Background Pulmonary Function Testing (PFT) is widely accepted as the gold standard for diagnosing Chronic Obstructive Pulmonary Disease (COPD). However, its use for large-scale population screening is limited due to the need for specialized equipment and personnel to operate. We developed novel algorithms using supervised machine learning to assess VF (Ventilatory Function) and screen for COPD based on signals detected by smartwatch sensors.Methods Two batches of participants were recruited: 476 for the training dataset, and 118 for the validation dataset. Smartwatch-recorded cough sounds(CS) and physiological parameters (PP) were collected after spirometry as inputs, and the desired outputs included spirometry results after bronchodilator administration, as well as COPD diagnosis by two independent respirology physicians. The prediction of VF indicators such as FEV1/FVC (Forced Expiratory Volume in One Second/Forced Vital Volume) and FVC using CS was evaluated against spirometry results. Subsequently,the accuracy of the model that incorporated PP to predict the participants' COPD status was verified against the physicians' diagnosis.Results Using CS alone, our model had a Mean Absolute Error (MAE) of 7.4% and 10.6% for FEV1/FVC and FVC% prediction, respectively, compared to spirometry. A significant correlation was found between the predicted FVC and measured FVC (r=0.806, p<0.001), as well as predicted FEV1/FVC and measured FEV1/FVC (r=0.763, p<0.001). Combined with PP, our model had an overall accuracy, sensitivity, and specificity of 88.0%, 91.5%, and 84.5%, respectively, in distinguishing COPD and normal controls.Conclusion Our algorithm showed efficacy in predicting VF and screening COPD.