An Unsupervised Graph Neural Network Approach to Deceive UAV Network Reconnaissance Attack
Adrien Njanko, Danda B. Rawat, Charles Kamhoua, Ahmed H. Anwar Hemida · 2024
Unmanned Aerial Vehicles (UAVs) networks are subjected to cyber attacks similarly to any networked systems. Network reconnaissance attacks (NRA) are important first steps in the cyber kill chain. To counter these attacks without dis-rupting the system is difficult due to the automatic response to requests. The Cyber Deception field provides a framework to lure attackers despite the automatic response to requests. Our approach to mitigate NRA consists in responding to requests with fake UAV s network configurations that do not match the defender's configurations. The fake configurations come from other networks that are different from the defender's UAV s network. Within a pool of networks, we find the best fake network candidate using a Graph Neural Network (GNN) model. The GNN model compares each network from the pool against the defender's network. The best candidate is close to the defender's network while having a complete different set of features. The end result is a GNN model that finds the best fake network given a reference one and, a system response to NRA request with deceptive information.