Database Reference
In-Depth Information
Figure 3.22 Further growing of the decision tree by partitioning a parent node.
splitting the ''men'' branch of the tree into smaller subgroups and child nodes
(Figure 3.22).
At this level of the tree, occupation category is the best predictor since it
results in optimal separation. The parent branch is partitioned into two child nodes:
blue-collar and white-collar men. The node of blue-collar men presents absolute
cohesion since it only contains (100%) non-buyers. Thus one more rule has been
identified, which classifies all blue-collar men as non-buyers with a confidence of
1.0. The percentage of buyers rises at about 67% (2/3) among white-collar men
who seem to constitute the target group of the promoted service. In our fictional
example the purity of this node can be further improved and this last partitioning
is presented in Figure 3.23.
Finally, white-collar men are segmented according to their SMS usage,
resulting in two completely homogeneous terminal nodels. The percentage of
buyers reaches 100% among white-collar men with high SMS usage, inducing a
confident rule for identification of good Internet service prospects.
The above oversimplified example may be useful in clarifying the way that
decision trees work, yet it cannot be considered as realistic. On the contrary we
Search WWH ::




Custom Search