Design and evaluation of distributed machine learning approaches for cyber attack detection on resource-constrained devices

Ieropoulos, Vasilis ORCID: https://orcid.org/0009-0002-0196-6571 2025. Design and evaluation of distributed machine learning approaches for cyber attack detection on resource-constrained devices. PhD Thesis, Cardiff University. Item availability restricted.

Preview	PDF - Accepted Post-Print Version Available under License Creative Commons Attribution No Derivatives. Download (8MB) | Preview
	PDF (Cardiff University Electronic Publication Form) - Supplemental Material Restricted to Repository staff only Download (224kB) | Request a copy

Abstract

The rapid growth of Internet of Things (IoT) devices presents a significant security challenge, as their resource constraints make them susceptible to cyberattacks. This thesis introduces a machine-learning-based intrusion detection system (IDS) specifically designed and evaluated for deployment on these resource-constrained devices. Unlike much prior work that often relied on more powerful hardware, this research demonstrates the practical feasibility of deploying a complete IDS on a resource-constrained microcontroller. A central contribution is a novel decentralised collaborative inference architecture that allows devices to pool intermediate detection results to achieve a more robust and accurate conclusion. This approach enhances detection accuracy and improves resilience against both individual and coordinated adversarial attacks. The study validates that key adversarial attacks, including model poisoning, can be countered through lightweight integrity verification and selective aggregation mechanisms. The implemented models achieved high F1-scores of up to 99% in specific scenarios, demonstrating that accurate detection is achievable at the device level. Furthermore, the thesis provides a systematic analysis of the real-world computational and energy costs associated with the system’s operation, addressing a gap in existing literature. Findings show that the proposed framework achieves high detection accuracy whilst operating within the strict computational and energy budgets of IoT devices. This work acknowledges a trade-off between security and resource constraints, as the system cannot implement complex encryption and hashing algorithms due to its limited resources. Additionally, the connectivity speed and bandwidth of these devices limit the amount of data that can be shared in a given timeframe. This research provides a foundation for developing practical, scalable, and adaptable security solutions for large-scale IoT networks

Item Type:	Thesis (PhD)
Date Type:	Completion
Status:	Unpublished
Schools:	Schools > Computer Science & Informatics
Subjects:	Q Science > QA Mathematics > QA76 Computer software
Funders:	10.13039/501100000726-Toshiba of Europe, 10.13039/100014013-UK Research and Innovation
Date of First Compliant Deposit:	17 April 2026
Date of Acceptance:	15 April 2026
Last Modified:	17 Apr 2026 16:36
URI:	https://orca.cardiff.ac.uk/id/eprint/186463

Actions (repository staff only)

Edit Item