I have a customer that is predicting staffing needs at several locations of a health care provider in response to COVID19. Numbers of infections in their area are dropping (happy dance!) and as a result the model is predicting negative numbers of nurses when projecting out six months. They would like to set a floor (e.g. min. number of nurses), is there a way to do this or is there an alternative workaround to this? I have not been able to come up with anything. Thanks.
@DonMcCormack the standard Time Series platform (Analyze
EDIT:
**** I did not read your question carefully. Please disregard the info below, as it does not apply to the question you asked. I will leave it, though, in case anyone reading this is interested in an option if the predicted # of new patients is negative.
Apologies,
Brady
************************************************************
************************************************************
Another option is to have them predict total cases, instead of additional cases. (Dataset with script attached.)
Here is a chart with the typical logistic growth, and an upper asymptote (which is typically unknown) of 1e6. (i.e., the "truth" is that a max of 1 million people will be infected.)
The "total" model makes predictions above the asymptote (chiefly because only data occurring AFTER maximum infection rate was included in the model), but these don't appear nonsensical, as in practice, the asymptote is unknown.
The "new cases" model, as you've mentioned, makes predictions below 0... which IS nonsensical.
Note that in each case, the confidence bounds DO include the truth... but we are forecasting a huge number of points relative to the number of points in the training data--a good way to produce poor forecasts.
Cheers,
Brady
