Short-term forecasting of emergency medical services demand exploring machine learning

This study addresses the challenge of forecasting short-term demand in Emergency Medical Services (EMS) using machine learning techniques, which is essential for improving resource allocation, optimizing response times, and enhancing overall system efficiency. In emergency situations, the swift allo...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Computers & industrial engineering Jg. 200; S. 110765
Hauptverfasser: Shahidian, Nika, Abreu, Paulo, Santos, Daniel, Barbosa-Povoa, Ana
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Elsevier Ltd 01.02.2025
Schlagworte:
Boosting algorithms
Computational complexity analysis
Genetic algorithms
Machine learning
Short-term forecasting
Computational complexity analysis
Short-term forecasting
Machine learning
Boosting algorithms
Genetic algorithms
ISSN:0360-8352
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:This study addresses the challenge of forecasting short-term demand in Emergency Medical Services (EMS) using machine learning techniques, which is essential for improving resource allocation, optimizing response times, and enhancing overall system efficiency. In emergency situations, the swift allocation of medical personnel and vehicles is crucial for ensuring faster response times and increasing survival rates. To tackle this, the study explores nine machine learning algorithms, including Gaussian Mixture Models, Naive Bayes, K-Nearest-Neighbors, Support Vector Machine, Random Forest, Extremely Randomized Trees, Gradient Boosting, Adaptive Boosting, and Bagging, to provide accurate forecasts. Model performance is analyzed using two variable selection methods: filter and wrapper methods. The best-performing algorithms, Gradient Boosting and AdaBoost, undergo further analysis after hyperparameter fine-tuning. Principal results show that Gradient Boosting and AdaBoost outperformed other algorithms, with Random Search and Genetic Algorithm methods enhancing model performance. The study also demonstrates how the combination of different methods influences both performance and computational complexity. In this context, the developed models not only achieve high levels of accuracy in predicting short-term EMS demand but also serve as versatile tools. They distinguish priority calls and vehicle types when predicting call volume per hour from different districts and the expected number of dispatches at each base per shift, respectively. These models can be used as stand-alone tools or integrated with other optimization approaches, providing valuable inputs for optimizing staff scheduling, location, and relocation of emergency vehicles, thus contributing to the efficiency and performance improvement of EMS systems. •EMS demand forecasting approached as a machine learning classification problem.•Gradient Boosting and AdaBoost were identified as the most accurate algorithms.•Hyperparameter tuning performed using Random Search and Genetic Algorithm.•Filter and wrapper methods applied for variable selection.
ISSN:0360-8352
DOI:10.1016/j.cie.2024.110765