Scholarlink

<p>This study aims to optimize emergency department (ED) triage processes using machine learning (ML) and deep learning (DL) models. Traditional triage systems are often based on subjective assessments of clinical staff, which could lead to inconsistencies and inefficiencies. This research aims to develop a more accurate and objective triage classification system by combining structured numerical data and unstructured textual patient reports. In this study, the data of 7,000 patients applying to the emergency department of a hospital during January–February 2024 are used. The dataset includes vital and demographic variables such as age, pulse, blood pressure, temperature, and textual data containing patient complaints. Feature scaling and natural language processing (NLP) techniques have been applied in the data preprocessing stage. Subsequently, the proposed CNN, LSTM deep learning, and machine learning models consisting of XGBoost, Random Forest, Logistic Regression, SVM, KNN, and Decision Tree are trained and tested. The performance of the proposed models is evaluated through metrics, which include accuracy, precision, sensitivity, F1-score, and AUC. XGBoost has the highest accuracy (85%) and AUC score (0.963) among the tested models. CNN and LSTM deep learning models achieved (83%) accuracy by processing textual and numerical data together. Compared to traditional rule-based triage methods, artificial intelligence-based models are more successful in prioritizing critical cases. The findings suggest that integrating ML and DL models into emergency department triage processes can improve decision-making, reduce subjectivity, and improve patient prioritization. Textual data integration significantly improves classification success.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Medical Decision Support Systems in Emergency Departments: An NLP and Machine Learning-Based Triage Model

  • 期刊论文

  • 发表日期：

    2025年12月4日

查看全文
  • Yasin Kırelli

摘要

This study aims to optimize emergency department (ED) triage processes using machine learning (ML) and deep learning (DL) models. Traditional triage systems are often based on subjective assessments of clinical staff, which could lead to inconsistencies and inefficiencies. This research aims to develop a more accurate and objective triage classification system by combining structured numerical data and unstructured textual patient reports. In this study, the data of 7,000 patients applying to the emergency department of a hospital during January–February 2024 are used. The dataset includes vital and demographic variables such as age, pulse, blood pressure, temperature, and textual data containing patient complaints. Feature scaling and natural language processing (NLP) techniques have been applied in the data preprocessing stage. Subsequently, the proposed CNN, LSTM deep learning, and machine learning models consisting of XGBoost, Random Forest, Logistic Regression, SVM, KNN, and Decision Tree are trained and tested. The performance of the proposed models is evaluated through metrics, which include accuracy, precision, sensitivity, F1-score, and AUC. XGBoost has the highest accuracy (85%) and AUC score (0.963) among the tested models. CNN and LSTM deep learning models achieved (83%) accuracy by processing textual and numerical data together. Compared to traditional rule-based triage methods, artificial intelligence-based models are more successful in prioritizing critical cases. The findings suggest that integrating ML and DL models into emergency department triage processes can improve decision-making, reduce subjectivity, and improve patient prioritization. Textual data integration significantly improves classification success.