<![CDATA[This research aims to evaluate the performance of the Multi-Layer Perceptron (MLP) for classifying eye diseases from fundus images in the ODIR dataset, which comprises four classes: Normal, Diabetic, Glaucoma, and Cataract. The methodology includes feature extraction using GLCM and Gabor, data pre-processing through cleaning, augmentation, and undersampling, and testing 16 model scenarios with variations in the number of hidden layers (2 and 3) and neuron configurations. The results show that data balance and dataset size are the most influential factors affecting model performance, with the best results achieved through the combination of undersampling and augmentation. The optimal architecture was obtained with the 64–32-neuron configuration, yielding a mean accuracy of 73.06%. Overall, this study concludes that combining a balanced dataset with a proportional MLP architecture significantly improves the model’s ability to classify eye diseases from fundus images.]]>
Copyrights © 2026
