Machine Learning-based Diabetic Retinopathy Classification using Novel Pre-processing Technique: A Comprehensive Approach

Diabetic Retinopathy (DR), a critical eye condition impacting individuals with diabetes, manifests through progressive damage to retinal blood vessels, leading to varying degrees of vision impairment. This research introduces an innovative approach to automate DR detection and grading, focusing on enhanced pre-processing techniques coupled with machine learning classification methods. Our study developed a unique three-stage pre-processing pipeline integrating colour space conversion, LAB and Adaptive Contrast Enhancement, The preprocessing workflow specifically addresses the challenges of non-uniform illumination and varying image quality in fundus photographs, resulting in standardized images suitable for automated analysis. This approach differs from conventional methods by incorporating dynamic thresholding based on local intensity distributions. Evaluated multiple ma-chine learning classifiers on a retinal fundus images collected from publically available databases. The Random Forest classifier emerged as the superior model, achieving 96% accuracy in DR grading. This performance improve-ment can be attributed to our novel pre-processing strategy and optimized feature selection method. The model demonstrated robust performance across different severity levels of DR, with particularly strong results in identifying early-stage manifestations.