Hyper-rectangular space partitioning trees: practical approach (Q2488385)

Hyper-rectangular space partitioning (HSP) classification trees are considered in which, at a given node, partitioning is made by hyper-rectangles rather then hyper-planes (as in most popular ordinary binary classification trees). An HSP tree growing algorithm is described. Advantages of HSP trees in solving some complex (for ordinary binary trees) problems are considered. It is noted that HSP trees are more stable than ordinary binary trees. HSP is compared to CART and C4.5 procedures via simulation and on real data.