Prediction for diabetes by a weighted voting classifier based on normalized AUC with balancing factors

Abstract

Diabetes is a metabolic disease characterized by hyperglycemia caused by insulin deficiency or resistance. Diabetes can lead to various co-morbidities and is a common medical problem worldwide. The prevention and control of diabetes would benefit from accurately identifying diabetic individuals. In this study, we test the best pipeline of different machine learning (ML) models (K-Nearest Neighbor, Random Forest, Support Vector Machine, XGBoost) under different pre-processing. Also, this study investigates and proposes a weighted soft-voting classification model (NAWVE), which uses the AUC obtained from the base model on the training set and adds a balancing factor as a weight after normalization. This integrated model can well balance different performance classifiers. In the prediction of diabetes, NAWVE received the highest scores in four metrics, with accuracy, AUC, f1, and recall of 0.9606, 0.9621, 0.9472, and 0.9677, respectively. Our proposed fusion model outperformed any individual classification model and Stacking with meta-learner as logistic regression.