Federated Learning and Edge AI for Privacy-Preserving Diabetes Prediction in Healthcare

The rapid expansion of digital health systems has resulted in large volumes of sensitive patient data, which if analyzed effectively, can significantly enhance disease prediction, diagnosis, and treatment approaches. Conventional machine learning models require centralized data collection, which poses serious concerns about patient privacy, data breaches, and compliance with regulations such as HIPAA and GDPR. This study presents a novel predictive healthcare framework that utilizes Federated Learning (FL) and Edge Artificial Intelligence (Edge AI) for decentralized and privacy-preserving analytics, specifically on diabetes prediction. To overcome these issues, the proposed solution is to train machine learning models locally in healthcare institutions and/or on IoT-enabled edge devices, using real-time patient data. The central federated server then receives only encrypted model updates with no raw patient data and performs Federated Averaging to create a global model. To ensure strict data protection, differential privacy techniques are employed to guard against re-identification and membership inference attacks. The system was evaluated using a partition of 327 of the PIMA Indian Diabetes Dataset, distributed across five virtual hospital nodes. Using the federated model accuracy was found to be 91.2%, precision 89.5%, recall 88.0%, and f1-score 88.7%. Inference latency was kept below 50 milliseconds and the bandwidth utilization decreased by 87% when compared to centralized techniques. The findings show that the proposed architecture maintains competitive performance while significantly improving security, compliance, and real-time usability. The study outlines a model for scalable, smart, and ethical healthcare systems. The combination of Federated Learning and Edge AI makes the proposed model deployable in many clinical settings especially in sensitive and resource-constrained environments, revolutionizing the field of digital health prediction and management.