Toward the Design of a Stack-Ensemble Model for Type II Diabetes Prediction in Rural Areas

Abstract

This research presents a stack-ensemble method designed to forecast Type II diabetes in rural populations, addressing limitations in existing approaches, such as dataset imbalance, limited accuracy, Considering the fact that diabetes patients in remote areas lack access to Type II diabetes screening instruments. Leveraging the Behavioural Risk Factor Surveillance System (BRFSS) dataset for model training and local Nigerian hospital data for model validation, the framework proposes SMOTE for class balancing, Grid search for hyperparameter tuning and wrapper-based feature selection methods. Logistic Regression, KNN, AdaBoost, Naive Bayes, and a Random Forest meta-learner are strategically used as base learners in a way that can enhance the overall performance of the predictive model. The model can be evaluated in terms of accuracy, AUC-ROC, sensitivity, specificity, and precision in order to enhance early diagnosis while meeting local healthcare demands. The expected results are expected to be effective in terms of model generalization, reducing overfitting, and enhancing model performance compared to others or individual learning techniques, making it applicable for scalability in a diabetes prediction model.

Keywords: Stack-ensemble learning, Type II diabetes prediction, SMOTE, Feature selection, Hyperparameter optimization.