Development and Evaluation of Machine Learning Models for Early Detection of Asymptomatic COVID-19 Patients Using Heart Rate and Oxygen Levels

Abstract

Since the coronavirus disease 2019 (COVID-19) spread across the world in late December 2019, it has taken significant harm and major challenges in over 190+ countries all over the world. The research that is being done today is getting all the chest X-ray images and lung images. With the help of this, researchers predicting Covid-19. There is mounting evidence that many COVID-19 patients are asymptomatic or have only minor symptoms but may still spread the virus to others. In the existing system, simple pulse oximeters were used to diagnose infectious disease early. Oxygen level and heart rate can be used to detect virus related infections, including asymptomatic patient infections. Screening for asymptomatic infections is difficult, which makes national prevention and control of the outbreak more difficult. In this research we predict the asymptomatic COVID-19 patients with the help of their oxygen level and heart rate level. To build the machine learning model we use SVM, Naïve Bayes, KNN and logistic regression algorithms on the collected dataset. The model predicts the asymptomatic COVID-19 patients early. The dataset contains 105,609 cases with 16 attributes, including information of patients with COVID-19 RT-PCR test results. There are ten key features to be selected from the given dataset for the experiment. First, we analyze the features of the dataset to find most important features. Heart rate and SPO2 are the most important features of the dataset for predicting the asymptomatic COVID-19 patients. Our machine learning technique uses four ML algorithms. Through feature correlation, we improved accuracy by using ten main features. Following that, we trained and evaluated the data with 80-20% splits. In this study compares the results of the model with other studies and find that our technique is achieve best results from others. The current study's findings show that the model developed with the KNN algorithm is more effective at detecting the likelihood of the infected patients and achieved highest 98% accuracy, 87% precision, 97% recall, 92% f1 score and making it the best model among those that have been developed with other algorithms such as support vector machine, naïve bayes and logistic regression