AI-driven resilience: advancing DDoS detection in virtualised IoT environments through an FCNN-based framework

Abstract

The rapid adoption of Internet of Things systems has accelerated the use of virtualisation technologies, including software-defined networking and cloud-based sensor platforms, to enhance scalability and manageability. However, these technologies introduce new security vulnerabilities, particularly distributed denial-of-service attacks, which pose significant risks to the stability and performance of virtualised IoT environments. This thesis addresses these challenges by proposing a novel AI-based DDoS detection framework specifically designed for virtualised IoT networks.A lightweight five-layer fully connected neural network comprising 49,729 parameters is developed and evaluated using eight benchmark datasets, including a custom virtualised IoT testbed with 168 hours of baseline traffic characterisation. The framework integrates batch normalisation, adaptive dropout regularisation, Adam optimisation, and SMOTE-based class balancing to enhance training stability and generalisation. The experimental results demonstrate a mean detection accuracy of 99.85% across all datasets, achieving a peak accuracy of 99.92% on the virtualised IoT testbed. The proposed model significantly outperforms established approaches, including LSTM, Random Forest, and Support Vector Machine classifiers, with all improvements statistically significant (p < 0.001) and showing large effect sizes.This research contributes to IoT security by quantifying virtualisation-specific vulnerabilities, demonstrating the effectiveness of lightweight AI-based detection in resource-constrained environments, and establishing experimentally validated scalability limits for deployment. The findings provide practical insights for designing efficient, adaptive DDoS detection systems for virtualised IoT infrastructures, while highlighting future directions in model adaptability, cross-domain evaluation, and privacy-preserving collaborative learning.

More information

Identifiers

ORCID for Belal Asad: orcid.org/0000-0001-6310-0170

ORCID for Nawfal Al Hashimy: orcid.org/0000-0002-1129-5217

Catalogue record

Export record