Diabetic retinopathy (DR) is a common eye condition that can lead to blurred vision or even blindness. Therefore, early detection is crucial to prevent or mitigate its effects. However, the subtlety of early symptoms makes it challenging for doctors to identify the condition. To address this, numerous predictive models utilizing machine learning (ML) and deep learning (DL) have been developed to assess the presence or absence of DR. Existing classification models exhibit varying levels of accuracy, influenced by the preprocessing and processing techniques applied to DR data. This research proposes computationally efficient methods for DR classification that leverage preprocessing models to prepare the data, reduce training time through feature selection, and enhance clarity by applying clustering after feature selection. This integrated approach offers a robust solution for accurate DR grading. Specifically, we utilize rough set theory (RS) and stability‐correlation and correlation (ScC) to eliminate irrelevant features and apply K‐means clustering on the results to manage indiscernibility relations among data points. We introduce two innovative solutions, FSBC‐RS and FSBC‐ScC, developed to mitigate overfitting while ensuring accurate early detection of DR grading. Both approaches exhibit state‐of‐the‐art performance in DL, achieving accuracies of 98.8% and 99.1% on the Messidor dataset, respectively, and 98.6% and 95.3% on the IDRiD dataset.
Al-Shalabi et al. (2026) studied this question.