Triage Categorization in Emergency Departments Using Ensemble Learning: Health Care

Abstract

Emergency department (ED) triage is a critical process that determines patient treatment priority based on medical urgency. While EDs worldwide struggle with overcrowding and resource constraints, these challenges are particularly acute in Sri Lanka, where inconsistent manual triage decisions and limited healthcare infrastructure can compromise patient care. This study develops an advanced triage prediction system using ensemble learning techniques to address these challenges. Through expert consultation, we identified key clinical indicators including vital signs, heart rate, blood pressure, respiratory rate, and oxygen saturation. Our approach implements a Stacking Classifier that combines Logistic Regression, Support Vector Machine (SVM), and Light Gradient Boosting Machine (LightGBM) algorithms, with a tuned LightGBM model serving as the final estimator. This ensemble method achieved 90.88% accuracy in predicting triage categories, offering a robust solution for optimizing patient prioritization and resource allocation in emergency settings.