Electronic Thesis/Dissertation
 

Unsupervised Machine Learning for Intrusion Detection of Cyber-Attacks in Unmanned Aerial Vehicles

Open Access Deposited

This paper presents a reliable and extensible ensemble framework for unsupervised anomaly detection in Unmanned Aerial Vehicles (UAVs). This framework aims at detecting cyber-attacks without relying on labeled data for training. Using the IEEE UAV Attack Dataset, the system integrates diverse models and techniques like Autoencoders, Isolation Forests, One-Class SVMs, Principal Component Analysis and pretrained Transformer-based detectors over multi-sensor, time-aligned telemetry. In the ensemble, model contributions are adaptively weighted via a hybrid scheme combining supervised trust (F1-score) with unsupervised reliability metrics (kurtosis, entropy, Median Absolute Deviation). The resulting weights are fused using a data-driven coefficient α, derived through Spearman rank correlation between supervised and unsupervised scores. This research uses MOMENT a pretrained Transformer introduced by Goswami et al (2024) for gating consensus mechanism. MOMENT refines final decisions using temporal smoothing and inter-model agreement measured by Jaccard similarity. Unlike prior single-detector methods, the proposed architecture is modular, sensor-agnostic, and resilient to sensor dropout. Experiments show improved anomaly detection accuracy, reduced false positives, and enhanced generalization across spoofing, jamming, and evolving threats without requiring threshold tuning.

Author Language Keyword Date created Type of Work License
  • All rights reserved
Rights statement GW Unit Degree Advisor Committee Member(s) Persistent URL

Notice to Authors

If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.

Thumbnail Title Date Uploaded Visibility Actions
Preview of Ahmed_gwu_0075A_17407.pdf Ahmed_gwu_0075A_17407.pdf 2025-12-11 Open Access