Databases Reference
In-Depth Information
Figure 5.3
Tree Graph - Iris data
Let's look at the second objective of our evaluation - understanding input
contribution.
Drag the decision tree model up to an available display and select “Tree
Graph”.
In the tree (see Figure 5.3) all nodes are sized and color encoded according to
the contents of the node. The top bar in the tree represents the entire dataset. All
varieties are equally represented given that the red, blue, and green bars are of
equal length. In processing, the decision tree classifier chose PetalLength as the
most discriminating attribute in making the first split. The split cut-off was set at
2.18 centimeters. The second level in the tree depicts this split. All Setosa
observations went into the left node, while the right node contains both the
Versicolor and Virginica. Look back at Figure 2.14 to understand the rationale
for the chosen split. The next split based on PetalWidth did not generate any
totally homogeneous nodes and thus required additional splitting.
