A forecasting tool for SWAST. Implemented in Python.

The tool predicts the number of calls that result in the dispatch of one or more ambulances.

Forecasting is an ensemble of Regression with ARIMA Errors and FBProphet. The regression model and prophet explicitly model new year's day as a holiday.

All dependencies are contained in the binder/environment.yml. It is recommended that the user first install conda (via Anaconda). Once installed open a terminal (or Anaconda Prompt on windows), navigate to the swast-forecast-tool directory and run the following command to create the swast_forecast virtual environment.

• conda env create -f binder/environment.yml

To activate the virtual environment run the following command:

• conda activate swast_forecast

The virtual environment includes Jupyter-Lab. To load it run the following command (or open from Anaconda Navigator):

• jupyter-lab

The repository contains a notebook tutorial.ipynb. This contains detailed instructions for training a model and forecasting.

In the example below assume that y_train is a pandas.Series with a DateTimeIndex and time series of historical observations.

from swast_forecast.utility import forecast

#return a 7-step forecast with 95% prediction interval
results = forecast(y_train, horizon=7)

from swast_forecast.utility import default_ensemble()

#create an ensemble with default parameters
model = default_ensemble()

#fit the model to the training data
model.fit(y_train)

#return a 7-step forecast with 95% prediction interval
model.predict(horizon=7)

Output:

By default three columns of data are returned in a pandas.DataFrame:

• yhat: the mean of the forecast distribution (the point forecast)
• yhat_lower_95: the lower bound of the 95% prediction interval
• yhat_upper_95: the upper bound of the 95% prediction interval

The index of the dataframe is a pandas.DataTimeIndex