Synergistic ensemble classification framework: utilizing a soft voting algorithm for enhanced prediction and diagnosis of diabetes mellitus

Abstract

Diabetes, a serious condition characterized by elevated blood glucose levels, can be effectively identified, and predicted early using machine learning (ML) algorithms. The research provides a comprehensive assessment of three ensemble ML models-stacking, soft voting, and hard voting-focused on enhancing diabetes diagnosis among Pima Indian women dataset taken from the National Institute of Diabetes and Digestive and Kidney Diseases, this study focuses on Pima Indian women aged 21 and older, with the dataset comprising critical diagnostic measurements. Two ensemble models were developed and evaluated on various evaluation parameters. The stacking model combines predictions from various classifiers using a meta-classifier, leveraging their strengths for final decision-making. In contrast, the voting model aggregates probability estimates from each classifier, providing nuanced predictions. Both models were rigorously evaluated on a validation dataset, emphasizing accuracy, specificity, sensitivity, and the receiver operating characteristic (ROC) area under the curve (AUC). Notably, the voting-based ensemble methods demonstrated superior performance in predicting diabetes for this cohort. However, their effectiveness heavily relies on preprocessing, base model selection, and hyperparameter optimization. This study underscores the potential of ensemble models in medical diagnostics, highlighting the critical role of data preprocessing, and configuration in enhancing predictive accuracy.