Citation

Umair, Muhammad and Tan, Wooi Haw and Foo, Yee Loo (2026) Hierarchical Federated Learning Approach for IoT Attacks Classification. IEEE Access. p. 1. ISSN 2169-3536

Text 14.pdf - Published Version Restricted to Repository staff only Download (6MB)

Abstract

The worldwide spread of internet access has driven a surge in devices connected to the Internet of Things (IoT), but their connectivity also opens a challenge such as network based cyber attacks. Existing solutions utilizes machine learning & deep learning techniques to identify these attacks, however, because these techniques rely on a centralized approach where the dataset needs to be processed and trained in a single node makes them non-viable solution because of data privacy and resource limitation. Thus, a decentralized solution known as Federated Learning (FL) exists and has been a promising solution for this challenge. However, FL face challenges in training and aggregating models because poorly performing trained model often negatively impact the overall aggregated model. Improving FL performance is the main target of this study, whereas countering IoT attacks is the use case. To address this issue in FL, the Hierarchical Federated Learning (HFL) technique has been utilized in this study, which train models on local data that are distributed among clients using a decentralized approach. The proposed methodology involves training a one-dimensional convolutional neural network across 11 local clients on the N-BaIoT and CSE-CIC IDS 2018 dataset using HFL. HFL aggregates client models based on a threshold validation accuracy of 80% accordingly. The experimental results show that the global model achieved test accuracies of 81.05% in the CSE-CIC IDS 2018 dataset and 98.94% in the N-BaIoT datasets. In comparison, traditional FL methods such as FedAvg, FedSGD, and FedProx achieved 97.54%, 90.26%, and 98.65% accuracies on N-BaIoT dataset and 75.56%, 79.02%, and 77.93% accuracies on CSE-CIC IDS 2018 dataset accordingly. The superior performance of the proposed HFL approach demonstrates its effectiveness for decentralized training for IoT attack classification.

Item Type:	Article
Uncontrolled Keywords:	IoT attacks, deep learning
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5101-6720 Telecommunication. Including telegraphy, telephone, radio, radar, television
Divisions:	Faculty of Artificial Intelligence & Engineering (FAIE)
Depositing User:	Ms Rosnani Abd Wahab
Date Deposited:	04 May 2026 06:13
Last Modified:	04 May 2026 06:13
URII:	http://shdl.mmu.edu.my/id/eprint/15885

Downloads

Edit (login required)