Abstract
It is an extremely important to have an AI-based system that can assist specialties to correctly identify and diagnosis diabetic retinopathy (DR). In this study, we introduce an accurate approach for DR diagnosis using machine learning (ML) techniques and a modified golf optimization algorithm (mGOA). The mGOA optimizes ML classifiers through finding the best available parameters with respect to objective functions, hence decreases the number of features and increases the classifier’s accuracy. A fitness function is employed to minimize the feature number of the medical dataset. The obtained results showed superiority of the mGOA with higher convergence speeds without extra processing costs across the datasets compared with several competitors. Also, the mGOA attained maximum accuracy and optimally reduced the number of features in the binary and multi-class datasets achieving the best CEC’2022 benchmark results compared with other metaheuristic algorithms. Based on this findings, three optimized ML classifiers called mGOA-SVM, mGOA-radial SVM,and mGOA-kNN were introduced as tools for classification of diabetic retinopathy disease and their performance was assessed on Messidor and EyePACS1 datasets. Experimental results demonstrated that mGOA-SVM and mGOA-radial SVM achieved remarkable accuracy in classification of DR disease with an average accuracy of 98.5% and precision of 97.4%.