Diabetes is the most common disease, and it develops when the body stops making insulin and blood sugar levels rise. One of the main conditions causing diabetic individuals to lose their vision is diabetic retinopathy (DR). It is an eye condition that affects the retina. The early detection of DR is essential for preventing total blindness in patients.This work concentrates on a cost-effective and early diagnosis of diabetic retinopathy. The study took into account the 1151- record Messidor imaging collection for diabetic retinopathy. By using machine learning classification algorithms (Logistic Regression, K Nearest Neighbor, SVM, bagged trees) on the features that are extracted from the results of various retinal images from the image dataset, the ultimate goal of this research is to investigate whether there is an existence of diabetic retinopathy. Cross validation and hold-out validation have both been used to validate the data because it may contain outliers and noisy numbers. Principal Component Analysis-based dimensionality reduction criteria have also been used to achieve the best results. In this study, logistic regression had the highest accuracy, scoring 77.2% in cross validation and 83.8% in the case of hold-out validation. The best method for DR prediction, according to the findings, is Logistic Regression.Likewise, bagged trees have also turned up with 80% accuracy.

Diabetic Retinopathy, Machine Learning, Classification, Logistic regression, SVM, KNN, Bagged trees