Emergency departments (EDs) worldwide face persistent crowding and long waiting times, which are strongly associated with adverse clinical outcomes and reduced patient satisfaction. This study proposes a pre-admission decision-support system that classifies patients at the moment of registration using readily available demographic, symptomatic, and vital-sign data, before formal triage is performed. The approach leverages an interpretable decision-tree classifier designed to emphasise sensitivity, thereby reducing the risk of under-triage, while still maintaining competitive precision and balanced F1 scores. The model was trained on a dataset of 1,267 adult visits labelled under the Korean Triage and Acuity Scale (KTAS), and achieved high recall with acceptable overall accuracy. A conservative “lives saved” proxy demonstrates the model’s capacity to meaningfully prioritise urgent cases within the test cohort. Feature-importance analysis revealed that injury status, arrival mode, heart rate, respiratory rate, systolic blood pressure, age, body temperature, and pain score were the most influential predictors, aligning well with clinical expectations. The findings suggest that interpretable machine-learning tools can complement traditional triage processes, supporting operational efficiency while safeguarding clinical reasoning. This framework is also consistent with continuous-improvement initiatives in emergency care, including Lean healthcare and tele-emergency strategies.