Diabetes was a disease that occurs due to high blood-glucose levels. Researchers tried to prevent complications from developing by using data mining techniques. One of the techniques used in data mining was classification. The purpose of this study improves the accuracy of the classification of diabetes for a better and optimal result. The method in this study is Correlation Feature Selection (CFS) as attribute selection, Random Over Sampling to handle unbalanced data and AdaBoost to improve the performance of the J48 algorithm so the result obtained best. Based on the result of this study, showed that Correlation Feature Selection for attribute selection and Random Over Sampling to handle imbalance's class with the Adaboost-based J48 algorithm proved can increase the results of the diabetes classification with an accuracy of 92.3%. For the further research recommended to apply other methods so that the accuracy results obtained are more optimal for comparison.