COVFILTER: A Low-cost Portable Device for the Prediction of Covid-19 for Resource-Constrained Rural Communities

Abstract

Early identification of COVID-19 is critical for preventing death and significant illness. People living in remote parts of resource-constrained countries find it more difficult to get tested due to a lack of adequate testing. As a result, having a primary filtering tool that can assist us in simplifying bulk COVID testing to prevent community spread is vital. In this paper, we introduce CovFilter, a low-cost portable device for COVID-19 prediction for resource-constrained rural communities, with the goal of encouraging people to be tested for COVID-19 in a more informed manner. CovFilter Hardware Module collects health parameters from three sensors while the CovFilter Prediction Module predicts COVID-19 status using the health data. We train supervised learning algorithms and an artificial neural network to predict COVID-19 from vital sign readings where MultilayerPerceptron outperformed ANN, NaiveBayes, Logistic, SGD, DecisionStump, and SVM with an F1 of 93.22%. We further show that a weighted majority voting ensemble classifier can outperform all single classifiers achieving an F1 of over 94%.