Data science appositeness in diabetes mellitus diagnosis for healthcare systems of developing nations

Abstract

One of the capacious applications of data science could be its use in bioinformatics. With its proper implementation, chronic diseases like diabetes mellitus, responsible for millions of deaths worldwide, could be diagnosed and predicted with high efficacy. But if not attended, could lead to fatal issues such as kidney failures, heart diseases, and even limb amputation. Diabetic cases have only elevated in numbers in the recent past. The authors use various machine learning, deep learning, and data dimensionality reduction techniques to detect diabetes mellitus. The research is principally conducted on two datasets, first from the Frankfurt Hospital, Germany, second from the University of California, Irvine repository. Models such as support vector machines, Naïve Bayes, and Random Forests were implemented to classify diabetic patients from non-diabetic ones. Subsequently, after hyperparameter tuning, a comparative study on the results was done and the most prominent model was promoted. This process was repeated for the datasets with reduced dimensionality using linear discriminant analysis and principal component analysis. For the Frankfurt, Germany, dataset, K-nearest neighbours showed the best accuracy of 98.2%, and the Random Forest classifier for the University of California, Irvine, repository showed 99.2%. With such proficiency, the authors thereby propose a statistical approach for the prediction of diabetes in its early stages. They hope to counter the concern of undiagnosed diabetic cases in developing nations where there is a lack of a basic healthcare system.