Explainable Federated Machine Learning for Diabetes Risk Stratification: A Multi-Algorithm Evaluation with SHAP-Guided Clinical Interpretability
Year: 2026
Type: article
Abstract: The escalating global burden of type 2 diabetes mellitus (T2DM) demands robust, interpretable, and privacy-preserving predictive frameworks that can operate reliably in federated clinical environments. Existing machine learning (ML) approaches for diabetes prediction are predominantly developed on single, demographically constrained datasets and lack mechanisms for transparent clinical decision support. This study introduces an Explainable Federated Machine Learning (EFML) framework that integrates five classification algorithms—Logistic Regression, Support Vector Machine (SVM), Random Forest, XGBoost, and Multi-Layer Perceptron (MLP)—applied to the Pima Indians Diabetes dataset with systematic preprocessing, domain-driven feature engineering, and multi-strategy class imbalance remediation. The proposed framework incorporates SHAP (Shapley Additive Explanations) and LIME (Local Interpretable Model-Agnostic Explanations) as dual-layer explainability components, enabling both global and instance-level transparency. The MLP model achieved the highest classification accuracy of 88.6% with an AUC of 0.91, while XGBoost delivered superior precision-recall balance (F1 = 0.80, AUC = 0.83) with markedly lower computational overhead. SHAP analysis confirmed glucose concentration, BMI, and insulin resistance indices as the dominant predictive biomarkers. The study further delineates a federated learning integration pathway for privacy-compliant multi-institutional deployment using the FedAvg protocol, contributing a deployable, regulation-aware blueprint for AI-augmented diabetes screening systems.
Language: English
Subfield: Health Information Management
Field: Health Professions
Domain: Health Sciences
Open Access status: diamond
OpenAlex ID: W7160052691