Implementation of Efficientnet-B0 CNN Architecture for Classification of Eye Diseases Based on Fundus Image: Normal, Cataract, Diabetic Retinopathy, and Glaucoma

Eye is one of the most vital organs for humans. Humans can obtain 80% of information solely through vision. Loss of vision has various causes that require comprehensive prevention, treatment, and care, making early identification and diagnosis of eye diseases crucial. In detecting eye diseases, experts and doctors use various methods, one of which is fundoscopy. This research proposes the development of a model for classifying eye diseases using the CNN method with EfficientNet-B0 Architecture. The model is trained using a dataset containing 4217 fundus eye images divided into 4 classes: Normal, Cataract, Diabetic Retinopathy, and Glaucoma. Subsequently, the dataset is split into 70% training data, 15% test data, and 15% validation data. The performance results obtained include an accuracy of 0.86 or 86%, precision of 0.88 or 88%, recall of 0.87 or 87%, and an f1-score of 0.87 or 87%.