Chapter 9. Tree-based regression

The linear regression methods we looked at in chapter 8 contain some powerful methods. These methods create a model that needs to work for all of the data points (locally weighted linear regression is the exception). When the data has many features that interact in complicated ways, building a global model can be difficult if not foolish. We know there are many nonlinearities in real life. How can we expect to model everything with a global linear model?

One way to build a model for our data is to subdivide the data into sections that can be modeled easily. These partitions can then be modeled with linear regression techniques from chapter 8. If we first partition the data and the results don’t fit a linear model, then we can partition the partitions. Trees and recursion are useful tools for this sort of portioning.

9.1. Locally modeling complex data

9.2. Building trees with continuous and discrete features

9.3. Using CART for regression

9.4. Tree pruning

9.5. Model trees

9.6. Example: comparing tree methods to standard regression

9.7. Using Tkinter to create a GUI in Python

9.8. Summary