Deep Learning based Approach for Prediction of Diabetes

Diabetes is a metabolic condition that raises when the body’s glucose levels are too high and it affects people of all age groups. Therefore, early diagnosis and an appropriate treatment are essential to minimize the consequences. If it is not encountered as a serious issue, it may result in a variety of complications, including kidney problems, damage to the eyes, high blood pressure, heart problems, and affect other body parts depending on the immune system of the individual person. Every year, more people die from diabetes, making it necessary to create a system that can accurately diagnose the disease in individuals. Consequently, an effective medical decision system based on a deep neural network was used in this work for the prediction of diabetes. For training and testing, data from diabetic and non-diabetic patients were collected from the PIMA Indian database and MESSIDOR dataset. This method uses parameters such diastolic blood pressure, body mass index (BMI), triceps skin thickness, heredity, and age factor from PIMA dataset and retinal images from MESSIDOR dataset to achieve a high level of accuracy. Instead of the conventional phase of manually extracting features at various resolutions, the deep learning model extracts features and classifies them within the same network of data categorization. The results of this approach offer better categorization outcomes than the standard methods for a more optimal diabetes diagnosis.