Predicting patient risk of leaving without being seen using machine learning: a retrospective study in a single overcrowded emergency department

Emergency department (ED) overcrowding has become a critical issue in hospital management, leading to increased patient wait times and higher rates of individuals leaving without being seen (LWBS). This study aims to identify key factors influencing LWBS rates and to develop a predictive model using machine learning (ML) techniques. A retrospective analysis was conducted on 80,614 ED visits recorded at Maresca Hospital in Torre del Greco, Italy, between 2019 and 2023. Statistical analyses were performed to examine correlations between patient characteristics, operational variables, and LWBS occurrences. Four ML classification algorithms—Random Forest, Naïve Bayes, Decision Tree, and Logistic Regression—were evaluated for their predictive capabilities. Random Forest demonstrated the highest performance on the minority class, achieving an overall accuracy of 72%. Feature importance analysis highlighted waiting time, triage score, and access mode as significant predictors. These findings suggest that predictive modeling may support hospital resource planning and patient flow management strategies to reduce LWBS rates.