Comparison of Machine Learning Techniques Based on Feature Reduction to Detect Diabetic Retinopathy

Abstract

Diabetic retinopathy is considered as an eye disorder condition that can affect persons who are diagnosed positive against diabetes. This condition causes damage to the retina's blood vessels. Diabetic retinopathy detection at an early stage may help patients protect their vision. This disease is often asymptomatic in the early stages but has a significant impact in the latter stages. The timely identification and intervention of diabetic retinopathy (DR) can greatly mitigate the potential for visual impairment. It takes time, effort, and money for ophthalmologists to manually diagnose DR retina fundus images. So there is a need for a screening system that can detect the disease in its early stages. There are various machine learning-based approaches available to detect diabetic retinopathy. In our proposed work first, we have to select the optimal features then we perform the comparative analysis with the help of various machine learning algorithms. We have used Random Forest, Logistic Regression, Gradient Boosting, Support Vector Machine (SVM), Naïve Bayes, and AdaBoost. In our experiment, we observed that Logistic Regression achieved the highest accuracy of 74.2 with optimal features.