Is It Possible to Distinguish COVID-19 Cases and Influenza with Wearable Devices? Analysis with Machine Learning

Abstract

The COVID-19 situation is enforcing the creation of the diagnosis and supporting methods for early detection, which could serve as screening tools. In this paper, we introduced the methodologies based on wearable devices and machine learning, which distinguishes between COVID-19 disease and two types of Influenza. We checked the results of binary classification for various scenarios and multiclass classification. The results were evaluated separately for the cases before the pandemic and in the middle of the pandemic. In the middle of the pandemic, the best classification accuracy was achieved when distinguishing between COVID-19 and Influenza cases with k-NN (the balanced accuracy was equal to 73%). The highest sensitivity was achieved for Logistic Regression - 61%. The successful distinction between Influenza types was achieved in 80 % for XGBoost and Decision Tree. Additionally, the balanced accuracy for multiclass classification was equal to 69 % for k-NN.The COVID-19 situation is enforcing the creation of the diagnosis and supporting methods for early detection, which could serve as screening tools. In this paper, we introduced the methodologies based on wearable devices and machine learning, which distinguishes between COVID-19 disease and two types of Influenza. We checked the results of binary classification for various scenarios and multiclass classification. The results were evaluated separately for the cases before the pandemic and in the middle of the pandemic. In the middle of the pandemic, the best classification accuracy was achieved when distinguishing between COVID-19 and Influenza cases with k-NN (the balanced accuracy was equal to 73%). The highest sensitivity was achieved for Logistic Regression - 61%. The successful distinction between Influenza types was achieved in 80 % for XGBoost and Decision Tree. Additionally, the balanced accuracy for multiclass classification was equal to 69 % for k-NN.