Machine Learning to Predict COVID-19 and ICU Requirement

This paper focuses on the application of machine learning (ML) algorithms to manage novel coronavirus disease (COVID-19). For this, different ML classifiers are used for two cases, one for the prediction of COVID-19 patients, and another for the prediction of the intensive care unit (ICU) requirement. A dataset of 5644 samples and 111 attributes collected at Hospital Israelita Albert Einstein, Brazil is considered in this paper. After necessary preprocessing 57 attributes are used for COVID-19 detection, while 67 attributes are considered for ICU requirement prediction. Using scikit-learn library of Python programming language, the most important features for both cases are found out. A number of base as well as ensemble classifiers are applied to the resultant datasets for the two cases. Results show that COVID-19 detection can be predicted with an accuracy of 94.39% and recall of 92% using stacking ensemble with random forest (RF), XGBoost (XGB) and logistic regression (LR). Results also show that ICU requirement can be predicted with an accuracy of 98.13% and recall of 99% using stacking ensemble with RF, extra trees and LR.