The Boston Housing data includes information from the 1970 Census for the city of Boston and surrounding area. Note that there are two variables representing value; one is “Medv” which is a dollar value. The other is “Cat Medv” which is a binary variable indicating whether the house is of “high value” (signified by “1”) or “lower value” (identified by “0”).

Estimate a classification tree using the Cat Medv variable as the target. You are trying to classify home as either high value or lower value by using the given attributes (such as the number of rooms in the house, the crime rate in the local area, and the age of the houses in the area). Use all of the attributes in the file to estimate the model requesting the “best pruned tree” to prevent overfitting. For display, however, request the “full tree.” 

Evaluate the estimate for the best pruned tree using the confusion matrix, the misclassification rate, and, most importantly, the lift chart.

By examining the full tree you should be able to see how the CART algorithm will attempt to perfectly classify the records. In doing so it may overfit the data and that is the reason for using the pruning method.

Now re-estimate the algorithm using the target Medv this time. In order to do so you will have to use the “Prediction” menu in XLMiner and select “Regression Tree. “The method is similar to the classification tree estimated earlier, but now an actual numerical prediction is being requested. Overfitting remains a possible problem and so it will again be necessary to prune using either the best pruned tree or the minimum error tree.