Leveraging Open Data and Machine Learning to Minimize the Risks of Aviation Satellite Navigation Interference

• Celso Guilherme Ferreira de Moura
• João Souza Dias Garcia
• Andrei Piccinini Legg

Embry-Riddle Aeronautical University · Universidade Federal de Santa Maria

Event

XXII SITRAER – Brasília, DF, Brasil, October 22–24, 2025

Volume

Anais do XXII Simpósio de Transporte Aéreo – SITRAER 2025 · ISBN 978-65-272-2024-4 · DOI 10.29327/9786527220244

Publisher

Even3 Publicações

Editors

Fabiana Serra de Arruda, Camila Lucena, Giovanna Miceli Ronzani Borille, Evandro José da Silva

Abstract

Global Navigation Satellite Systems (GNSS) are essential for modern aviation and provide crucial safety information in support of aircraft operations like timing, positioning, and navigation support. In recent years, the reliability of such systems has been tested by a significant increase in the cases of unintentional and intentional interference, commonly referred to as jamming and spoofing. Some progress has been made in detecting GNSS interference, but practical challenges remain. Existing approaches based on broadcast low Navigation Integrity Category (NIC) reports show promise but can sometimes miss smaller, localized, or intermittent jamming events that do not produce overt disruptions in satellite navigation systems, potentially leaving threats undetected in some operational environments. This research explores a machine learning framework to enhance the detection of GPS interference in aviation, particularly leveraging open Automatic Dependent SurveillanceBroadcast (ADS-B) to identify new reliable predictors of these events. The project also explores logistic regression with L1 regularization and LightGBM algorithms. The authors examine various predictors, including NIC-based metrics, aircraft movement variations, and detailed altitude bin distributions. Beyond commonly used variables like the percent of low NIC values, this study introduces spatial neighbor features to capture additional anomalies in regional traffic patterns and other metrics. Data from actual 2024 GNSS disruption events collected from official investigation reports by aviation and telecommunications authorities support the current analysis. The most promising results came from LightGBM models enhanced with spatial neighbor statistics, trained on a dataset that included two confirmed interference events. These models had a F1-score of 79.67% alongside an ROC AUC of 0.9582. Key predictors included mean altitude neighbor statistics and positional jump metrics. These findings show that, while existing methods provide valuable insights, there is room for refinement in detecting intermittent and more localized jamming events. The research suggests a new approach that leverages novel variables and advanced machine learning techniques, contributing to the ongoing effort to enhance aviation safety and resilience against GNSS interference.

Keywords: Aviation safety; GNSS interference; Machine Learning; ADS-B; GPS jamming detection