An explainable machine learning approach for hospital emergency department visits forecasting using continuous...

Area of Science:

• Healthcare Operations Research
• Applied Machine Learning
• Health Informatics

Background:

• Emergency Department (ED) visits have steadily increased since the late 1990s, a trend significantly amplified by the COVID-19 pandemic.
• Accurate forecasting of ED visits is essential for enhancing hospital operational efficiency, patient care quality, and outcomes.

Purpose of the Study:

• To develop and evaluate two novel, interpretable Machine Learning (ML) approaches for accurate forecasting of hospital Emergency Department (ED) visits.
• To improve short-term and long-term prediction accuracy for ED visit volumes.

Main Methods:

• Proposed two distinct ML-based forecasting strategies: a threshold-based data segmentation approach and a cluster-based ensemble learning method.
• Evaluated methodologies using real-world ED visit data from two Spanish hospitals.

Main Results:

• Both proposed approaches demonstrated accurate ED visit forecasting capabilities for both short-term and long-term horizons (up to one week).
• The novel methods outperformed alternative prediction techniques in terms of forecasting accuracy and efficiency.

Conclusions:

• The developed forecasting models offer strong explainability, identifying key variables influencing ED visit predictions.
• The interpretable ML approaches provide valuable insights for hospital resource management and strategic planning.