Federated Learning with Privacy-Preserving Big Data Analytics for Distributed Healthcare Systems

Abstract

An architecture for privacy-preserving federated learning that can classify chest X-ray pictures of tuberculosis (TB) in decentralized healthcare settings. The suggested solution protects patient privacy and ensures regulatory compliance by facilitating the cooperative training of ML models across many healthcare organizations without necessitating direct access to private patient data. A structured data pre-processing pipeline is implemented, including initial inspection, image resizing to 224×224 pixels, normalization, and class balancing using Synthetic Minority Oversampling Technique (SMOTE) to manage non-IID distributions across federated clients. The federated setup simulates realistic clinical environments where each node holds a portion of the dataset and only shares model updates with the central server. The ResNet deep learning model is deployed as the primary classifier and its performance is evaluated against Dense Net and Squeeze Net using four key evaluation metrics: accuracy, precision, recall, and F1-score. Compared to the comparison models, the suggested ResNet model achieves better performance according to accuracy (96.7%), precision (96.8%), recall (98.0%), and F1-score (97.4%). Squeeze Net achieved a rate of 94.18% accuracy and Dense Net 94%. This framework's integration with cloud-based platforms increases its scalability and real-time applicability. It offers a secure, scalable, and high-performance solution for tuberculosis (TB) diagnosis in healthcare environments through the use of federated learning and big data analytics. The results validate its potential as a foundation for broader applications in privacy-aware medical AI systems.