Few-Shot Learning for Adaptive Sybil Attack Detection in the Internet of Vehicles

The Internet of Vehicles (IoV) enables safety-critical cooperative driving applications through continuousexchange of Basic Safety Messages (BSMs). However, these networks remain vulnerable to Sybil attacks, inwhich a single malicious node fabricates multiple identities to disrupt traffic information, manipulate votingprotocols, and create phantom congestion. Existing machine learning defences rely on large labelled datasetsand must be fully retrained whenever attackers develop new variants—a response time incompatible withthe dynamic nature of vehicular environments. This paper presents FSL-Sybil, a Prototypical Network-based few-shot learning framework that detects novel Sybil attack variants with as few as two labelledsamples, without retraining the underlying model. We construct a realistic simulation environment usingOMNeT++ 6.1, the veins_inet framework, and SUMO 1.20.0 on a real-world road network of Indiranagar,Bengaluru (India), and implement three distinct Sybil attack types: random position fabrication (SA-RND), trajectory fabrication (SA-TRJ), and coordinated sophisticated attacks (SA-CRD). We extract 20discriminativefeaturesacrossBSM,mobility,andnetwork-layerdomainsandevaluatetheproposedapproachagainst Random Forest, SVM, and k-NN baselines across 10 scenarios (65,087 samples total). Under a novelattackadaptationscenarioinwhichSA-CRDattacksarewithheldfromtraining, ourmethodachieves99.88%detection recall with only one support sample per class, while the naive baseline Random Forest recall dropsto 82.23%. Computational profiling on CPU hardware demonstrates that the proposed model occupies only70 KB and sustains 1.4 million inference samples per second—1.7×faster than Random Forest and 26×faster than k-NN—making it suitable for deployment on resource-constrained Road Side Units. We’ve alsoidentified a critical vulnerability: while RSSI-based features are highly discriminative under ideal conditionsthey represent a single point of failure in adaptive spoofing attacks.