<![CDATA[Eye diseases such as cataracts, diabetic retinopathy, and glaucoma are among the leading causes of visual impairment and can result in blindness if not detected at an early stage. The application of Deep Learning technology, particularly Convolutional Neural Networks (CNNs), provides an opportunity to support the automatic classification of eye diseases based on retinal fundus images. This study aims to develop an eye diseaseclassification model using the EfficientNetB0 architecture with a transfer learning approach. The dataset used was the Eye Diseases Classification dataset from Kaggle, consisting of 4,217 retinal fundus images categorized into four classes: Cataract, Diabetic Retinopathy, Glaucoma, and Normal. Before the training process, the images underwent preprocessing, including image resizing, normalization, and data augmentation. The model was trained using the Tensor Flow/Keras framework in the Google Colab environment and evaluated using accuracy, precision, recall, F1-score, and confusion matrix metrics. The experimental results showed that the proposed model achieved an overall accuracy of 73.23%. The Cataract class demonstrated the best performance, with a precision of 0.78, a recall of 0.92, and an F1-score of 0.84. In contrast, the Glaucoma class obtained the lowest recall of 0.36, indicating that the model still encountered difficulties in correctly identifying a considerable number of glaucoma images. Overall, the results indicate that the EfficientNetB0 architecture is capable of classifying eye diseases from retinal fundus images with satisfactory performance. However, increasing the dataset size, optimizing preprocessing techniques, and applying more extensive fine-tuning are recommended to further improve the model's performance, particularly in glaucoma classification]]>
Copyrights © 2026
