Hyperpolarized noble-gas pulmonary imaging provides a way to measure ventilation and perfusion in patients. The potential of highly sensitive MRI biomarkers of lung function has not yet been exploited using machine-learning. Ensemble machine-learning merges diverse classifiers to improve classification accuracy and reduce the potential for misclassification. Our aim was to evaluate the performance of ensemble machine-learning algorithms and hyperpolarized gas MRI features for predicting worsening airflow measured using spirometry. This proof-of-concept study revealed that MRI ventilation combined with ensemble machine-learning predicted small changes in airflow limitation (∆FEV1%pred=5%) over relatively short time period (2.5 yr) in ex-smokers with and without COPD.