Comparative study of machine learning algorithms for diabetes detection using binary data

Source abstract

The prevalence of diabetes is constantly increasing, and timely diagnosis of this disease is extremely important in the health care system. Nowadays, in addition to advanced medical technologies, there is an opportunity to process large volumes of information quickly and efficiently, in particular medical information. Patients with diabetes, even in the early stages, have certain similar symptoms and patterns, which makes it possible to effectively diagnose the disease, based on easily obtained clinical data. In this work, we compare the accuracy of different machine learning algorithms for diagnosing diabetes on binary data that do not include laboratory tests, and can be obtained by patient questionnaires. All the classification models give good results and show the feasibility of identifying individuals probably having undiagnosed diabetes, based on questionnaire data. The best results are provided by random forest classifier. In order to improve the accuracy of the classification, an analysis of the relationships between the variables of the dataset was carried out. This made it possible to reduce the dimensionality of the model by removing variables that do not carry useful information. Moreover, the data set is additionally balanced. The resulting models demonstrate higher efficiency than the classification models built on the base dataset.