Enhancing Data Poisoning Detection With Federated Ensemble Learning on E-Health IoT Data
Mushira Mustafa Freihat, Tarek Moulahi, Pascal Lorenz · 2025
Federated Learning (FL) has emerged as a promising approach for privacy-preserving collaborative model training, especially in high-stakes applications such as e-health. However, FL is highly vulnerable to data poisoning attacks, where malicious clients introduce corrupted updates to compromise model integrity. This paper proposes a Federated Ensemble Learning (FEL) framework to detect and mitigate such poisoning attempts using ensemble learning techniques. FEL leverages multiple independent models to identify adversarial behavior while preserving privacy. Our method predicts and filters poisoned data to enhance disease detection accuracy. We demonstrate that FEL offers superior robustness against labelflipping attacks compared to baseline FL models. Experiments on clinical datasets confirm FEL's effectiveness in rejecting poisoned updates while maintaining high prediction accuracy (97%). FEL enhances FL security mechanisms and provides a scalable solution for decentralized healthcare systems.