Optimizing Diabetic Retinopathy Screening with Machine Learning: A Data-Driven Approach

Diabetic Retinopathy (DR) ranks among the major causes of preventable blindness in regions with an inadequate specialist-based care. A two-stage approach combining deep learning for DR grading from fundus images and machine learning for forecasting diabetes from clinic data is submitted in this work. A ResNet152 Convolutional Neural Network (CNN) with fine-tuning classifies retinal images into five levels of DR severity, and an XGBoost model predicts the risk of diabetes based on input features like glucose, insulin, BMI, and age. Preprocessing involves resizing, normalization, fundus image data augmentation, and tabular data feature scaling. Transfer learning and fine-tuning improve accuracy even with a relatively small combined dataset of retinal images and clinical health records. The fusion of DR detection and diabetes screening facilitates an early warning system that can reduce the need for ophthalmologists and provide cost-efficient, effective specialist services. The automated process helps mitigate the impact of the rising prevalence of diabetes globally, especially in underprivileged areas, and allows early medical treatment before permanent damage to the retina. Experimental findings indicate that the integrated method attains a high accuracy rate of 98% for classification.