Abstract

This paper explores the application of Federated Learning (FL) with Differential Privacy (DP) in healthcare predictive modeling. The inherent sensitivity of healthcare data necessitates robust privacy-preserving techniques. Federated learning enables collaborative model training across multiple healthcare institutions without direct data sharing, while differential privacy adds noise to the model updates to further protect individual patient data. This research investigates the trade-off between privacy protection (measured by the privacy budget, epsilon) and model accuracy (data utility) in the context of predicting patient readmission rates. We present a novel framework integrating federated averaging with Gaussian differential privacy and evaluate its performance on a synthetic healthcare dataset. The results demonstrate the feasibility of achieving acceptable prediction accuracy while maintaining a reasonable level of privacy protection, highlighting the potential of this approach for advancing collaborative healthcare research in a privacy-conscious manner.