SaFe: A Stacked Ensemble Fingerprinting Approach for UAV Classification Fusing Mechanical and 5G Communication Data

The increasing deployment of Unmanned Aerial Vehicles (UAVs) in aerospace and defense operations necessitates robust identification mechanisms to ensure secure and resilient communications in contested environments. Traditional softwarebased authentication methods, such as cryptographic key exchanges and protocol-level security, remain vulnerable to spoofing attacks and adversarial manipulation. To address these challenges, we propose SaFe, a stacked ensemble fingerprinting approach that utilizes 5G communication performance metrics and mechanical vibration data for passive, tamper-resistant UAV classification. SaFe integrates UAV-specific signatures from both mechanical and communication domains to enhance classification accuracy. In the first stage, Extreme Gradient Boosting (XGBoost) models independently process mechanical vibration and 5G link performance data, capturing unique UAV characteristics. In the second stage, a Logistic Regression meta-network fuses these outputs to generate the final classification decision. Extensive real-world experimental evaluations demonstrate that SaFe achieves an AUC of 0.9901 and an accuracy of 94.75%, outperforming single-domain fingerprinting methods by 4.63% (communication-only) and 8.76% (vibrationonly). Additionally, our method reduces computational complexity and training data requirements, making it suitable for low-latency deployment in aerospace and defense applications. By leveraging intrinsic hardware characteristics, SaFe enhances UAV classification and aerospace communication security, ensuring resilient operations in military and critical infrastructure networks.