Diabetes is a disease which is affecting many people now-a-days. Most of research is happening in this area. In this paper, we proposed a model to solve the problems in existing system in applying data mining techniques namely clustering and classifications which are applied to diagnose the type of diabetes and its severity level for every patient from the data collected. This paper tries to diagnose diabetes based on the 650 patient’s data with which we analyzed and identified severity of the diabetes. As part of procedure Simple k-means algorithm is used for clustering the entire dataset into 3 clusters i.e., cluster-0 - for gestational diabetes, cluster-1 for type-1 diabetes (juvenile diabetes), cluster-2 for type-2 diabetes. This clustered dataset was given as input to the classification model which further classifies each patient’s risk levels of diabetes as mild, moderate and severe. Further, performance analysis of different algorithms has been done on this data to diagnose diabetes. The achieved results show the performance of each classification algorithm.