Predicting daily emergency department visits using machine learning could increase accuracy.

Area of Science:

• Health Informatics
• Data Science
• Emergency Medicine

Background:

• Predicting emergency department (ED) visit volume is crucial for efficient healthcare resource allocation.
• Accurate forecasting enables better staffing, operating room scheduling, and resource management.

Purpose of the Study:

• To evaluate machine learning models for predicting daily ED visits.
• To compare the accuracy of machine learning against univariate time series models.
• To identify key factors influencing ED patient volumes.

Main Methods:

• Univariate time series models (ARIMA, Exponential Smoothing, Prophet) were used as a baseline.
• Machine learning models (Random Forests, Gradient Boosted Machines) were trained on 2017-2018 data.
• Model performance was assessed using out-of-sample data from 2019 and Root Mean Squared Error (RMSE).

Main Results:

• Random Forest and GBM models showed higher predictive accuracy than univariate models.
• Day of the week was the most significant predictor of ED patient volume.
• Weather variables like temperature and pressure also showed some correlation with visit trends.

Conclusions:

• Machine learning models offer a slight improvement over simple time series models for ED visit prediction.
• Further research with more data and feature engineering could enhance predictive accuracy.
• Day of the week is a critical factor for forecasting ED volumes.