An Efficient Feature Selection in Classifying Diabetes Mellitus

Abstract

Machine learning (ML) is one of the most popular fields in medical research. Correct identification of the presence of diabetes is essential for providing efficient treatment. Numerous machine learning models were developed to predict diabetes. However, many ML models suffer from misclassification due to a lack of proper feature selection methods. How to select the best features is still a significant problem in the classification domain. To address the problem, an ensemble-based feature selection is proposed. The proposed feature selection is then evaluated in five machine learning models. The experimental results show that the proposed feature selection is 36% more efficient than existing feature selection methods.