Felgnn-Dp: Federated Ensemble Learning-Based Graph Neural Network for Disease Prediction

Recent advancements in artificial intelligence and machine learning have transformed healthcare analytics, enabling insights that improve outcomes. However, privacy concerns and stringent data governance regulations pose significant challenges to sharing medical data across institutions. To address this, we propose a novel Federated Ensemble Learning-based Graph Neural Network (FELGNN-DP) framework for disease prediction tailored for medical datasets with graph structures. By leveraging the combined power of Graph Neural Networks (GNNs) and Federated Learning (FL), our framework facilitates collaborative analysis while preserving data privacy. The proposed FGNN framework integrates adaptive client selection, identifying the most informative and representative clients to enhance model performance. Experimental results demonstrate the efficacy of FELGNN-DP in predicting Alzheimer's disease, achieving an accuracy of 85% on the ADNI benchmark dataset-an improvement of approximately 12 % over state-of-the-art methods. This work highlights the potential of FL-based GNNs to balance data utility, privacy, and model accuracy, marking a significant step forward in privacy-preserving healthcare analytics.