Risk prediction of chronic obstructive pulmonary disease: an explainable machine learning study

Chronic obstructive pulmonary disease (COPD) often progresses silently, making early risk identification clinically important. Using data from the National Health and Nutrition Examination Survey (NHANES), this study developed a machine learning framework for COPD risk prediction and further examined the key factors underlying model decisions. After data preprocessing, feature selection, and class balancing, multiple machine learning models were constructed and compared, together with an ensemble model. The ensemble achieved the best overall performance, while the best individual model also showed good discrimination. Feature selection and explainability analyses yielded broadly consistent results, identifying smoking exposure, age, asthma, and inflammation- and hematology-related indicators as major contributors to COPD risk prediction. Longer smoking duration and older age were particularly associated with higher predicted risk. These findings suggest that machine learning models built on multidimensional clinical and laboratory information can support COPD risk identification while providing explainable insights for early screening and risk stratification.