12 groups of experiments (T1-T12) were carried out on the unbalanced datasets. The H-SMOTE outperforms SMOTE on many indicators, e.g. precision, recall, F1 score, G-mean, AUC. The distribution of dataset oversampled by H-SMOTE is more stable and uniform. The center offset of H-SMOTE is smaller, which means the center of datasets before and after oversampling is more stable.