Privacy-Preserving Federated Learning With Blockchain Auditing for Consumer Electronics Applications

Consumer Electronics (CE) systems increasingly use artificial intelligence (AI), machine learning (ML), and federated learning (FL) for data-driven intelligence. This integration introduces risks such as privacy leakage during aggregation, poisoning of model updates, and integrity threats from unverified participants. Differentially private FL reduces leakage but often degrades accuracy by more than 10%, while blockchain (BC)-based FL improves integrity at the cost of up to 12.9% overhead. To address these issues, a hybrid FL framework with BC auditing is proposed. The design combines secure aggregation (SA), robust global update (RGU) rules , and lightweight BC consensus to defend against adaptive adversaries. Experimental results show an 8.3% accuracy gain under non-IID data, a 30.9% reduction in privacy leakage, and a poisoning-induced accuracy drop limited to 6.1% compared to 23.9% in standard FL. The framework maintains 97.8% BC integrity with only 3.2% overhead, demonstrating a balanced trade-off between accuracy, privacy, robustness, and integrity for real-world CE deployments.