ED Forecast Utility functions overview

This repo, which can be used for educational purposes, contains several functions which are used by the CHART team to explore forecast models for our emergency department volumes. While the repo primarily contains functions we have used to forecast ED volumes, we have included a couple of functions for extracting weather data along with a calendar dataset we use for holidays. To understand how we calculate ED volumes, you can see [this tutorial]

Getting started

To use any of the functions, you can clone the repository and install the necessary libraries with the renv package

Loading the project for the first time:

• renv::init() + select Restore the project from the lockfile

Functions included

• initialize_models for initializing the models we typically explore
• get_features for extracting the variables we use in our models
• pick_fourier_terms for exploring different fourier terms in ARIMA models
• fit_gam for fitting our generalized additive model
• forecast_models the function we have used for extracting forecasts and 95% confidence intervals
• get_raw_weather_forecast for extracting weather forecasts
• clean_raw_forecast for cleaning raw weather forecasts
• extract_sport_events for extracting sport event times (relies on an API (stattleshipR))

Data included

• data/calendar_variables.csv: a dataset of all holidays
• data/future_frame.csv: a dataset with horizons we will forecast in the future. This is for illustrative purposes. While we can't include our actual data in the repo, this is the format that our data takes with 1 row per 6 hour interval.