Secure And Auditable Traffic Analysis With Blockchain-Integrated Federated SVM

Abstract

Support Vector Machines (SVM) are widely applied in Intelligent Transportation Systems (ITS), but centralized training exposes sensitive traffic data to inference and poisoning attacks. Federated SVM mitigates direct data sharing, yet relies on a central aggregator and lacks transparent auditability. This paper introduces TABT-ML, a blockchain-enabled federated SVM framework that integrates differential privacy (DP) and homomorphic encryption (HE) to secure gradient exchange, while Ethereum smart contracts provide decentralized verification and tamper-resistant auditability. Experiments on benchmark datasets, PeMS-Bay and Kaggle Traffic Volume, show that TABT-ML achieves 93% accuracy, with only ~2.6% degradation compared to centralized SVM. Blockchain validation introduces minimal overhead (~0.002 ETH per update, ~15 s latency), confirming feasibility for near real-time ITS applications. TABT-ML unifies federated SVM, DP, HE, and blockchain under realistic cost constraints, providing robust resilience against inference and poisoning attacks, while establishing a scalable framework for privacy-preserving, auditable, and secure machine learning in ITS and other decentralized environments.