You’ve now seen a few different automatic classifiers, and this chapter will expand on them by introducing a very useful method called decision tree learning. Unlike most other classifiers, the models produced by decision trees are easy to interpret—the list of numbers in a Bayesian classifier will tell you how important each word is, but you really have to do the calculation to know what the outcome will be. A neural network is even more difficult to interpret, since the weight of the connection between two neurons has very little meaning on its own. You can understand the reasoning process of a decision tree just by looking at it, and you can even convert it to a simple series of if-then statements.
This chapter will cover three different examples that employ decision trees. The first shows how to predict which of a site’s users are likely to pay for premium access. Many online applications that are priced by subscription or on a per-use basis offer users a way to try the applications before spending money. In the case of subscriptions, the sites usually offer a time-limited free trial or a feature-limited free version. Sites that employ per-use pricing may offer a free session or similar incentive.
The other examples, covered later in the chapter, will use decision trees to model housing prices and “hotness.”
Sometimes when a high-traffic site links to a new application that offers free accounts and subscription accounts, ...