topic Re: Is there a way to constrain a model so predictions are limited within a certain response range? in Discussions

Is there a way to constrain a model so predictions are limited within a certain response range?

schilakwad — Fri, 09 Jun 2023 00:40:57 GMT

Hi,

I have a space-filling DoE with a response that ranges from 0 to 100%, is there a way to constrain the model (specifically a Gaussian Process model, if possible) so that model predictions are don't go below 0 and above 100%? I'm using JMP Pro 15, if that helps.

Thank you,

Shama

Re: Is there a way to constrain a model so predictions are limited within a certain response range?

Mark_Bailey — Tue, 26 Oct 2021 16:07:00 GMT

You might have to break this into several steps. You might apply the logit transform to your response (be sure to represent it as 0 to 1, not 0 to 100). Fit this new response, save the model, then back transform the predictions.

Re: Is there a way to constrain a model so predictions are limited within a certain response range?

Mark_Bailey — Tue, 26 Oct 2021 16:37:14 GMT

This suggestion is commonly performed in logistic regression, which is modeling the proportion (probability) of an outcome. The logit function converts the [0,1] variable to [-inf,+inf]. Here is the back transform:

The Lin[x] in this formula would be replace by your prediction formula column.

Re: Is there a way to constrain a model so predictions are limited within a certain response range?

schilakwad — Thu, 28 Oct 2021 13:23:10 GMT

That's great, thank you. I'll give that a try!

Re: Is there a way to constrain a model so predictions are limited within a certain response range?

schilakwad — Thu, 28 Oct 2021 13:23:37 GMT

Thank you, that's really helpful.