IMPROVING DIABETIC RETINOPATHY RECOGNITION WITH ADVANCED DEEP ENSEMBLE MACHINE LEARNING TECHNIQUES

Abstract

Diabetic retinopathy (DR), one of the most common and crippling eye diseases, affects a large percentage of people worldwide. For blindness prevention and appropriate treatment, DR must be evaluated quickly and precisely. ResNets50, VGG19, and ConvNet CNNs are examples of modern methods that aim to improve the precision and dependability of DR classification. A range of datasets, such as the Kaggle Diabetic Retinopathy Detection and Golonoma (Positive and Negative) Datasets are used in the evaluation in order to provide a comprehensive performance analysis of the proposed models. We highlight the many objectives that the research accomplishes by addressing the most common issues in DR classification, including interpretability and picture quality fluctuations. Improvements in classification performance across various DR severity stages, noise tolerance in real-world picture quality fluctuations, Deep Ensemble model explainability and deplorability regarding clinical problems encountered in the real-world healthcare context. We proposed a research strategy in this work that combines a thorough literature evaluation with data collecting and meticulous preprocessing. Using cross-validation and hyperparameter tuning, cutting-edge ensemble models such as Random Forest, Gradient Boosting Classifier, XGB Classifier, etc., developed and adjusted. Preprocessed datasets used to train and test these models, and various metrics like accuracy, precision, recall, and F1 score used to assess them.