Diabetic Retinopathy Detection Using Ensemble Learning Technique

Abstract

Diabetic Retinopathy, which mostly affects those with long-term diabetes is among the major causes of eye blindness. To effectively manage and prevent vision loss, early diagnosis of it is essential. However, traditional diagnostic methods are often time-consuming and require specialized expertise, posing challenges for large-scale screening. This work offers a hybrid machine learning framework that combines K-Nearest Neighbours, and Random Forest in an ensemble learning technique to overcome these drawbacks. The framework uses advanced feature extraction to find important retinal characteristics including vessel diameters and exudates after utilizing improved preprocessing approaches, such as adaptive median filtering and HSV colour space conversion, to improve image quality. Majority voting is used for classification in order to maximise robustness and accuracy. The ensemble model outperforms the standalone RF and KNN models in experiments, obtaining 92% accuracy, 93 % precision, and 96 % recall. These results demonstrate how the suggested approach could transform DR detection and aid in clinical decision-making.