Diabetic retinopathy (DR) is a leading cause of blindness among diabetic patients worldwide. Early and accurate detection of DR is critical to preventing vision loss. In today’s medicine field/computing field, many research are undergoing to prevent and detect DR at an early stage. With the global prevalence of diabetes on the rise, research in the field of computer vision and image processing for DR detection has become increasingly critical. This survey reviews state-of-the-art machine learning (ML) and transfer learning (TL) techniques applied to DR detection and severity grading. We analyze various convolutional neural networks (CNNs) and hybrid models like EfficientNet-B3, ResNet50, and ensemble methods that have shown significant improvement in DR classification accuracy. Additionally, this work explores dataset utilization, such as the APTOS 2019 and EyePACS datasets, to enhance the detection and grading of DR. Key gaps in existing research are identified, including the need for improved performance in the early stages of DR detection. Our findings highlight the transformative potential of ML and TL in automating DR diagnosis, ultimately reducing the burden on healthcare professionals and improving patient outcomes. Continued research is essential to refine these technologies for clinical integration.