SPSS Statistics for Data Analysis and Visualization

Chapter 14 Powerful and Intuitive: IBM SPSS Decision Trees

Now that we’ve seen Artificial Neural Nets, we are going to move on to another technique. Decision trees are more accurately thought of as a class of techniques as they represent multiple algorithms. The chapter that lays the groundwork for what we will see in this chapter, and all of Part III, is Chapter 11. If you are new to data mining, in general, you may want to start there. IBM SPSS Decision Trees offers four “Growing Methods”: CHAID, Exhaustive CHAID, CRT, and QUEST. The C5.0 Tree extension command offers a fifth possible option. Extension commands will be discussed in Chapter 18. We will demonstrate just CHAID and CRT, but running more than one iteration of each. CHAID and CRT provide a number of contrasts to each other so those two will give a good understanding of the decision tree approach. By altering the settings of both CHAID and CRT, it will allow the differences to become even more clear. A deeper understanding of two will prove a more satisfying introduction than a brief introduction of all five. (Note that Exhaustive CHAID, as the name implies, is quite similar to CHAID.) Finally, at the close of the chapter we will demonstrate the Scoring Wizard.

Building a Tree with the CHAID Algorithm

We’ll use the Titanic_Results.sav dataset (available in this chapter’s downloads), and the same partition variable, Train_Test, that was created near the end of Chapter 13. As shown in Figure 14.1, Pclass, Age, Sex, ...