This paper presents a comparative study on different machine learning algorithms to classify retinal fundus images of glaucoma, diabetic retinopathy, and healthy eyes. This study will aid the researchers to know about the reflections of different algorithms on retinal images. We attempted to perform binary classification and multi-class classification on the images acquired from various public repositories. The quality of the input images is enhanced by using contrast stretching and histogram equalization. From the enhanced images, features extraction and selection are carried out using SURF descriptor and k-means clustering, respectively. The extracted features are fed into perceptron, linear discriminant analysis (LDA), and support vector machines (SVM) for classification. A pretrained deep learning model, AlexNet is also used to classify the retinal fundus images. Among these models, SVM is trained with three different kernel functions and it does multi-class classification when it is modelled with Error Correcting Output Codes (ECOC). Comparative analysis shows that multi-class classification with ECOC-SVM has achieved high accuracy of 92%.