Proactive Threat Hunting and Anomaly Detection in Consumer Electronic Ecosystems

GPS-enabled consumer electronics in intelligent transportation systems are increasingly vulnerable to signal-level cyberattacks such as spoofing and jamming, which traditional intrusion detection systems struggle to address due to their reliance on extensive labeled data and retraining. This paper presents a proactive threat hunting framework based on uncertainty-aware few-shot learning for GNSS security. The proposed approach enables detection of novel attack patterns using only one to three labeled examples, without requiring model training or adaptation. We introduce an Uncertainty-Aware Prototypical Network (UAPN) that models class representations as Gaussian distributions and performs classification using a hybrid Euclidean-Mahalanobis distance with data-driven variance estimation. This design enables robust decision-making under limited data conditions. Experimental results show that UAPN achieves 80.6% accuracy at K = 1 and 88.3% at K = 3, consistently outperforming five state-of-the-art few-shot learning baselines. Evaluation on multiple GNSS datasets (GPS, BeiDou, and GLONASS) demonstrates strong generalization and robustness in low-data regimes. By operating in a non-training (cold-start) setting with competitive inference time, the proposed framework is well-suited for real-time deployment in resource-constrained consumer electronics.