Blockchain-Augmented FL IDS for Non-IID Edge-IoT Data Using Trimmed Mean Aggregation
Paper Title: Blockchain-Augmented FL IDS for Non-IID Edge-IoT Data Using Trimmed Mean Aggregation
Journal Name: IEEE Internet of Things Journal
Abstract: Abstract—The rapid expansion of the Internet of Things (IoT)
has introduced significant security challenges, particularly in
edge networks, where resource-constrained devices are deployed.
This paper presents a robust intrusion detection system (IDS) for
edge IoT networks, developed as a combination of blockchain
technology and Trimmed Mean Aggregation (TMA) within a
Federated Learning (FL) framework. Blockchain manages client
registration and stores the final global model, ensuring that
only authenticated clients participate in the training process and
preserving the model’s integrity. TMA is employed to mitigate the
impact of Byzantine attacks, discarding extreme model updates
that may originate from adversarial clients. The experimental
setup simulated non-independent and identically distributed
(non-IID) data among clients to evaluate the robustness of the
proposed system, achieving a validation accuracy of over 96%
and minimal error loss of 0.04 in all tested scenarios. The results
demonstrate the system’s effectiveness in maintaining a high
model accuracy and robustness under adversarial conditions,
making it suitable for a secure and efficient IDS in edge IoT
environments.