Development of machine learning techniques for wearable vital signs monitoring device

Abstract

High blood pressure or hypertension accounts for 45% of deaths due to heart disease and 51% of deaths due to stroke. Singapore National Health Survey estimates 27.3% of Singaporeans between the ages of 30 and 69 years, suffer from hypertension. A single wearable vital signs monitoring device is the need of the hour for home based non-invasive and low cost which can detect pulse rate and respiration rate with intelligent and autonomous decision. Currently, contact vital sign devices are incorporated into our electronic devices such as Apple watch, Garmin and other brands of smart watches. However, the availability of a non-invasive, noncontact vital sign device will allow greater ease of monitoring one’s vital sign. In this project, the author aims to test the range and accuracy of a non-contact vital signs monitoring device, the AWR1642BOOST. This was done by taking readings from different angles and distances then comparing its accuracy to that of an existing contact vital sign device, the Careech HR device. A few machine learning classifiers were also applied to the data. The author’s results have shown that the non-contact device is as accurate as the contact vital sign device at a range of 0.3 to 0.8 metres and from an angle of 0˚ to 40˚.