incorporating extra predictors in a forecasting model

Question

I have a Machine Learning demand forecasting task for a healthcare service. I have some columns such as date, diagnosis, hospital etc and each row represents a particular time that this service was used. Now I want to make weekly forecasting models for the usage of this service, so I group-by the date column and get the counts for each week, creating this way a 'demand' column. Now I have only a DateTime and demand columns. How I could incorporate my other predictors in a forecasting regression model. Maybe I am missing something as I am new to time-series.

Answer 1

One standard method would be regression with ARIMA errors. Take a look at "The ARIMAX model muddle", which has examples in R.

Actually, depending on your predictors, you might not have any temporal dynamics left over after the regression part, so the errors might be modeled as white noise. This can happen in particular if you explicitly model seasonality.

Note that you will need to forecast the values of your predictors themselves in some way (unless you only use lagged values with lags at least as large as your forecasting horizon).