Refining the Model

Figure 1. Tree Diagram for Credit Rating Model

Overall, the model has a correct classification rate of just under 80%. This is reflected in most of the terminal nodes, where the predicted category—the highlighted category in the node—is the same as the actual category for 80% or more of the cases.

There is, however, one terminal node where cases are fairly evenly split between good and bad credit ratings. In node 9, the predicted credit rating is "good," but only 56% of the cases in that node actually have a good credit rating. That means that almost half of the cases in that node (44%) will have the wrong predicted category. And if the primary concern is identifying bad credit risks, this node doesn't perform very well.