Analysis of machine learning classifiers for predicting diabetes mellitus in the preliminary stage

Abstract

Diabetes is the most common disease all over the world and it must be detected early to receive proper treatment, which can prevent the condition from becoming more severe. Automated detection plays an essential role in diabetes early diagnosis. Over the last few decades, many complicated machine learning algorithms and data analysis approaches have been applied for diabetes prediction. To determine the best model for early-stage diabetes prediction, ten different machine learning classifiers have been used in this study. These models were evaluated in terms of accuracy, precision, specificity, recall, F1-score, negative predictive value (NPV), false positive rate (FPR), rate of misclassification, and receiver operating characteristics (ROC) curve. The experimental findings indicated that all of the models performed well. Gradient boosting (GB), with 97.2% accuracy, is observed to show the best performance on the early-stage diabetes risk prediction dataset. Random forest (RF) and Adaboost performed similarly to the GB; however, RF and Adaboost's precision was not as good as the GB precision (GB’s).