The Identification of Respiratory Phase Using Support Vector Machines and Extreme Gradient Boosting

Respiration is often monitored by direct airflow measurement. Indirect approaches of monitoring breathing can be used when direct airflow access is inconvenient or impossible. Respiration tends to affect the seismocardiographic signal (SCG). Hence, SCG signals may be used to derive the respiratory phases noninvasively. In portable settings where SCG signals are readily available, the use of SCG would eliminate the need for direct airflow measurements. Previous studies [1]–[6] extracted respiratory phases from different physiological signals using traditional signal processing and machine learning techniques/methods. Solar et.al. [7] classified SCG events using support vector machine (SVM) algorithm, but only lung volume classes (i.e., Low and high lung volumes) were considered. The current study investigated both lung volume and inspiratory/expiratory phases and compared the performance of two methods (namely, SVM and extreme Gradient Boosting (XGBoost)).