Machine Learning Approach for Predicting COVID-19 Suspect Using Non-contact Vital Signs Monitoring System by RGB Camera

ABSTRACT

The emergence of several SARS-CoV-2 variants, especially, the new variant strains B.1.1.7 lineage and 20C/501Y.V2 have highly accelerated the COVID-19 pandemic. A large number of COVID-19 patients are not getting the chance of admitting to the hospitals. Therefore, this pandemic situation accelerates the method of non-contact evaluation of patients along with prediction system of COVID-19 suspects. To feed this interest, a non-contact vital signs monitoring and COVID-19 suspect prediction system is developed. The study can measure heart rate, respiratory rate, and blood oxygen saturation simultaneously using low-cost RGB camera. A predictive model is developed using supervised learning algorithms for predicting the COVID suspect. Among the implemented algorithms, Support Vector Machine ensures a high accuracy of 97.92%. The accuracy of the non-contact vital signs monitoring system is also compared with commercial sensors. Among seven participants, the absolute error (AE) of heart rate was 2.11 for two participants and AE ≤ 4.06 for the other five participants. AE for oxygen saturation was 0.0 for four participants and AE ≤ 1.0 for three participants. Commercially, mean bias for heart rate varies from 2.08 to 8.06 and mean bias for SpO2 is ± 2. The mean bias of the heart rate for this research varies from 2.11 to 4.06 and for SpO2 varies from 0 to 1. Both of them are in a commercially acceptable range. © 2022, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.