Abstract
The COVID-19 pandemic has placed immense pressure on global healthcare systems, underscoring the urgent need for early and accurate prediction of disease severity to improve patient care and optimize resource allocation. Failure in ward allocation can lead to wasted hospital resources and inadequate treatment. This study analyzes data from 806 COVID-19 patients admitted to the emergency room of Chungbuk National University Hospital, Korea, between January 2021 and December 2022, to develop machine learning models that predict which patients should be prioritized for intensive care unit (ICU) placement based on initial clinical information. Additionally, two different severity criteria were considered based on actual ICU level interventions (Criterion I) and based on national policy definitions (Criterion II). Single models of logistic regression, random forest, support vector machine, light gradient boosting, and extreme gradient boosting, as well as ensemble learning models using voting classifiers, were used. The ensemble model achieved the best performance, with recall rates of 96.2% and 88.2% for each criterion, respectively. Key features such as glucose level, neutrophil count, high sensitivity C-reactive protein (hsCRP) level, and albumin level were identified, improving model interpretability. This study provides valuable insights for healthcare professionals to support effective early ward allocation and treatment strategies.