A Deep Learning Framework for Diabetes Prediction: PIMA Indian Dataset

A growing number of people throughout the world are struggling with diabetes, which is already one of the most prevalent diseases. On the other hand, diabetes can be significantly slowed down in its progression if caught early. A novel deep learning-based approach to diabetes early detection is suggested in this paper. The PIMA dataset included in the research consists entirely of numerical values, similar to the majority of medical data sets. Thus, popular convolutional-neural-network (CNN) models have restricted applicability to this kind of data. For the purpose of early diabetes detection using the robust representation of CNN models, this study converts numerical data into visuals according to the feature importance. The diabetes picture data that is generated is subsequently classified using three distinct methods. The first uses convolutional neural network (CNN) models called ResNet18 and ResNet50 to sort through photos of people with diabetes. Second, using support vector machines (SVMs), we combine the deep features of the ResNet models for classification. The final method involves using SVM for feature classification on the chosen fusion features.