Choosing ER Time Series Models (part 2): How to Fairly Compare ARIMA and XGBoost?

An emergency department forecasting example Photo by Raquel Martínez on Unsplash How does a hospital decision maker know which model to trust for forecasting? In part 1 of my series, I explored this question with an emergency department example aimed at predicting patient arrivals by first understanding the building blocks of ARIMA, a traditional time series technique for forecasting. Here I compare ARIMA with a machine learning algorithm, XGBoost, which can capture non-linear patterns, handle missing values well and has the flexibility to add features to improve predictive accuracy.