One of the primary drivers of visual impairment in working-age grown-ups is diabetic retinopathy. For a good prognosis, early diagnosis of this illness is essential. By receiving therapy as soon as possible, many diabetic retinopathy complications can be avoided. The detection of diabetic retinopathy is quite challenging because of the variety and complexity of the condition. The goal of this paper is to detect diabetic retinopathy using Convolutional Neural Networks (CNNs) in conjunction with a transfer learning approach and hyperparameter tuning. To detect diabetic retinopathy, four different models are trained: Alex Net, Vgg Net, Google Net, and Res Net. This paper brings together a performance analysis of four different networks. The APTOS Blindness Dataset obtained from Kaggle was used for training and testing. The best classification accuracy of 95.21% was obtained with VGG Net model, thus demonstrating the better accuracy compared to the other of Convolutional Neural Networks (CNN’s) coupled with transfer learning for Diabetic Retinopathy detection.

Diabetic Retinopathy, Transfer Learning, Convolutional Neural Networks, Hyperparameter Tuning