The development of diabetes occurs due to elevated glucose levels in the bloodstream. Prevention of diabetes or the delayed onset of diabetes is crucial. It can be achieved if there exists a screening process that can accurately identify individuals who are at a higher risk of developing diabetes in the future. Although there are many works employing machine learning techniques in medical diagnostics, there is little work done regarding the long term prediction of disease, type 2 diabetes in particular. In this study, we propose a machine learning framework consists of finding the best features that are highly correlated with the future development of diabetes, followed by developing diabetes prediction models. The proposed models are evaluated using data from a longitudinal clinical study known as the San Antonio Heart Study. Our approach has managed to achieve a long-term prediction accuracy of 81.01%, a specificity of 81.2%, a sensitivity of 79.5%, and an AUC score of 87.1%.