Diabetes Classification Using a Framework Stacking of BiLSTM, Logistic Regression, and XGBoost

Abstract

Diabetes is a chronic condition that requires accurate and timely diagnosis for effective management and treatment. This study introduces an innovative approach to diabetes classification using a stacking framework that combines Bidirectional Long Short-Term Memory (BiLSTM), Logistic Regression, and XGBoost. The study employed an experimental approach by implementing the stacking framework. The two base models used were BiLSTM and Logistic Regression, with BiLSTM achieving an accuracy of 0.9935 and Logistic Regression reaching 0.9869. The stacking framework with XGBoost as the meta-learner achieved a perfect accuracy of 1.0. These findings demonstrate the potential of the stacking framework to improve diabetes classification performance compared to using individual models alone.