Monterey, CA; Naval Postgraduate School
FEDERATED LEARNING IN CYBERSECURITY: PERFORMANCE, ROBUSTNESS, AND ADVERSARIAL THREATS
Abstract
dc:description.abstractModern military operations require resilient, privacy-preserving data processing across distributed systems, where centralized learning often fails to meet security and operational demands. Federated learning (FL) provides a decentralized solution by enabling model training without transferring raw data. This thesis evaluates FL's performance, robustness, and vulnerability to adversarial attacks in network traffic classification at the tactical edge. Using supervised models—random forest, Extreme Gradient Boosting, and convolutional neural networks (CNNs)—this study compares FL to centralized learning using Internet of Things (IoT)-23 and Naval Postgraduate School (NPS) Enterprise Research Network (ERN) datasets. Results show FL matches centralized performance in binary classification but struggles with multiclass tasks, particularly under class imbalance. Label flipping attacks preserve overall accuracy while degrading minority-class performance. A novel backdoor method is introduced, embedding feature-level triggers without changing labels, reducing macro and weighted F1 scores by up to 10%. This work contributes a federated simulation framework, model comparison, and threat evaluation. Findings reveal the need for improved metrics, class balancing, and integrated defense mechanisms to enhance FL resilience. These insights inform future deployment of FL in military and coalition domains where data sensitivity, resource constraints, and operational security remain paramount.
Degree
thesis:*- Name thesis:degree_name
- Master of Science in Computer Science
- Level thesis:degree_level
- Masters
- Discipline thesis:degree_discipline
- Computer Science
- Department dc:contributor.department
- Computer Science (CS)
- Grantor dc:publisher
- Monterey, CA; Naval Postgraduate School
- Year dc:date.issued
- 2025
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Nkuako, Kojo A.
- Advisor dc:contributor.advisor
-
- Barton, Armon C.
- Committee member dc:contributor.committeemember
-
- Orescanin, Marko
Rights
dc:rights- Statement dc:rights
-
- This publication is a work of the U.S. Government as defined in Title 17, United States Code, Section 101. Copyright protection is not available for this work in the United States.
- Distribution Statement A. Approved for public release: Distribution is unlimited.
Identifiers
dc:identifier.*- Dc Identifier Programcode
- 326, Cyber Systems and Operations
- Thesis Thesis Identifier
- 40921