Skip to content
Open access

Privacy-Preserving Federated Learning for Healthcare Analytics: Integrating Differential Privacy and Secure Aggregation for Predictive Modeling

2026 · Journal of Machine Learning Innovations and Artificial Intelligence Horizons · 0 citations

Abstract

Healthcare data is often unique in that privacy laws often restrict the ability to share patient information centrately, and as a result, there's an increasing need for predictive healthcare analytics often limited by privacy regulations. We propose a federated learning based distributed system that exploits both the differential privacy technique and the secure aggregation to cooperatively train a binary disease classification system, without sharing private valuable patient information, across multiple hospitals. Each participating institution does a local stochastic gradient descent, and applies gaussian noise (clipping) to the updates as well as satisfies DP constraints in each round of communication. Additionally, protocols have been developed to allow for secure aggregation such that the central server only be able to determine the contribution as a sum of the individual contributions and not the contribution from any one hospital. The server then averages (federated averaging) each local model with its weighting based on the number of samples and gets an enhanced global model. A two-layered neural network with a sigmoid activation function output is trained using Adam and binary cross-entropy function for the classification problem.For the classification problem, a two-layered neural network with a sigmoid activation function output is trained with the Adam optimizer and binary cross-entropy function. This was enough security to make it difficult for the server or others to be able to “reconstruct” the patient-level data that was sent. In summary, the proposed principle is seemly practical and can be usefully used to collaboratively construct models across clinical settings, where privacy safeguards and prediction performance logic dictated in a delicate tradeoff fashion. What we provide is a system on how to integrate such complementary privacy mechanisms into a single federated pipeline, but that goal extends beyond the case where data cannot be shared out of the box as we did above.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.