Balancing Data through Data Augmentation Improves the Generality of Transfer Learning for Diabetic Retinopathy Classification

The incidence of diabetes in Mauritius is amongst the highest world. Diabetic retinopathy (DR), a complication resulting from disease, can lead to blindness if not detected early. aim this work was investigate use transfer learning and data augmentation for classification fundus images into five different stages diabetic retinopathy. are No DR, Mild nonproliferative Moderate Severe DR Proliferative. To end, deep three pre-trained models, VGG16, ResNet50 DenseNet169, were used classify APTOS dataset. preliminary experiments resulted low training validation accuracies, hence, dataset augmented while ensuring balance between classes. This then train best models blind Mauritian test datum. We found that model produced results out also achieved very good accuracies class-4 images, severe cases, some unexpected results, with being classified as mild, therefore needs be further investigated.