Graph Neural Network-Based Intrusion Detection System for a Swarm of UAVs

Unmanned aerial vehicle (UAV) swarms present significant potential in both civil and military applications, yet the security of swarm communications remains a critical challenge. While machine learning-based intrusion detection systems (IDS) have advanced, their effectiveness is often hindered by the reliance on simulated or irrelevant datasets that do not adequately capture the unique characteristics of UAV swarm communications. Furthermore, existing IDS have predominantly focused on temporal information, overlooking the potential of spatial relationships within the UAV network. To address these limitations, this research establishes a testbed of six UAVs forming an hexagonal graph where each UAV acts as a node and communicates with its immediate neighbors. We then execute various cyber-attacks such as false data injection, evil twin, replay, and denial-of-service attacks on each of the UAVs in the swarm. This allows us to collect spatial and temporal data under normal operations and attack conditions. We propose a graph neural network (GNN)-based IDS that exploits spatial and temporal information patterns. This research seeks to answer the following question: Can leveraging the spatial relationships within a UAV swarm improve detection performance compared to IDS relying solely on temporal information? Through extensive experiments and comparison with traditional deep neural network models, we evaluate the effectiveness of this topology-aware GNN approach in securing UAV swarm communications.