Predicting Patient Triage at the Emergency Department using Machine Learning Classification: The Case Study of UNS Hospital, Indonesia

Abstract

The triage of patients in the Emergency Department (ED) plays a critical role in determining the urgency and type of treatment that shall be administered. Therefore, an accurate prediction system for patient triage can be very helpful. This study aims to develop a Machine Learning (ML) classification model to predict triage decisions for patients admitted to the ED at UNS Hospital. Several classification models were evaluated, including Naïve Bayes (NB), Decision Trees (DT), Random Forest (RF), Support Vector Machines (SVM), and k-Nearest Neighbors (k-NN). To assess model performance, multiple metrics were employed: accuracy, precision, recall, F1-score, and the confusion matrix. Following an initial evaluation, hyperparameter tuning was conducted on the selected best-performing model to identify the optimal combination of parameters for improved predictive performance. RF emerged as the best-performing model for triage prediction. The parameters tuned included the number of estimators, criterion, maximum features, and maximum depth, using a 5-fold cross-validation strategy. The optimal parameter values were found to be 900 estimators, 'entropy' for the criterion, 'log2' for max features, and a maximum depth of 40. The results of this study demonstrate that hyperparameter tuning can significantly enhance model performance, reducing recall errors, improving the F1-score, and decreasing the number of mispredictions.