Stacked Ensemble Learning for Type 2 Diabetes Prediction Using the MIMIC-IV Dataset

Diabetes mellitus is a major public health concern, making early and accurate prediction critical for effective management and complication prevention. Existing studies have struggled to achieve the accuracy necessary for reliable clinical application. Therefore, we propose a machine learning framework for predicting type 2 diabetes using the Medical Information Mart for Intensive Care IV (MIMIC-IV) dataset. The pipeline includes rigorous preprocessing to handle missing values and outliers, application of the Synthetic Minority Over-sampling Technique to mitigate class imbalance, and training of multiple classifiers. To further enhance performance, we propose ensemble models based on voting and stacking, where KNN, RF, CatBoost, and Extra Trees serve as base learners and Logistic Regression is employed as the meta-learner. Experimental results demonstrate that the stacking-based ensemble consistently outperforms all traditional models, achieving accuracy of 85.94%. This study introduces a robust ensemble-based methodology that addresses class imbalance, offering a promising reliable tool in diabetes care.