In the prior chapter, we went through a discussion on MARS, how it works, why use it, and so on, so I won't duplicate that here; other than that, it can be applied in a classification problem as a generalized linear model. One of the key benefits is its power to conduct feature selection, so there's no need to run stepwise or IV—or even regularization, for that matter.
We'll train it with 5-fold cross-validation and set nprune = 15 to limit the maximum number of features at 15. Recall from the previous chapter that more than 15 terms are possible as it fits piecewise splines.
This code will give us our model object. Be advised that this may take some time to complete:
> set.seed(1972)> earth_fit <- ...
