Diabetic detection at early stage could lead to avoiding multiple critical diseases. To this end, technology plays critical role. Data mining is one of the systems that could be used for the detection of disease. Data mining requires dataset for operation. Real time information may not be usable under this situation. Dataset formation is a leading step in formation of the model for detection of diabetic. The proposed system uses kmeans clustering along with pre-processing mechanism to ensure high classification accuracy. the proposed approach works in phases. In first phase data collection is done. This phase is followed by pre-processing mechanism. This mechanism removes noise from the dataset. After this clustering mechanism is applied to determine the label for presented record. The result obtained is in the range of 80% that is better by 4% then existing k means clustering mechanism.