Privacy-Preserving Federated Learning for Big Data Analytics in Smart Healthcare

With the digital transformation of healthcare, vast amounts of data are generated, enabling predictive analytics and personalized care. However, centralizing this data introduces serious privacy risks governed by HIPAA and GDPR. To address these concerns, this paper presents a privacy-preserving federated learning (FL) framework for big data analytics in smart healthcare using the Diabetes Health Indicators Dataset. A lightweight deep neural network was deployed across five simulated healthcare clients, integrating techniques like secure preprocessing, federated averaging, and differential privacy. The proposed federated model achieved an accuracy of 83.7%, precision of 80.4%, and recall of 84.2%, competitive with the centralized model (85.2% accuracy). The clientwise accuracy ranged from 85.7% to 91.1%, and privacy-accuracy trade-offs were quantified under varying differential privacy levels. These results validate FL as a viable and secure alternative for collaborative healthcare analytics while safeguarding sensitive patient data.