A Hybrid Anomaly Characterisation and Sensitivity Analysis Framework for Software Defined Network-Aided Communication

Simple item page
dc.contributor.authorChekwube EZECHI
dc.date.accessioned2026-11-05T14:38:43Z
dc.date.available2026-11-05T14:38:43Z
dc.date.issued2025-12
dc.description.abstractContemporary communication infrastructures increasingly rely on Software-Defined Networking (SDN) to deliver programmability, scalability, and dynamic resource management. These strengths also introduce vulnerabilities, exposing SDN controllers and data planes to anomalies such as Distributed Denial of Service (DDoS) flooding attacks. These attacks threaten availability, degrade performance, and undermine network resilience. This research develops a hybrid anomaly characterisation and sensitivity analysis framework tailored for SDN-enabled Internet of Things (IoT) environments. The DOE-GAN-SA framework integrates design of experiments (DOE), generative adversarial networks (GANs), anomaly detection and characterisation (ADC), and sensitivity analysis (SA), providing a unified methodology for analysing and mitigating anomalous behaviour. The study simulated normal and malicious traffic within an emulated SDN-IoT testbed built with Mininet, Oracle VirtualBox, and Ubuntu Linux. Popular DDoS vectors, including HTTP, TCP, and UDP flooding attacks, were launched with the Low Orbit Ion Cannon (LOIC) tool against custom topologies containing OpenFlow-enabled switches and hosts. Performance indicators such as throughput, jitter, and response time were collected. Latin Hypercube Sampling (LHS) and GANs were used to augment datasets and ensure statistical robustness. CART supported anomaly detection and classification, while Artificial Neural Networks (ANNs) performed global sensitivity analysis. Repeated experiments and hypothesis tests ensured statistical credibility. Results demonstrated that the DOE-GAN-SA framework effectively captured service degradations across both normal and DDoS conditions. The hybrid model achieved 99.93 % accuracy and an F1 score of 1.000 for anomaly detection. Sensitivity analysis produced a mean squared error of 9.14 × 10⁻⁴, confirming that the ANN reliably learned the behaviour of throughput, jitter, and response time under multiple scenarios. Compared with existing methods, the framework delivered higher detection accuracy, improved dataset quality through synthetic augmentation, and reduced computational overhead, supporting its suitability for near real-time SDN-IoT deployment. Key contributions include a unified methodology for anomaly characterisation and sensitivity analysis in SDN, empirical validation of synthetic data augmentation for security studies, and identification of parameters that most influence network resilience. The study offers network administrators and security practitioners proactive tools for anomaly detection and performance optimisation, strengthening the development of resilient and secure SDN-IoT infrastructures. Future work will expand the framework to additional attack types and emerging 5G contexts. Keywords: Software-Defined Networking (SDN), Internet of Things (IoT), Distributed Denial of Service (DDoS), Generative Adversarial Networks (GANs), Design of Experiments (DOE), Latin Hypercube Sampling (LHS), Artificial Neural Networks (ANNs) Word Count: 355
dc.identifier.citationKate Turabian
dc.identifier.otherPh.D
dc.identifier.urihttps://repository.lcu.edu.ng/handle/123456789/1702
dc.language.isoen
dc.publisherLead City University, Ibadan
dc.relation.ispartofseriesPh.D
dc.subjectSoftware-Defined Networking (SDN)
dc.subjectInternet of Things (IoT)
dc.subjectDistributed Denial of Service (DDoS)
dc.subjectGenerative Adversarial Networks (GANs)
dc.subjectDesign of Experiments (DOE)
dc.subjectLatin Hypercube Sampling (LHS)
dc.subjectArtificial Neural Networks (ANNs)
dc.titleA Hybrid Anomaly Characterisation and Sensitivity Analysis Framework for Software Defined Network-Aided Communication
dc.typeThesis

Files

Original bundle
Now showing 1 - 1 of 1
No Thumbnail Available
Name:
EZECHI Chekwube Obinna PHD THESIS.pdf
Size:
7.09 MB
Format:
Adobe Portable Document Format
Download
License bundle
Now showing 1 - 1 of 1
No Thumbnail Available
Name:
license.txt
Size:
1.71 KB
Format:
Item-specific license agreed to upon submission
Description:
Download

Collections

Department of Computer Science;