International Journal of Latest Technology in Engineering Management & Applied Science · Nov 3, 2026

An Adaptive Federated Machine Learning Framework for Privacy-Preserving Prediction of Type 2 Diabetes Mellitus across Healthcare Institutions

By OLAOYE OJ, FOLARIN OO, Saheed OA

There are challenges in collaboration between health care institutions in the areas of confidentiality, governance considerations and information system integration that hinder the accurate prediction of Type 2 diabetes mellitus to enable timely preventive interventions. This is further complicated by differences in the populations of patients, number of patients in datasets, the computing resources available and participation. In this study, a novel adaptive federated machine learning approach for privacy-preserving diabetes onset prediction within three years is proposed. The methodological design calls for 5,000 synthetic records of adult patients, across five simulated hospitals. Preprocessing consists of duplicate removal, harmonisation of variables, missing value imputation, categorical encoding, numerical scaling, and class weighting, and is only applied to the training data. Prediction model gets local training, aggregation weight based on model validation performance and update reliability and participation history. Differential privacy and secure aggregation are defined as complementary protection measures. Discrimination, sensitivity, F1 score, calibration, institutional consistency, and computational efficiency are used as evaluation metrics for comparison between local-only, centralised, FedAvg, FedProx, and adaptive approaches. Illustrative performance targets are an AUROC of 0.9000, sensitivity of 0.8200, and F1 score of 0.8500. An efficiency illustration shows that the training time is reduced by 25%, the volume of communication is reduced by 33.3%, and convergence is achieved in 60 rounds, when compared to the efficiency illustrated with FedAvg. These numbers are not experimentally validated and differences in institution performance, confidence intervals and numerical privacy budgets are not measured. The contribution is an integrated design connecting adaptive collaboration, explicit privacy controls, and comprehensive evaluation. To draw conclusions on effectiveness and readiness for deployment in different clinical contexts, the implementation needs to be reproducible, privacy accounting must be done, and external clinical validation must be conducted.

doi.org/10.51583/ijltemas.2026.151000044Open accessBiomarkers

AgingResearch.News