Hi all,
i have a customer asking whether MARS (Multivariate Adaptative regression Spline) is or will be available in JMP. Thanks a lot in advance,
Cheers,
Valerie
A form of MARS is used to fit the penalized B-spline models in the Functional Data Explorer in route to the functional principal components, but it is not generally available as a stand-alone fitting platform.
One thing to be careful about: MARS is a registered trademark (originally for Salford Systems and now for Minitab, which owns/sells Salford Predictive Modeler). That would not prevent us from implementing it as an analytical technique, but we might have to come up with a different name for it. I looked in Jira and did not see any mention of this method. @Mark_Bailey can you check to see if we ever considered doing this or if there is any JIRA/DEFECT issues that mention this?
Also, if the customer is willing, you could ask them to go the JMP Wish List and enter a wish list item about this.
To add on to what @SamGardner mentions, would be also good to understand any use-cases that the customer has where MARS is performing better than some of the other predictive modeling tools in JMP Pro. Or if this request is more of a check-list item from this customer.
In case it helps: Salford/Minitab did not have many people using this. For the few that used MARS, the main reasons are that the model is interpretable in comparison with a lot of other methods, the results are easy enough to understand if you know regression, and the model can easily be deployed in Excel. I don't remember ever seeing MARS outperform random forests or boosted trees. The main drawback of MARS is that the model's performance deteriorates with a lot of predictors. The documentation recommends using a variable reduction technique first, and suggests that MARS performs best with about 5 or less predictors.
I used the broadest term I could think of (regression spline) and found nothing in Jira at this time.
That is some good information. Thanks!
