Inability in diabetes mellitus diagnosis in early stages of the disease is one of the main problems for diabetic patients. Hence, in this study we aim to improve the ability of diabetes detection by applying a combination of a classification method as a basic method and k-nearest neighbor and four clustering combination of basic models. The process of weighting features were repeated five times and each step, showed a progress in accuracy of results. Because of too many null data we have used a set of averaging methods, instance removal and a similar method to k-nearest neighbor algorithm in data preparation process.
