A Novel MCDM-Based Framework to Recommend Machine Learning Techniques for Diabetes Prediction

Abstract

Early detection of diabetes is crucial because of its incurable nature. Several diabetes prediction models have been developed using machine learning techniques (MLTs). The performance of MLTs varies for different accuracy measures. Thus, selecting appropriate MLTs for diabetes prediction is challenging. This paper proposes a multi-criteria decision-making (MCDM) based framework for evaluating MLTs applied to diabetes prediction. Initially, three MCDM methods—WSM, TOPSIS, and VIKOR—are used to determine the individual ranks of MLTs for diabetes prediction performance by using various comparable performance measures (PMs). Next, a fusion approach is used to determine the final rank of the MLTs. The proposed method is validated by assessing the performance of 10 MLTs on the Pima Indian diabetes dataset using eight evaluation metrics for diabetes prediction. Based on the final MCDM rankings, logistic regression is recommended for diabetes prediction modeling.