Is this project an undergraduate, graduate, or faculty project?
Undergraduate
Project Type
group
Campus
Daytona Beach
Authors' Class Standing
Nicolas Machado, Junior Jaxon Selzer
Lead Presenter's Name
Nicolas Machado
Lead Presenter's College
DB College of Engineering
Faculty Mentor Name
Dr. Omar Ochoa
Abstract
An Evaluation of Machine Learning Models' Efficacy in Determining UAV Spoofing Attacks - The rapid integration of Unmanned Aerial Vehicles (UAVs) into urban airspace has introduced significant cybersecurity concerns, particularly due to vulnerabilities in Automatic Dependent Surveillance–Broadcast (ADS-B), which lacks authentication and encryption. This project addresses the problem of detecting spoofing and data manipulation attacks that can compromise UAV safety and mission reliability. The objective of this work is to evaluate the effectiveness of machine learning–based anomaly detection, specifically Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) networks, as protocol-agnostic solutions for identifying anomalous UAV behavior. To achieve this, realistic UAV flight trajectories were generated using Microsoft AirSim, and multiple spoofing scenarios were simulated by injecting controlled deviations into telemetry data across varying magnitudes and durations. Both models were trained on time-series telemetry data and evaluated using performance metrics such as recall, precision, and F1-score. Results indicate that LSTM and GRU models can effectively detect moderate to large spoofing perturbations, with GRU demonstrating slightly improved recall and faster convergence. However, both models show limited sensitivity to subtle or single-point spoofing events. These findings highlight the feasibility of recurrent neural networks for UAV anomaly detection while emphasizing the need for further research into improving detection sensitivity and enabling real-time deployment in complex urban environments.
Did this research project receive funding support (Spark, SURF, Research Abroad, Student Internal Grants, Collaborative, Climbing, or Ignite Grants) from the Office of Undergraduate Research?
No
Included in
Artificial Intelligence and Robotics Commons, Aviation Safety and Security Commons, Cybersecurity Commons
An Evaluation of Machine Learning Models' Efficacy in Determining UAV Spoofing Attacks
An Evaluation of Machine Learning Models' Efficacy in Determining UAV Spoofing Attacks - The rapid integration of Unmanned Aerial Vehicles (UAVs) into urban airspace has introduced significant cybersecurity concerns, particularly due to vulnerabilities in Automatic Dependent Surveillance–Broadcast (ADS-B), which lacks authentication and encryption. This project addresses the problem of detecting spoofing and data manipulation attacks that can compromise UAV safety and mission reliability. The objective of this work is to evaluate the effectiveness of machine learning–based anomaly detection, specifically Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) networks, as protocol-agnostic solutions for identifying anomalous UAV behavior. To achieve this, realistic UAV flight trajectories were generated using Microsoft AirSim, and multiple spoofing scenarios were simulated by injecting controlled deviations into telemetry data across varying magnitudes and durations. Both models were trained on time-series telemetry data and evaluated using performance metrics such as recall, precision, and F1-score. Results indicate that LSTM and GRU models can effectively detect moderate to large spoofing perturbations, with GRU demonstrating slightly improved recall and faster convergence. However, both models show limited sensitivity to subtle or single-point spoofing events. These findings highlight the feasibility of recurrent neural networks for UAV anomaly detection while emphasizing the need for further research into improving detection sensitivity and enabling real-time deployment in complex urban environments.