ONYIBO, CHUKWUEBUKA GIDEON (2026) HYBRID AI-POWERED INTRUSION DETECTION SYSTEM FOR WIRELESS NETWORKS. Other thesis, Godfrey Okoye University, Enugu.
|
Text
Onyibo Chukwuebuka Final Year Project.pdf Download (1MB) | Preview |
Abstract
Conventional signature-based intrusion detection systems (IDSs) are not very effective at detecting subtle network-level attacks such as deauthentication, disassociation, and rogue access point impersonation in IEEE 802.11-based wireless networks. This research has devised and developed a Hybrid AI-Powered Intrusion Detection System (IDS) for wireless networks, consisting of two models: a supervised Random Forest model for known attack classification and an unsupervised Isolation Forest model for zero-day anomaly detection. It has been combined with a rule-based expert system and a SHAP (SHapley Additive exPlanations) explainability engine to provide structured explanations in plain English of the nature of each detected intrusion, and is presented via a Flask-based near-real-time Web Dashboard. The system was trained on the AWID3 wireless benchmark dataset with a 33/41/40 classification breakdown (Deauth, Disas, RogueAP) using a file-level train-test split (205730/313193 files), with class distribution matching. On the unseen data for testing, the Random Forest classifier had an accuracy of 99.44%, precision of 99.55%, recall of 99.44%, and F1 score of 99.47%. It successfully recalled 1.00% of instances from each of the three attack categories.. To test the generalization of the wireless network datasets beyond the controlled laboratory environment, five synthetic wireless network datasets were created from the statistical distribution of the AWID3 training datasets, and the accuracy was found to be 97.87% on average. The results indicate that the proposed hybrid architecture is capable of solving the three identified research challenges that wireless IDS research lacks: a lack of wireless-specific hybrid IDS implementation, a lack of dual supervised and unsupervised detection architectures in IEEE 802.11 environments, and the universal absence of explainability mechanisms in wireless IDS research.
| Item Type: | Thesis (Other) |
|---|---|
| Subjects: | Q Science > Q Science (General) |
| Divisions: | Faculty of Engineering, Science and Mathematics > School of Electronics and Computer Science |
| Depositing User: | MICHAEL MADUBUKO |
| Date Deposited: | 22 Jul 2026 11:59 |
| Last Modified: | 22 Jul 2026 11:59 |
| URI: | http://eprints.gouni.edu.ng/id/eprint/6007 |
Actions (login required)
 |
View Item |
