O'Reilly logo
  • Heather Finn thinks this is interesting:

This chapter uses an algorithm called CART (Classification and Regression Trees). To build the decision tree, the algorithm first creates a root node. By considering all the observations in the table, it chooses the best variable to divide up the data. To do this, it looks at all the different variables and decides which condition (for example, "Did the user read the FAQ?") would separate the outcomes (which service the user signed up for) in a way that makes it easier to guess what the user will do.


Cover of Programming Collective Intelligence


how does it decide?