Abstract:

Diabetic Retinopathy (DR) is the leading cause of vision loss and blindness, affecting millions of people worldwide. There are many different scientific and medical approaches that use retinal fundus images for DR detection. In most of these approaches, various machine learning and deep learning approaches have been applied to diabetic retinopathy datasets without the feature selection step. The DR dataset obtained from the UCI machine learning repository was used in the study. In this article, Atom Search Optimization (ASO) algorithm, a new population-based metaheuristic method proposed by utilizing atom dynamics, is used for the first time as a feature selection method for the DR dataset. Applied the ASO algorithm to the normalized dataset, the new dataset was tested by six classification algorithms: Bagging, CvR, Ibk, JRip, Kstar, and SimpleCart. The same classification algorithms were applied to the original DR dataset. The results obtained were compared with the data set that was selected with the ASO algorithm. The performance of the proposed model was evaluated in terms of accuracy, sensitivity, specificity, precision, f-measure, and roc curve values. The results show that better values were obtained with Bagging, CvR, Ibk, JRip, Kstar, and SimpleCart algorithms on the dataset selected with the ASO algorithm. In this regard, an increase of 2.7% for the average accuracy, 3.5% for the sensitivity, and 2% for the specificity were achieved in the classification rates obtained without feature selection of the algorithms with the proposed feature selection.

Keywords: