Hierarchical Federated Learning for Collaborative IDS in IoT Applications

As the Internet-of-Things devices are being very widely adopted in all fields, such as smart houses, healthcare, and transportation, extremely huge amounts of data are being gathered, shared, and processed. This fact raises many challenges on how to make the best use of this amount of data to improve the IoT systems' security using artificial intelligence, with taking into consideration the resource limitations in IoT devices and issues regarding data privacy. Different techniques have been studied and developed throughout the years. For example, Federated Learning (FL), which is an emerging learning technique that is very well known for preserving and respecting the privacy of the collaborating clients' data during model training. Therefore, in this paper, the concepts of FL and Hierarchical Federated Learning (HFL) are evaluated and compared with respect of detection accuracy and speed of convergence, through simulating an Intrusion Detection System for Internet-of-Things applications. The imbalanced NSL-KDD dataset was used in this work. Despite its infrastructure overhead, HFL proved its superiority over FL in terms of training loss, testing accuracy, and speed of convergence in three study cases. HFL also showed its efficiency over FL in reducing the effect of the non-identically and independently (non-iid) distributed data on the collaborative learning process.