Paper Title
Transparent Diagnosis for Diabetic Retinopathy

Abstract
Diabetic Retinopathy (DR) is a sight-threatening complication of diabetes that demands early and accurate diagnosis. Manual screening methods are often time-consuming, expertise-dependent, and subject to inter-observer variability. To address this, we propose a transparent, explainable deep learning-based system for automated DR diagnosis using retinal fundus images. The system combines EfficientNet and MobileNet feature extractors with an ensemble Random Forest classifier. Enhanced preprocessing, including Gaussian filtering and HSV transformations, is employed alongside robust data augmentation to overcome dataset imbalance. Explainable AI techniques like Grad-CAM are integrated to visualize and justify predictions. The model is deployed through a Flask-powered web application, enabling real-time, interpretable DR diagnosis. This solution aims to bridge the gap between AI efficacy and clinical trust. Keywords - Diabetic Retinopathy, Deep Learning, Explainable AI, EfficientNet, MobileNet, Grad-CAM, Fundus Images, Ensemble Learning, Computer Vision, Medical Imaging.