Generative AI-driven secure and privacy-preserving multi-source data fusion for intelligent transport systems

Intelligent Transport Systems (ITS) use different kinds of data from cars, road infrastructure, and users to make traffic safer, more efficient, and smarter. However, the integration of different sources of data, including IoT sensor measurements, GPS tracks, CCTV signals, and vehicle telematics, which renders ITS systems vulnerable to considerable security and privacy threats, such as data poisoning, inference-based attacks, unauthorized profiling, and personal data leaks. The current solutions typically address these challenges individually and often employ centralized or partially decentralized architectures, which limit scalability and resilience. To address these shortcomings, this paper proposes a new generative AI-based model for secure and privacy-preserving data fusion in ITS. In contrast to previous models, the proposed architecture combines federated learning, differential privacy, generative models, and blockchain technology into a single, decentralized security system. Federated learning allows collaborative training of models without the transfer of raw data, differential privacy is a formal constraint on the leakage of information, generative models under non-IID data conditions enhance robustness to artificial data along with anomaly detection, blockchain is a formal guarantee that the message is not tampered with during aggregation, a transparent audit is offered, and distributed ITS stakeholders can trust each other. This close adaptation of technologies offers simultaneous immunity against data poisoning, model inversion, and membership inference attacks, an issue that has not been fully considered in the current literature on ITS data fusion. The suggested framework is tested on three benchmark datasets, including i-VANET, ToN-IoT, and CIC-IDS 2017, which yield average classification rates of 97.21%, 97.75%, and 96.66%, respectively. The proposed method is better than the best current solutions when it comes to accuracy in detection, privacy protection, and assurance of security. This shows that it works well, can be used in real-world ITS systems, and can be used in many different ways.