Ensemble-Based Weighted Voting Approach for the Early Diagnosis of Diabetes Mellitus

Abstract

AbstractDiabetes mellitus, shortly diabetes, is a fearsome disorder that can be characterized by elevated blood glucose levels. The appropriate use of machine learning (ML) techniques aid in the earlier diagnosis of diabetes. The main goal of this research is to use an ensemble of ML algorithms for the better prediction of diabetes mellitus. For this, the work utilizes the Pima Indians Diabetes (PID) database. The ensemble-based approach of weighted voting classifier employs an ensemble of three ML algorithms for providing binary classification that includes logistic regression, random forest, and extreme gradient boosting classifiers. Here, the performance of the above three ML algorithms are individually assessed, and then the weighted voting-based ensembled approach is performed by considering standard benchmark metrics such as accuracy, precision, and F1 score. And finally, the above-said performance is validated using Matthews correlation coefficient. In this way, the proposed ensemble-based weighted voting approach of diabetic classification provides a supreme performance of 92.21% classification accuracy over other individual ML algorithms used.KeywordsDiabetesGlucoseEnsemble classifierInsulinVotingMachine learning