Wearable Analytics and Early Diagnostic of COVID-19 Based on Two Cohorts

Abstract

The outbreak of the COVID-19 pandemic forced a need to create screening tests to diagnose the disease. To answer this challenge, this paper introduces the support methodology for COVID-19 early detection based on wearable and machine learning likewise on two various cohorts. We compare the level of detection of the COVID-19 disease, Influenza, and Healthy Control (HC) thanks to the usage of machine learning classifiers likewise changes in heart rate and daily activity. The features obtained as the parameters of the ratio of heart rate to the<br/>variable of the number of steps proved to have the highest statistical importance. The COVID-19 cases versus HC were possible to be distinguished with 0.73 accuracy by the XGBoost algorithm, whereas COVID-19 cases, Influenza vs. HC were able to be differentiated on similar level of accuracy: in 0.72 by Support Vector Machine. The multiclass classification between the cases achieved a 0.57 F1-score for three classes by XGBoost. For early diagnosis, this solution could serve as an extra test for clinicians during the pandemic, and the result shows which metric could be useful for creating the machine learning model.