OPTIMIZING DIABETES DETECTION USING STACKING OF ADVANCED ENSEMBLE MODELS

Abstract

This paper explores the effectiveness of stacking ensemble models for improving the accuracy of diabetes detection using machine learning. We utilized Random Forest, Gradient Boosting, and AdaBoost as base learners, with Logistic Regression serving as the meta-learner. After applying hyperparameter tuning with GridSearchCV, the performance of the stacked model improved substantially. The results demonstrate a clear progression in model accuracy, from individual base models (achieving a maximum of 74.2%) to the final optimized stacked model, which achieved an accuracy of 91.3%. This research highlights the potential of stacking and hyperparameter optimization in enhancing the predictive accuracy of medical diagnosis systems, particularly for diabetes detection.