Diabetic Retinopathy Classification Using Vision Transformers and XGBoost Optimized with Hippopotamus and Blue Whale Algorithms

• Published in: Journal of information systems engineering & management Vol. 10; no. 40s; pp. 458 - 476
• Main Author: R. Sindhujha
• Format: Journal Article
• Language: English
• Published: 26.04.2025
• ISSN: 2468-4376; 2468-4376
• DOI: 10.52783/jisem.v10i40s.7316

Diabetic retinopathy is a significant public health concern and one of the leading causes of blindness globally, particularly among individuals with diabetes. With the increasing prevalence of diabetes, early detection and timely intervention for diabetic retinopathy have become critical for preventing vision loss. Traditional screening methods often rely on manual inspection of retinal images, which can be time-consuming and subject to human error. To address these challenges, this project focuses on developing an advanced classification system to identify diabetic retinopathy using a hybrid model that combines the Vision Transformer (ViT) architecture with XGBoost, implemented in MATLAB. The study employs robust pre-processing techniques, including conversion to HSV color space and advanced classification methods, to enhance image quality, which is critical for accurate feature extraction. The hybrid model leverages the strengths of ViT in capturing intricate patterns in the images while utilizing XGBoost for efficient classification. Hyper parameter tuning is performed using optimization algorithms such as the Hippopotamus Algorithm and the Blue Whale Algorithm, which significantly improve the model's classification performance. The proposed system achieves an impressive accuracy of 99.5%, showcasing its potential as an effective tool for early diabetic retinopathy classification which is 13.33 % higher when compared to the traditional methods like GCN, Convolutional Auto Encoder and GA-SSAE. The findings indicate that the integration of advanced deep learning architectures and optimization techniques can lead to significant improvements in classification tasks. Furthermore, this project highlights the importance of utilizing adaptive methodologies to enhance the robustness of machine learning models. Ultimately, this research contributes to the development of efficient diagnostic tools that could enhance patient outcomes through timely detection and treatment of diabetic retinopathy.