Automatic Screening of Diabetic Retinopathy Using Different Data Mining Classifier Techniques

Diabetic retinopathy (DR) is a progressive disease that occurs in patients suffering from diabetes, which ultimately leads to blindness. The presence of biomarkers namely microaneurysms, exudates, and hemorrhages in the retina attributes to this disease. Screening of retinal images could be a helpful strategy in early diagnosis of DR. Literature suggests that early diagnosis could prove vital for detection and treatment to avoid blindness in diabetic patients. Computer-aided diagnosis and data mining techniques are very promising in the identification of DR. Here, in the present study, we used six different classifiers such as k-nearest neighbor (K-NN), random forest, regression tree, support vector machine (SVM), logistic regression and the Naïve Bayes theorem to evaluate the occurrence of DR from the dataset. Based on the random forest classification, our results revealed the significance of different microaneurysms and exudates and the impact of each attribute in the diagnosis of DR. Among all the classifications studied here, remarkably the Naïve Bayes method of classification shows the highest accuracy in the diagnosis of DR at 99.8%. We speculate this approach could serve as an early diagnostic tool for ophthalmologists.