The HPFOREST Procedure

No SAS procedure other than PROC HPFOREST creates a forest of decision trees for predictive modeling. PROC HPSPLIT in SAS HPSTAT creates a single decision tree, as does PROC ARBORETUM in SAS Enterprise Miner. These procedures search for a split on every variable in every node; the HPFOREST procedure searches for a split on only one variable in a node: the variable that has the largest association with the target among candidates randomly selected in that node. Consequently, the HPFOREST procedure creates different trees than the other procedures.

The ARBORETUM procedure distinguishes between split-based and observation-based variable importance measures. The HPFOREST procedure calls these measures loss reduction and Breiman’s method of variable importance, respectively.

The HPFOREST and HPSPLIT procedures are high-performance analytical procedures; the ARBORETUM procedure is not.