Chronic Obstructive Pulmonary Disease (COPD) is one of the leading causes of chronic diseases and deaths worldwide. When acute exacerbation of COPD (AECOPD) occurs, the frequency and severity of malignant attacks are highly correlated with the mortality rate. The purpose of this study is to use wearable devices to collect the physiological parameters of patients for early warning and prevention of complications of possible AECOPD attacks in the future. The subjects used wearable devices to measure Heart Rate Variability (HRV) at home. Physiological data and health assessment scales of 13 COPD patients were collected during the 6-month study period. According to the scale responses, the severity of the condition was classified into mild AE and no AE. If the subject needed emergency medical treatment due to COPD, it was classified as AE. With the scale classification method, a machine-learning Random Forest (RF) algorithm is used to predict the occurrence of AECOPD in the next 7 days, so as to prevent the deterioration of the disease in advance. The results of the study show that the accuracy of the model is more than 92% according to different classification methods, and using the mixed-parameter model as a feature for the prediction can improve the sensitivity of the original warning mechanism. In order to provide predictive results to the nursing staff at any time, the user interface of our system would transmit a warning message to remind the nursing staff to ensure early medical intervention for patients to avoid the occurrence of AECOPD.