ABSTRACT
This chapter discusses with two main issues: getting the right sized tree T and getting more accurate estimates of the true probability of misclassification or of the true expected misclassification cost R* (T). The stepwise tree structure does an optimization at each step over a large number of possible splits of the data. If only resubstitution estimates are used, the usual results are too much splitting, trees that are much larger than the data warrant, and a resubstitution estimate R(T) that is biased downward. The first step is to grow a very large tree Tmax by letting the splitting procedure continue until all terminal nodes are either small or pure or contain only identical measurement vectors. The tree Tmax is grown using only L1 and pruned upward to give the sequence T1 > T2 > … > {t1}.
