Given numerous efforts of genome analysis and disease prediction studies on Chronic obstructive pulmonary disease (COPD), however, there was no model for predicting COPD severity. In this study, we constructed the prediction model for COPD severity using various machine learning techniques. By analyzing 36S samples of mild and severe COPD groups, we observed that the model using random forest performed the best (AUC =0.886) and Diffusing capacity of Lung CO, modified medical research council, and age were the most important features of the model. These results would provide valuable scientific evidence for predicting COPD severity.