Diagnosis of Diabetes Using a Weight-Adjusted Voting Approach

Abstract

Diabetes is a worldwide public health challenge with a yearly increasing incidence. Many approaches using different machine learning classifiers have been developed for automatic diagnosis of diabetes. However, they mostly rely on a single classifier or a hybrid model to make the diagnosis decision, which might be weaker than a voted decision of multiple classifiers. In this study, we present an approach that combines three classifiers (i.e. Support vector machine, artificial neural network, and naive bayes) to diagnose diabetes. The approach can adjust each classifier's weight based on their ability and history of making correct predictions. A rule that mixes majority voting and weights of classifiers was proposed and applied for the final diagnosis decision. The Pima Indians diabetes data set (268 diabetes patients and 500 normal subjects) was used in the work. A wrapper method was adopted to select features for classification. An experimental comparison of our method with other voting strategies and each single classifier used in our study demonstrated that our approach performed better in sensitivity.