A Robust Machine Learning Framework for Diabetes Prediction

Abstract

Diabetes mellitus is a metabolic disorder characterized by hyperglycemia which results from the inadequacy of the body to secret and responds to insulin. If not properly managed or diagnosed on time, diabetes can pose a risk to vital body organs such as the eyes, kidneys, nerves, heart, and blood vessels and can be life-threatening. From the many years of research in computational diagnosis of diabetes, machine learning has been proven to be a viable solution for the prediction of diabetes. However, the accuracy rate to date suggests that there is still much room for improvement. In this paper, we are proposing a machine learning framework to improve the performance of diabetes prediction with the PIMA Indian dataset. Through analysis, we observe that the main challenges of the dataset, which flaws learning, are feature selection and missing values. For each of these challenges, we propose a working solution that incorporates, Spearman Correlation and polynomial regression from a new perspective. Further, we optimize the random forest classifier by tuning its hyperparameters using grid search and repeated stratified k-fold cross-validation to build a robust random forest model that scales to the prediction problem. Finally, through exhaustive experiments, we demonstrate that our proposed data preparation approaches lead to a robust machine learning framework for the diagnosis of diabetes mellitus with train accuracy, and test-accuracy values that range from 98.96% to 100% and 97.92% to 100%, respectively, which outperforms all the state-of-the-art results. The source code for the proposed machine learning framework is made publicly available.