<![CDATA[The high prevalence of preventable eye diseases, such as cataracts, glaucoma, and diabetic retinopathy, emphasizes the importance of accessible and efficient diagnostic solutions. This research aims to develop a web-based eye disease classification system using a lightweight Convolutional Neural Network (CNN) architecture, MobileNetV2, to overcome computational limitations in real-time applications. CRISP-DM methodology is applied, including dataset preparation, transfer learning with MobileNetV2 and VGG16, model evaluation, and implementation using Flask. The dataset from Kaggle consisting of 4,217 eye fundus images with four classes (cataract, glaucoma, diabetic retinopathy, and normal) was divided into 80% training, 10% validation, and 10% testing. Data augmentation and normalization were performed to improve model generalization. The results showed MobileNetV2 achieved the highest accuracy (90.14%) with low computational requirements, outperforming VGG16 (89.66%) and CNN (86.78%). MobileNetV2 displays balanced precision (89-99%), recall (74-96%), and F1-score (81-99%) across all classes, especially excelling in diabetic retinopathy detection. Its efficiency on resource-constrained environments makes it ideal for web integration. The developed Flask-based application allows users to upload images for instant classification, bridging the healthcare access gap. This research proves the effectiveness of MobileNetV2 in combining high accuracy and computational efficiency, offering a scalable solution for early screening of eye diseases, especially in remote areas.]]>
Copyrights © 2025
