Leakage-safe receiver-derived GNSS interference recognition using navigation-level observables

Abstract

Global Navigation Satellite Systems (GNSS) constitute a critical positioning layer for intelligent transportation systems, including road mobility, aviation navigation, maritime operations, and emerging autonomous transport applications. The growing exposure of GNSS-enabled transport services to spoofing, jamming, and meaconing attacks creates a pressing need for deployable integrity-monitoring approaches capable of identifying interference without relying on specialized radio-frequency hardware or raw IQ signal access. This paper proposes a leakage-safe receiver-derived framework for GNSS interference recognition based exclusively on navigation-level and receiver-level observables available from standard commercial GNSS devices. The methodology integrates positioning trajectories, radial drift descriptors, local instability metrics, horizontal accuracy estimates, pseudorange-based statistics, Doppler-related variables, and integrity-oriented risk indicators extracted from NAV-PVT, RINEX, and MON-RF measurements. Using the publicly available JammerTest 2024 dataset, a comprehensive preprocessing and feature-engineering pipeline was implemented to construct window-level representations of interference-induced positioning degradation. To avoid overly optimistic performance estimates caused by temporal and scenario overlaps, model evaluation was performed using repeated scenario-group holdout splitting, ensuring strict separation between training and testing scenarios. Window-level classification was subsequently carried out using an ExtraTrees ensemble classifier on 2,939 integrity windows derived from 44,639 positioning samples across 24 spoofing, jamming, and meaconing scenarios. Under this leakage-safe evaluation protocol, the proposed framework achieved a mean classification accuracy of 62.86 % ± 9.58 % and a mean macro-F1 score of 0.5490 ± 0.0980 across repeated scenario-group splits. Feature-importance analysis identified radial drift descriptors, Doppler-related observables, and pseudorange-based statistics among the most discriminative receiver-derived variables. Although classification performance remained lower for meaconing than for spoofing and jamming, the results demonstrate that standard receiver outputs retain operationally meaningful information for interference-aware positioning monitoring even without direct signal-level access. The proposed framework provides a reproducible baseline for resilient GNSS integrity monitoring and practical interference recognition in intelligent transportation systems operating under degraded navigation conditions.