Scholarlink

<p>The patient’s length of stay is an essential indicator for performance measurement in hospitals. The prolonged stay results in more consumption of hospital resources. It keeps patients at risk of infection and adversely affects the admission of critically ill patients due to limited bed availability. An accurate prediction of length of stay at admission time can help the medical specialist to judiciously allocate the clinical staff, plan personnel care, and optimize the use of available hospital resources. Short hospital stays help to reduce costs and the demand for insurance. For example, an AI-based predictive system proposed in Australia claims a 6.5% reduction in length of stay and a 2.1% reduction in 30-day readmission. It projected approximately $10&#xa0;M in savings annually. The research literature provides evidence for the use of machine learning algorithms to predict length of stay more quickly. However, there is limited focus on selecting the most relevant features and on an appropriate model to handle missing values, data imbalance, irrelevant or non-correlated features, and noisy data. This leaves scope for improving the reliability and accuracy of these models. In this paper, a framework is proposed to predict patients’ length of stay using five machine learning models and to identify the parameters responsible for longer stays. In addition, hyperparameter tuning has been performed to identify optimal parameters for the data, including medical history, gender, and laboratory measurements. The random forest model, integrated with a random forest-based feature selection technique, outperformed the other models, achieving an accuracy of 91.34% and an AUC of 97.51%. The integration of feature selection led to an improvement of 3.30% in accuracy and 1.33% in AUC. It also reduces computational time. This approach may be useful to improve hospital resource utilization and patient care planning. It may assist patients and insurance companies in managing their finances within budget.</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

An Integrated Feature Selection and Machine Learning Framework for Hospital Length of Stay Prediction

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月22日

查看全文
  • Jagriti Gupta,
  • Naresh Sharma,
  • Geeta Rani

摘要

The patient’s length of stay is an essential indicator for performance measurement in hospitals. The prolonged stay results in more consumption of hospital resources. It keeps patients at risk of infection and adversely affects the admission of critically ill patients due to limited bed availability. An accurate prediction of length of stay at admission time can help the medical specialist to judiciously allocate the clinical staff, plan personnel care, and optimize the use of available hospital resources. Short hospital stays help to reduce costs and the demand for insurance. For example, an AI-based predictive system proposed in Australia claims a 6.5% reduction in length of stay and a 2.1% reduction in 30-day readmission. It projected approximately $10 M in savings annually. The research literature provides evidence for the use of machine learning algorithms to predict length of stay more quickly. However, there is limited focus on selecting the most relevant features and on an appropriate model to handle missing values, data imbalance, irrelevant or non-correlated features, and noisy data. This leaves scope for improving the reliability and accuracy of these models. In this paper, a framework is proposed to predict patients’ length of stay using five machine learning models and to identify the parameters responsible for longer stays. In addition, hyperparameter tuning has been performed to identify optimal parameters for the data, including medical history, gender, and laboratory measurements. The random forest model, integrated with a random forest-based feature selection technique, outperformed the other models, achieving an accuracy of 91.34% and an AUC of 97.51%. The integration of feature selection led to an improvement of 3.30% in accuracy and 1.33% in AUC. It also reduces computational time. This approach may be useful to improve hospital resource utilization and patient care planning. It may assist patients and insurance companies in managing their finances within budget.