A dataset of 1,249 uveitis cases with unknown etiology was used, with 766 (61.3%) receiving an etiological diagnosis. The study employed the ULISSE screening protocol and SUN classification for uveitis. A neural network model, Multilayer Perceptron (MLP), was developed to predict uveitis etiologies, outperforming SVM and RF in accuracy. The study highlighted the risks of oversampling methods like SMOTE, which altered the dataset's real distribution, affecting diagnostic accuracy.