MACHINE-LEARNING-BASED SELECTION OF ENCRYPTION ALGORITHMS FOR UAV

Abstract

Unmanned Aerial Vehicles (UAVs) generate heterogeneous data streams, including real-time video, telemetry, and
command-and-control messages, while operating under strict constraints on communication bandwidth, energy consumption,
computational resources, and mission risk levels. The use of a single universal encryption algorithm across all UAV scenarios is
inefficient, as it either results in excessive computational and energy overhead or provides an insufficient level of cryptographic protection in critical conditions. In this work, a set of eight criteria (K₁–K₈) for the informed selection of symmetric encryption algorithms is formalized. Based on these criteria, a machine learning (ML)-driven approach is proposed to enable the automated selection of the most appropriate cryptographic algorithm according to the parameters of a given UAV deployment scenario.
The key scientific contribution lies in the development of a UAV-oriented dataset that integrates mission context parameters
(operational conditions, resource constraints, threat levels) with cryptographic characteristics of algorithms (security strength, performance, and resource efficiency). The proposed feature structure provides a formalized representation of the algorithm selection problem and establishes a foundation for training intelligent decision-making models.
Each feature group is grounded in state-of-the-art research in UAV cybersecurity, lightweight cryptography, and ML-based adaptive
encryption. Furthermore, this work extends the authors’ previous studies related to the development of cryptographic algorithm datasets and the application of neural networks for ensuring image confidentiality in UAV systems.