Diabetes prediction needs tools that protect privacy and scale across varied hospitals, clinics, and wearables with nonidentical data distributions. Traditional machine learning uses central data hubs. This raises privacy risks and limits generalizability in heterogeneous settings. Institutions training alone miss collaborative power.
Objective Builds Secure Collaborative Framework
In the study published in the JMIR Diabetes, the researchers aimed to create FedEnTrust, a federated ensemble system for diabetes prediction. It handles data heterogeneity and varying compute power without raw data sharing. Blockchain adds access control, while knowledge distillation boosts model consensus across participants.
Methods Enable Decentralized Training
FedEnTrust lets sites train locally and share only soft labels for weighted voting into a global model. It assigns architectures by local capacity and uses blockchain smart contracts for registration, roles, and secure submissions. Tests used PIMA Indians Diabetes Dataset for accuracy, efficiency, and blockchain metrics.
Results Show Strong Performance Gains
The framework reached 84.2% accuracy, with precision at 84.6%, recall at 88.6%, and F1-score at 86.4%. It beat other decentralized models and matched centralized deep learning. Blockchain hit 100% secure transactions with 210 ms latency and ~107,940 gas units. Privacy stayed intact via metadata only.
Conclusions Pave Way for Real-World Use
FedEnTrust delivers accurate, scalable diabetes prediction with top privacy and trust. It proves federated ensembles can replace central AI in diverse healthcare, balancing ethics and performance.
Featured
Off
Page Content
#ffffff
Anonymous user
On
Authenticated user
On
Premium
On
Paid / Sponsored
On
Key highlights
- FedEnTrust achieved 84.2% accuracy on PIMA dataset, with 84.6% precision, 88.6% recall, and 86.4% F1-score, rivaling centralized models.
- Blockchain smart contracts ensured 100% success for authorized transactions while blocking all malicious or unauthorized attempts.
- Average blockchain latency was 210 ms with ~107,940 gas units, supporting real-time secure coordination across participants.
- Framework preserves privacy by sharing only model metadata and soft labels, never raw patient data.
- It handles heterogeneous data and compute via adaptive architectures and distillation for scalable diabetes prediction.
Source
Hasan MR, Li J. Privacy-Preserving Collaborative Diabetes Prediction in Heterogeneous Health Care Systems: Algorithm Development and Validation of a Secure Federated Ensemble Framework. JMIR Diabetes. 2026;11:e79166-e79166. doi: https://doi.org/10.2196/79166
Thumbnail
Speciality
Currency
Sub Speciality
Sub Sub Speciality
Short Description
FedEnTrust integrates federated learning, ensembles, blockchain, and distillation for privacy-safe diabetes prediction across diverse healthcare sites, hitting 84.2% accuracy without sharing raw patient data.
User Segments
Release Date
Featured Order
0
Is Paid
0
Send Notification
Off