DR, a major cause of vision loss in diabetics, requires early recognition for useful treatment. This work explores the effectiveness of ML (machine-learning) techniques, particularly the VGG16, and CNN (Convolutional Neural Network), for computerized DR detection and classification using retinal fundus images. We evaluate the effectiveness of the VGG16 model, which was trained on an extensive dataset to categorize the severity of DR into discrete levels. We compare the model's performance with that of alternative techniques, including KNN (K-Nearest Neighbours), SVM (Support Vector Machines), and several DenseNet designs. Based on this investigation, VGG16 can accurately classify DR data, indicating performance that is on par with current deep-learning models. We highlight our method's effectiveness in feature extraction and its strong performance at various DR degrees of rigorousness. Draw attention to the important research gaps and difficulties in applying deep learning (DL) for DR diagnosis, highlighting the significance of larger datasets and the interpretability of model predictions. For researchers and medical professionals looking to improve automated systems for early detection and treatment of depression and anxiety, this work offers insightful recommendations.