AI Analytics for V2X Misbehavior Recognition Using Qubit-Aided Feature Encoding

Abstract

Vehicle-to-everything (V2X) communication extends automated driving perception beyond line of sight, yet it also creates a data-trust problem: authenticated messages can still contain false, stale, or strategically manipulated mobility information. This article develops an AI analytics framework for V2X misbehavior recognition using qubit-aided feature encoding (QAFE), a quantum-inspired representation layer that maps normalized kinematic, temporal, spatial-consensus, and radio-context indicators into nonlinear interaction features before classical classification. The study is framed as a simulation-based benchmark using a VeReMi-style message corpus with 1.25 million cooperative awareness and basic safety messages, six label categories, scenario-holdout validation, and operational metrics that include macro-F1, attack recall, false-report rate, calibration error, and edge inference latency. Results show that the proposed QAFE-gradient boosting model improves macro-F1 from 0.909 for an XGBoost baseline and 0.925 for a compact Transformer to 0.935, while reducing minority-class recall gaps under sudden-stop and Sybil-like replay attacks. Ablation analysis indicates that pairwise phase terms and temporal consistency windows account for most of the performance gain. The framework does not claim near-term quantum hardware advantage; instead, it demonstrates how qubit-inspired feature maps can provide a disciplined way to construct high-order interactions for safety-critical V2X analytics under limited latency and interpretability constraints

Keywords: V2X security; misbehavior detection; AI analytics; quantum-inspired learning; qubit feature maps; cooperative intelligent transportation systems; edge inference