On the Intensive Care Unit Admission During the COVID-19 Pandemic in the Region of Lleida, Spain: A Machine Learning Study

The coronavirus disease 2019 (COVID-19) is an infectious and high transmissible disease that may cause severe illness. Some severe cases might require intensive care unit (ICU) admission because, in some cases, patients had comorbidities and several previous symptoms that can worst their condition. Therefore, it is of paramount importance to find strategies for helping manage the occupancy of ICU units. In this paper, we present a predictive model about ICU admission. We trained different machine learning (ML) models using a dataset containing previous symptoms and risk factors, such as demographics. The algorithms used were Random Forest (RF), Logistic Regression (LR), and Extreme Learning Machine (ELM), and the metrics used to evaluate them were accuracy, balanced accuracy, sensitivity, and specificity. We used RandomOverSampling (ROS), NearMiss (NM), and SMOTE algorithms to balance the dataset at different proportions. The best RF obtained model got a specificity of around 97%. The best LR model gives specificity about 93% and, the best ELM obtained a specificity of 94%. These results demonstrate the excellent performance of these algorithms in these kinds of datasets. Moreover, our findings show that ROS and SMOTE performed better than NM.