What question did this study set out to answer?

This research aims to enhance diabetes risk prediction using an explainable ensemble learning approach.

March 22, 2026Open Access

Explainable Ensemble Learning for Diabetes Risk Prediction Using SHAP Stability Analysis

AMAniket MishraEindhoven University of Technology

Key Points

This research aims to enhance diabetes risk prediction using an explainable ensemble learning approach.
Utilized an ensemble learning strategy with the Pima Indians Diabetes dataset.
Implemented median imputation for handling missing values and removed 'SkinThickness' feature.
Evaluated Random Forest model and employed soft voting ensemble for final predictions.
Achieved an ROC-AUC of 0.821, recall of 0.685, and F1 score of 0.667.
Identified Glucose as a dominant predictor through SHAP-based stability analysis.
Demonstrated clinically interpretable results with effective risk stratification.

Abstract

In this study, we use an ensemble learning approach for diabetes risk prediction using the Pima Indians Diabetes dataset.Missing values were handled using median imputation, and the feature "SkinThickness" was removed as it had very little impact on modelperformance. Random Forest, XGBoost, and LightGBM models were evaluated, and the final prediction was obtained using a soft votingensemble strategy. The proposed ensemble achieved ROC-AUC = 0.821, recall = 0.685, and F1 = 0.667. SHAP-based interpretation andstability analysis consistently identified Glucose, BMI, DiabetesPedigreeFunction, Age, and Pregnancies as dominant predictors. Riskstratification, fairness evaluation, and decision curve analysis demonstrated clinically interpretable and reliable behavior. The resultssuggest that explainable ensemble models provide effective support for early diabetes screening.

AI से पूछें

Bookmark

View Full Paper