In the case study that follows, we're going to look at the application of some exciting methods on an interesting dataset. Like in the previous chapter, once the data is loaded we'll treat it, but unlike the previous example, we'll split it into training and testing sets. Given the dimensionality of the data, feature reduction and selection are critical.

We'll explore the oft-maligned stepwise selection, then move on to one of my favorite methodologies, which is Multivariate Adaptive Regression Splines (MARS). If you're not using MARS, I highly recommend it. I've been told, but cannot verify it, that Max Kuhn stated in a conference that it's his starting procedure. I'm not surprised if it's true. I learned the technique from a former Senior Director of Analytics at one of the largest banks in the world and haven't looked back since.