To judge the benefit of picking a particular
attribute and condition for partitioning of the data at a node, we measure the
purity of the data at the children resulting from partitioning by that attribute. The
attribute and condition that result in the maximum purity are chosen.