Diabetes has become a disease that seriously endangers people’s health, then how to control the content of glycemic is an important issue. Since the treatment scheme of patient is usually a combination of multiple hypoglycemic drugs, multi-label learning is an effective method to solve this problem. By analyzing the type 2 diabetes data set including 2443 diabetics provided by the Chinese People’s Liberation Army General Hospital, we find that the defined daily dose system (DDDs) of drugs is an imbalanced problem, traditional multi-label methods easily leads to poor prediction results. In order to overcome the shortcoming, a weighted rank support vector machine (WRank-SVM) is proposed in this paper. We firstly define the weight of each label and then give each sample different weight according to relevant-irrelevant label pair. This method ensures that the prediction results on drugs with higher DDDs are as accurate as possible. Compared with the other six popular multi-label methods, our WRank-SVM can effectively predict the schemes for hypoglycemic drugs of type 2 diabetes. Meanwhile, receiver operating characteristic (ROC) curve is employed to statistically show the effectiveness of the model. Finally, the correlation between labels and features is further analyzed, and 13 important features are selected to improve the average precision of our proposed algorithm.