MCBM: IMPLEMENTATION OF MULTICLASS AND TRANSFER LEARNING ALGORITHM BASED ON DEEP LEARNING MODEL FOR EARLY DETECTION OF DIABETIC RETINOPATHY

Abstract

Diabetic retinopathy (DR), the primary cause of a visible disease in working-age adults, is often controlled with the aid of early detection to prevent sight loss. This research proposes a collection of automated deep-learning techniques for DR screening. In this paper, we collected total 3662 Images from the Kaggle. Out of the total 3662 images, 90% (3295) images taken for the training purpose and 10% (367) for the testing purpose. This study measured the performance and comparative study of five Deep Learning models such as CNN, Efficient Net, VGG 16, Mobile Net, and RESNET 50 are demonstrated to improve the accuracy by changing various parameters of these models to classify DR in different stages. Out of the total images. Our finding shows that Efficient Net achieved a training accuracy of 0.9342 and a testing accuracy is 0.8181 and RESNET 50 achieved 0.9329 accuracies for the train data set and the test data set with 0.8116 accuracies. Efficient Net and Res Net 50 have achieved better accuracy out of the five models. Hence these two models perform well as compared to the other 3 Models.