A Secure Federated GNN Framework for Healthcare Prediction
Year: 2026
Type: conference-paper
Abstract: Artificial Intelligence (AI) and Machine Learning (ML) have greatly enhanced several areas of healthcare, including disease prediction and clinical decision-making, but centralised models are negatively affecting patient data privacy and security. This paper presents a secure healthcare prediction framework based on Federated Learning (FL) and Graph Neural Networks (GNNs). This proposed framework enables healthcare providers to collectively build a global model without any shared personal patient information. FedAvg (Federated Averaging) is utilised for efficient model aggregation, while GNNs improve prediction accuracy by capturing complex, interconnected dependencies across heterogeneous healthcare data. The experimental evaluation shows that the proposed framework provides satisfactory prediction accuracy when compared with centralised approaches, along with consistent convergence during federated training rounds. The obtained results indicate that the framework effectively addresses several major challenges present in healthcare systems, including but not limited to data heterogeneity, privacy preservation, and collaboration among many cooperating healthcare providers. Tests conducted using the UCI Heart Disease dataset show that the developed model attains nearly 84% accuracy with consistent performance throughout federated training rounds. In comparison with existing approaches, the proposed FL-GNN attains 83.84% accuracy, exceeding the conventional federated learning baseline (81.84%) by nearly 2% and approaching the centralised upper bound (89.23%) without sharing raw patient data.
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W7225199442