Graphics Reference
In-Depth Information
Note that the color scheme chosen for the nodes ranges from a very dark
blue to a very light green—creating a conundrum for labeling. Reading text
depends on the contrast between the text and the background. One solution
is to adjust the range of colors for the nodes so as not to be so dark. But this
is not feasible here because the colors should be the same across all three
different visualizations in order to be comparable. Instead, a white outline
is added around the text to clearly separate it from dark nodes. When using
text with outlines, use a heavy font such as bold or black; otherwise, thin
font details can disappear.
Note that the leaf nodes are now much smaller and more difficult to read. If
detail is required, interactive zooming is needed (for example, PDF output
can be easily zoomed) or printed out to poster size.
Decision Trees
Decision trees
predict an outcome by dividing data by successive criteria,
forming a hierarchy. The game “Twenty Questions” works like a decision
tree. With each successive question, the player attempts to eliminate a
significant portion of possible answers. Each decision point is a branch in
the tree. The best strategy is to ask questions that quickly narrow down the
possible answers.
The approach is often used in database marketing, where data attributes
such as age, income, gender, and employment are used to make marketing
decisions such as which credit card offer to mail out to a prospective
customer. Perhaps you've been on a phone call with a cable service provider
to address some concern and been offered an Internet upgrade, followed
by an offer for additional channels. A decision tree has been created and is
being used to prompt the call center operator to make successive offers. If
done well, the caller is receiving offers that are meaningful and relevant.
Decisiontreescanbecreatedoutofawiderangeofdatawhereasequenceof
decisions can be derived either directly from a given observed sequence or
generatedalgorithmically(forexample,instatisticalsoftwaresolutionssuch
as R or as libraries for general programming languages such as the package
DecisionTree for Python).
One interesting example of the former comes from professional sports.
Players make decisions during gameplay, and a sequence of decisions can
then be analyzed to see if there is some commonality. Offensive strategies
