Diabetes complication prediction using deep learning-based analytics

Abstract

The high levels of blood sugar (or glucose) that occur in diabetes can damage organs such as the heart, blood vessels, eyes, kidneys, and nerves in time. Type 2 diabetes typically affects adults and is most prevalent in adults due to an insufficient supply of insulin. On the other hand, Diabetes type 1, also known as juvenile diabetes or insulin-dependent diabetes, is a chronic disease in which the body cannot produce insulin on its own. Diabetes prevalence has increased over the past three decades at every income level. Affordable treatment is vital for those with diabetes. Several cost-effective interventions can improve patient outcomes. However, a diagnosis of this disease can be costly and difficult. The aim of this research is, therefore, to demonstrate a comparative analysis and improved performance using deep learning to classify diabetic and non-diabetic patients that will provide a feasible way to diagnose this chronic disease. In this work, we used a neural network model with very low variance applying the synthetic minority oversampling technique to augment and improve the variety of data. By removing imbalances and classifying diabetes based on different features, our model achieved an accuracy of approximately 99 % for training and 98 % for validation.