A Movement-Artefact-Free Heart-Rate Prediction System

Continuous automatic heart rate (HR) monitoring plays a crucial role in timely intervention for postoperative patients. However, for effective alarm management, patients' activities of daily living need to be considered as they influence HR. This explorative study aimed to develop a heartrate prediction system while performing six activities. An experiment with fourteen participants was conducted to gather data to build a system. This system consisted of a support-vector machine classifier for activity recognition and a k-Nearest Neighbors regressor for HR prediction. The R-squared (a goodness-of-fit measure) of the HR predictor is 79% on average. Given the heterogeneity of different populations, the system will be further tested and developed using patient datasets in future towards clinicalpractice applications.