In this study, it is aimed to generate improved models for detecting undiagnosed type II diabetes mellitus patients by employing data from both undiagnosed and diagnosed patients. The main motivation is that, with the widespread use of electronic health records, increasing amount of data is accumulating for diagnosed patients. In our simulations, the training sets are allowed to include data from both undiagnosed and diagnosed patients. The models generated are then evaluated using the undiagnosed patients. Experimental results have shown that the data from diagnosed patients do not help to improve the detection performance of undiagnosed patients.