Author: Lanxiang Zhang
Supervisor: Mubashar Iqbal
Abstract: The rapid development of the Internet of Vehicles (IoV) has created unprecedented demands for secure, scalable, and privacy-preserving edge computing solutions. While Federated Learning (FL) offers a promising approach to collaboratively train machine learning models across distributed devices without exposing raw data, traditional FL frameworks remain dependent on centralized aggregators, introducing single points of failure, data poisoning attacks, and trust issues in dynamic vehicular environments. To address these challenges, we propose an AI-driven blockchain-based FL framework for IoV that integrates decentralized consensus mechanisms, cryptographic validation, and robust aggregation techniques to enable secure and efficient collaborative learning among untrusted edge devices. The framework leverages a Hyperledger Fabric-based permissioned blockchain network to manage tamper-resistant records of model updates, enforce dynamic reputation systems, and distribute incentives for participation. By combining differential privacy and Byzantine-resilient aggregation, our method significantly reduces the risks of data leakage and model poisoning. Experimental evaluations in a simulated IoV environment demonstrate that the proposed approach achieves comparable accuracy to centralized learning (92.4\% mAP vs. 93.1\% mAP) while reducing attack success rates from 78.5\% to 3.2\% and preserving strong privacy guarantees. This work advances the state of the art in decentralized machine learning and provides a practical foundation for privacy-preserving, trustworthy intelligence in intelligent transportation systems. While existing approaches, such as differential privacy and secure multi-party computation, offer partial protection, they often introduce high computational costs or rely on unrealistic trust assumptions. Blockchain technology, though promising for decentralizing FL workflows, is still challenged by trade-offs between security and scalability—especially in highly mobile, intermittently connected IoV settings.