Early Triage Prediction for Outpatient Care Based on Heterogeneous Medical Data Utilizing Machine Learning

Abstract

Traditional triage tools hospitals use face limitations in handling the increasing number of patients and analyzing complex data. These ongoing challenges in patient triage necessitate the development of more effective prediction methods. This study aims to use machine learning (ML) to create an automated triage model for remote patients in telemedicine systems, providing more accurate health services and health assessments of urgent cases in real time. A comparative study was conducted to ascertain how well different supervised machine learning models, like SVM, RF, DT, LR, NB, and KNN, evaluated patient triage outcomes for outpatient care. Hence, data from diverse, rapidly generated sources is crucial for informed patient triage decisions. Collected through IoMT-enabled sensors, it includes sensory data (ECG, blood pressure, SpO2, temperature) and non-sensory text frame measurements. The study examined six supervised machine learning algorithms. These models were trained using patient medical data and validated by assessing their performance. Supervised ML technology was implemented in Hadoop and Spark environments to identify individuals with chronic illnesses accurately. A dataset of 55,680 patient records was used to evaluate methods and determine the best match for disease prediction. The simulation results highlight the powerful integration of ML in telemedicine to analyze data from heterogeneous IoMT devices, indicating that the Decision Tree (DT) algorithm outperformed the other five machine learning algorithms by 93.50% in terms of performance and accuracy metrics. This result provides practical insights for developing automated triage models in telemedicine systems.