Privacy-Preserving Chronic Kidney Disease Prediction Through Federated Learning Optimization Techniques

Abstract

Chronic Kidney Disease (CKD) is a major global health concern that requires accurate and timely prediction for effective diagnosis and clinical decision-making. However, conventional centralized machine learning approaches require sensitive patient data to be collected and shared at a central location, raising significant privacy and security concerns. To address these challenges, this paper proposes a Federated Learning (FL)-based framework for privacy-preserving CKD prediction, integrating multiple machine learning classifiers with advanced federated optimization techniques. The proposed framework enables collaborative model training across distributed healthcare environments without directly sharing patient data. Several machine learning classifiers are evaluated under different federated optimization strategies to identify the most effective combination for CKD prediction. Experimental results demonstrate that the FL-FedProx with XGBoost model achieves the best performance, obtaining an accuracy of 99.76%, precision of 99.58%, recall of 99.42%, and F1-score of 99.50%. The results indicate that the FedProx strategy effectively addresses data heterogeneity and non-IID characteristics while minimizing client drift during distributed training. Furthermore, the integration of XGBoost enhances the predictive capability of the proposed framework. Overall, the proposed FL-FedProx–XGBoost framework provides an accurate, scalable, and privacy-preserving solution for CKD prediction, demonstrating its potential for deployment in distributed healthcare environments and real-world clinical decision-support systems.