Diabetic Retinopathy (DR) is an eye disease mainly generated due to diabetes. It affects vision and causes blindness problems because it weakens the blood vessels in retina. The statistical evidence indicates that the persons who are all having diabetics for more than 15 years also have DR. It now presents a major hazard to people's lives and health as a result. While physical disease identification can be used to combat DR, it is both intimidating and laborious, thus a new method must be developed. Therefore, initial detection and diagnosis are crucial for DR to prevent blindness and stop the condition from emerging into more severe stages. Many researchers generated countless Machine Learning (ML) methods for the extraction of DR features for initial findings. The typical machine learning models unfortunately exposed small generalization in feature abstraction and classification when applied to smaller datasets, or they have obligatory training time, which results in poor prediction when applied to higher datasets. Thus, deep learning (DL), the newest field in machine learning, is utilized to treat lower dataset based on competent data processing techniques. The deep architectures employ larger datasets in order to boost performance in feature extraction and image arrangement process. This paper presents a comprehensive examination of DR, causes of DR, its traits, ML and DL models, challenges, similarities, and possible research directions for DR early detection.