Main article

Nguyễn Văn Hùng*
Faculty of Information Technology, Hanoi University of Industry, Hanoi 100000, Vietnam
hung.nguyenvan@haui.edu.vn
Trần Thị Mai
Faculty of Computer Science and Engineering, Industrial University of Ho Chi Minh City, Ho Chi Minh City 700000, Vietnam
Lê Quang Minh
Faculty of Information Technology, University of Information and Communications Technology (Thai Nguyen University), Thai Nguyen 250000, Vietnam

Abstract

Vehicular edge computing has emerged as a foundational technology for supporting low-latency analytics in connected-vehicle environments, but the open membership of these networks creates persistent risks from misbehaving participants whose actions degrade resource allocation outcomes for benign users. This article proposes an integrated framework that combines AI-driven trust scoring with a blockchain-anchored interaction history and a smart-contract allocation policy. The trust scorer is a small feed-forward neural network whose features summaries each vehicle's recent ledger history; the allocation policy combines the trust score with a confidence-aware override to handle short-history vehicles fairly. We evaluate the framework using a discrete-event simulation of a six-RSU corridor under a heterogeneous mix of reputable, new, and suspect vehicles. The proposed framework reduces successful task completion degradation from 30 percentage points (priority-only baseline) to under 7 percentage points across an adversary fraction range of 5% to 40%, achieves a discrimination AUC of 0.989 on the trust classification task, and preserves Jain's fairness index above 0.79 even under 90% offered load. The total per-allocation latency overhead is bounded at approximately 58 MS relative to the priority baseline, which corresponds to less than one meter of additional vehicular travel at typical urban speeds. We discuss the deployment considerations that determine production viability, including the choice of permissioned ledger, the model retraining cadence, the privacy implications, and the relationship to existing lower-layer security primitives. The framework is intended as a workable design pattern rather than a final proposal, and we identify federated training, reinforcement-learning extensions to the policy, and additional resource types as natural directions for future work.

Article details

How to Cite

Hùng, N. V., Mai, T. T., & Minh, L. Q. (2023). AI-Enhanced Trust Scoring for Blockchain-Based Vehicular Edge Resource Allocation. Journal of AI Analytics and Applications, 1(3), 23-41. https://doi.org/10.63646/jaiaa.2023.010302