Diabetes is a prevalent condition that is associated with a significant global mortality rate, diabetes affects millions of individuals globally; India has the second highest diabetes patients in the world. The recent survey says 77 million people suffering from diabetes (Type-2) and nearly 25 million are prediabetes. Because of the busy schedule individuals won’t visit the diagnostic center to test the diabetes, many people who have diabetes or may have diabetes don’t know about their health problems as well as risk factors until they are diagnosed. So that it is important to identify this disease at an early stage, which can reduce mortality in people. In recent years’ artificial intelligence methods have been widely used to predict diabetes earlier on, and the outcomes have been quite good. Many authors trying to address this problem based on patient’s standard UCI dataset, and applied many Machine Learning and Deep Learning techniques used to prediction. This paper reviews existing Machine Learning and Deep Learning techniques used for early diabetes prediction by using the standard dataset of UCI early diabetes risk prediction dataset, comparative study of these techniques is also conducted, and the study also provides an analysis of existing methodologies to inform prospects for the design and development of new diabetes prediction models. Most of the Authors used standard UCI dataset and analyzed one or more feature selection methods. But no one author tried to apply all feature selection methods. In this review came to know that need to select more feature selection methods and apply ML, DL techniques to resolve this issue.