Comparison of Image Classification Algorithms Performance for Eye Diseases in Humans

Abstract

This research focus on a comparative analysis of the efficiency of image classification algorithms for eye diseases in humans using the Orange Data Mining program, which serves as a tool for machine learning to aid in the classification of various types of eye diseases, including Cataract, Diabetic retinopathy, Glaucoma, and normal eyes. A total of 4,217 images were used as training and testing datasets. The objective of comparing the performance and accuracy of these algorithms for classifying eye diseases. The supervised learning methods used for classifying the different types of eye disease datasets included Neural Network, Logistic Regression, Gradient Boosting, Support Vector Machine, k-Nearest Neighbor, Random Forest, AdaBoost, Naive Bayes, and Decision Trees. The experimental results showed that the algorithm with the highest performance in classifying eye diseases is the Neural Network with ReLU activation function, achieving an accuracy of 97.87%.