Privacy-Preserving Collaborative Diabetes Prediction in Heterogeneous Health Care Systems: Algorithm Development and Validation of a Secure Federated Ensemble Framework | JoVE Visualize

Area of Science:

Artificial Intelligence in Healthcare
Decentralized Machine Learning
Health Data Security and Privacy

Background:

Centralized machine learning models for diabetes prediction face challenges with data privacy, regulatory compliance, and heterogeneity in healthcare data.
Existing federated learning (FL) approaches struggle with data heterogeneity, security vulnerabilities, and system coordination in decentralized healthcare environments.
Isolated model training limits generalizability and effectiveness due to non-independent and identically distributed (non-IID) data and varied computational constraints.

Purpose of the Study:

To develop a secure, scalable, and privacy-preserving framework for diabetes prediction using integrated federated learning (FL).
To address data heterogeneity, non-IID data distributions, and varying computational capacities across diverse healthcare participants.
To enhance data privacy, security, and trust in collaborative model training through ensemble modeling, blockchain, and knowledge distillation.

Main Methods:

Proposed FedEnTrust: a federated ensemble learning framework enabling collaborative model training without raw data sharing.
Utilized knowledge distillation with adaptive weighted voting for aggregating soft label outputs into a global consensus.
Implemented blockchain-enabled smart contracts for secure and transparent participant registration, role assignment, and access control.

Main Results:

Achieved 84.2% accuracy in diabetes prediction, outperforming existing decentralized models and approaching centralized deep learning benchmarks.
Blockchain smart contract demonstrated 100% success for authorized transactions and rejection of unauthorized attempts, ensuring robust security.
Preserved patient privacy by exclusively exchanging model metadata, not raw patient data, with low blockchain latency and manageable gas costs.

Conclusions:

FedEnTrust provides a deployable, privacy-preserving solution for decentralized healthcare prediction by integrating FL, ensemble methods, and blockchain.
The framework effectively balances predictive accuracy, scalability, and ethical data usage while enhancing security and trust.
Demonstrated the viability of secure federated ensemble systems as practical alternatives to centralized AI models in real-world healthcare.