Abstarct: Prevention, early detection, and automated classification of eye diseases are critical and challenging topics in the medical field. Given the importance of this subject, numerous studies have been conducted using deep learning techniques to achieve accurate and reliable results. Although deep learning, due to its ability to automatically extract features and the flexibility of various architectures, has made significant advancements in this field, challenges such as the need for large amounts of data, heavy computational load, and optimal hyperparameter tuning still exist, requiring a significant amount of time and expertise. In contrast, machine learning methods, due to their higher speed and lower computational load, are a suitable choice for some applications. In this research, six different transfer learning models were used to extract features from eye images. For final classification, several ensemble learning algorithms such as Random Forest, Bagging, XGBoost, and hard and soft voting methods, as well as Stacking, were utilized. Although some combinations and models have led to improved diagnostic accuracy compared to previous methods, this improvement has not been observed in all cases, and some models have shown lower performance. In the proposed method, the DenseNet121 model for OCT images achieved an accuracy of 99.87%, and for fundus images, an accuracy of 96.44%. Additionally, in the Stacking method for fundus images, the DenseNet121 model achieved an accuracy of 94.66%, and for OCT images, the VGG19 model achieved an accuracy of 98.41%.