Diabetes is one of the prevalent diseases in the word with a high mortality rate. This disease has created several health problems and side effects on other organs of the human body. Therefore, diagnosis of this disease at early stage is essential that can reduce the fatal rate of humans. There are several ways to diagnose the diabetes but early diagnosis is quite challenging task for the medical practitioners. Recently, data mining based techniques are widely used for early prediction of diabetes that gives promising results in diabetes prediction. This paper presents the detailed review of existing data mining techniques used for diabetes prediction with their comparative study. This study also provides analysis of existing methodologies that will help in future perspective for designing and developing novel diabetes predictive models.