Detection of Attacks to ADS-B Protocol through Graph Neural Networks and Anomaly Detection Techniques using Crowdsourced Sensors Network

The Automatic Dependent Surveillance-Broadcast (ADS-B) protocol is a cornerstone of modern air traffic management, providing real-time aircraft position data to improve safety and efficiency. However, its lack of robust security mechanisms makes it vulnerable to various cyberattacks, potentially compromising aviation safety. For this reason, this thesis introduces a novel detection system that leverages Graph Neural Networks (GNNs) to analyze data from crowdsourced network of ADS-B receivers. The methodology transforms ADS-B messages into graph-based representations, enabling comprehensive analysis of spatial and temporal relationships to identify threats. Unlike existing solutions that address individual or only a subset of security challenges, our implementation provides a framework that simultaneously tackles system scalability, adaptability to dynamic network configurations, and protection against insider threats. Experimental validation demonstrates that our GNN-based approach effectively detects most attack types, particularly those that modify sensor reception patterns. While the system shows limitations in areas with sparse sensor coverage and in detecting sophisticated spoofing attacks, it establishes a promising foundation for enhanced ADS-B security. The results suggest that GNN-based architectures offer distinct advantages over traditional machine learning approaches.