet of Things (IoT) is a promising paradigm that is considered as major enabler of smart cities. However, with the emergence of IoT botnets, the number of unsecured IoT devices is increasing rapidly. This can give attackers more advanced tools to carry out large scale damaging IoT attacks. Advanced Machine Learning (ML) techniques can help enhance the effectiveness of conventional intrusion detection systems (IDS) to accurately detect IoT attacks. But there are ongoing challenges with centralized learning as well as the lack of up-to-date/ new datasets, covering key IoT attacks. In this context, we design a novel Multiple access Edge Computing (MEC) architecture to secure IoT applications with Federated Learning (FL). In particular, we propose a promising eDge-based architEcTure to sEcure IoT appliCations using FL, called DETECT. DETECT allows multiple MEC domains to collaboratively and securely mitigate IoT attacks, while ensuring the privacy of the MEC collaborator and consequently the privacy of IoT devices. The in-depth experiments results with well- known IoT attack using, the Edge-IIoTset and NSL-KDD datastets, show the significant accuracy of DETECT in terms of Accuracy (86 percent in NSL-KDD and 99 percent in Edge-IIoTset) and F1 score (87 percent in NSL-KDD and 99 percent in Edge-IIoTset).