There are differences in how triage is done in emergency rooms, and triage nurses often don’t follow through with the standard triage tools. This has led researchers to look for more accurate and reliable ways to triage so that patients are better prioritised based on their health conditions. This research aims to define optimal methodological strategies for utilising machine learning (ML) approaches in constructing an automated triage system to achieve heightened precision in assessment. Integrating IoMT in telemedicine systems is a potentially revolutionary approach to prehospital emergency triage. To this end, the study develops a framework to augment the triage processes in emergency departments using a hybrid approach. A Principal Component Analysis technique was employed to reduce the number of input features using a dataset with 50,000 patient data collected from medical sensors. Some of the tested machine learning models achieved a promising outcome: Extra Trees achieved 88% accuracy and 98% ROC AUC, Gradient Boosting 87% accuracy and 98% ROC AUC, XGBoost 90% accuracy and 99% ROC AUC, and the Stacking Classifier, where accuracy was 95% and ROC AUC 99%, was the best classifier. These metrics demonstrate the model’s superior abilities in accurately classifying the patient’s triage level compared to individual classifiers. The outcomes of the system we propose have a tremendous deal of promise for improving medical emergency prediction capabilities and providing powerful decision-making tools that can enhance patient triage, assist staff in the triage process, and ultimately reduce the effects of ED overcrowding while improving patient outcomes.