@Article{jiot.2026.078248,
AUTHOR = {Nomangwane Angelina Tshabalala, Ping Guo},
TITLE = {A Federated Learning Framework with Blockchain for Privacy-Preserving Continuous Glucose Monitoring in Type 2 Diabetes},
JOURNAL = {Journal on Internet of Things},
VOLUME = {8},
YEAR = {2026},
NUMBER = {1},
PAGES = {87--107},
URL = {http://www.techscience.com/jiot/v8n1/67236},
ISSN = {2579-0080},
ABSTRACT = {Type 2 Diabetes mellitus is a disease that afflicts approximately 537 million individuals all over the world, and continuous glucose monitoring (CGM) systems have become very important in the management of the disease. Nonetheless, the existing centralized data architecture of CGM generates high privacy and security risks, as sensitive patient health data can be easily abused. This paper introduces an original structure that incorporates both federated learning and blockchain technology and allows for predicting glucose safely and preserving privacy without affecting the integrity of the data. Our model uses the Long Short-Term Memory (LSTM) neural networks that are trained through the Federated Averaging (FedAvg) algorithm across distributed patient devices, such that the raw CGM data does not leave local storage. An algorithmic blockchain based on Hyperledger Fabric captures cryptographic hashes of model updates, generating an unalterable audit trail that prevents model poisoning and provides integrity verification. We applied and tested a full prototype on 10 simulated patients (modeled on OhioT1DM patterns) through various rounds of federated learning. The experimental findings indicate that our method has a Root Mean Square error (RMSE) of 11.37 ± 0.85 mg/dL and a Mean Absolute error (MAE) of 9.09 ± 0.68 mg/dL in predicting glucose and only ~12% privacy overhead. The blockchain element supports both transaction latencies of 8–12 ms with cryptographic guarantees of model integrity. Model-only transmission saves 78 percent on the cost of communication in ongoing continuous learning scenarios when compared to centralized methods. This paper offers a practical, proof-of-concept privacy protecting diabetes management solution that balances clinical utility with patient privacy.},
DOI = {10.32604/jiot.2026.078248}
}