Comprehensive Performance Analysis based on Classical Machine Learning and Deep Learning Methods for Predicting the COVID-19 Infections

Abstract

The COVID-19 (coronavirus disease) has been declared as pandemic throughout the world by the WHO (World Health Organization). The numbers of active COVID-19 cases are increasing day by day and clinical laboratory findings consume more time while interpreting the COVID-19 infected result. There are limited treatment facilities and proper guidelines for reducing infection rates. To overcome these limitations, the requirement of clinical decision support systems embedded with prediction algorithms are raised. In our study, we have architect the clinical prediction system using classical machine learning, deep learning algorithms, and experimental laboratory data. Our model estimated that which patients likely infected with COVID-19 disease. The prediction performances of our models are evaluated based on the accuracy score. The experimental dataset has been provided by Hospital Israelita Albert Einstein at Sao Paulo, Brazil, which included the records of 600 patients from 18 laboratory findings with 10% COVID-19 disease infected patients. Our model has been validated with a train-test split approach, 10-folds approach, and AUC-ROC curve score. The experimental results show that the infected patients with COVID-19 disease are identified at an accuracy of 91.88% through deep learning method (CNN) and 89.79 % through classical machine learning (Logistic Regression), respectively. We can observe that our prediction model could be used as clinical assistance for predicting the COVID-19 infections. We would like to invite medical experts to recommend our model for clinical studies.