Classification Trees Results - Tree Plot Tab

Select the Tree plot tab of the Classification Trees Results dialog box to access the options described here.

Classification tree plot. Click the Classification tree plot button to display a graph of the classification tree selected in the analysis. Note that all labels and legends for the graph are produced as custom text, and can be edited, moved, or deleted via the Graph Options dialog box. Use the options under Tree graph options (see below) to control which labels, legends, and node plot type are produced for the graph.

Tree graph options. Under Tree graph options are format choices for the resulting tree plots, including a group of options for the type of node plot (if any). See below for details on each of these options.

Angled branches. Select the Angled branches check box to produce a tree plot with angled branches. Clear the Angled branches check box to produce a plot with straight branches.

Node numbers. Select the Node numbers check box to produce a tree plot with node numbers as labels for each node. Clear the Node numbers check box to produce a plot without node number labels.

If the tree plot is not specified to have labels for Node assignments (see below) and the No node plot option button is selected in the Node plot type group box, the node number labels will be positioned at the center of the node. If the tree plot is specified to have labels for Node assignments and the No node plot option button is selected, the node number labels will be positioned at the upper middle part of the node. If you selected Deviation plots or Histograms as the Node plot type, the node number labels will be positioned at the upper left part of the node.

Split conditions. Select the Split conditions check box to produce a tree plot with text descriptions of the split condition for each non-terminal node. Clear the Split conditions check box to produce the graph without descriptions of the split conditions.

The descriptions of the split conditions for non-terminal nodes will be printed below the nodes at which the corresponding splits are made. For univariate splits on ordered predictor variables, the description gives the split variable's name, the less than or equal to sign, and the negative of the split constant for the split function. For univariate splits for categorical predictor variables, the description gives the split variable's name, the equal sign, and the category or categories on the predictor variable that satisfy the split. For linear combination splits on ordered predictor variables, the description gives F(0) as an abbreviation for the linear function of the predictor variables, the less than or equal to sign, and the negative of the split constant for the split function. For all types of splits, the description of the split condition indicates that all cases or objects at the split node that satisfy the split condition are sent to the left child node; all cases or objects at the split node that do not satisfy the split condition are sent to the right child node.  

Node assignments. Select the Node assignments check box to produce a tree plot with node assignment labels indicating the predicted class for cases or objects assigned to each node. Clear the Node assignments check box to produce the graph without node assignment labels.

If the tree plot is not specified to have labels for Node numbers and the No node plot option button is selected, the node assignment labels will be positioned at the center of the node. If the tree plot is specified to have labels for Node numbers and the No node plot option is selected, the node assignment labels will be positioned at the lower middle part of the node. If Deviation plots or Histograms have been selected as the Node plot type, the node assignment labels will be positioned at the upper right part of the node.

Node n's. Select the Node n's check box to produce a tree plot with custom text for each branch of the tree specifying the number of cases or objects sent down that branch. Clear the Node n's check box to produce the plot without descriptions of the number of cases sent down each branch to a node.

The number of cases or objects sent to each node is printed above the branch leading to the node. If the branch is the left branch from a split node to its left child node, the number of cases sent down the branch is the number of cases assigned to the split node that satisfy the split condition. If the branch is the right branch from a split node to its right child node, the number of cases sent down the branch is the number of cases assigned to the split node that do not satisfy the split condition.

Node plot type. Use the options under Node plot type to specify which node plot type is produced on the tree plot. See Node numbers and Node assignments (above) for descriptions of how the positioning of node number and node assignment labels is affected by the Node plot type.

Deviation plots. Select the Deviation plots option button to produce a tree plot in which each node will contain a deviation plot. The deviation node plot type is similar to the histogram node plot type, in that the observed class Ns for each node are represented by vertical bars; however, the bars connect the observed class Ns to a baseline defined as the average observed class N for the node. The bars accordingly will extend either up or down, depending on the direction of the deviations of the observed class Ns from the average class N for the node. In the ideal situation where all of the cases or objects sent to a node are perfectly classified, the bar for the predicted class for the node would extend upward and the bars for the remaining classes would extend downward, giving a quick visual indication of good classification.

A custom text legend, giving the correspondence between observed classes and the line colors and line types used to draw each bar in the node plot, is produced in the upper left corner of the tree plot. This custom text can easily be moved, edited, or deleted, as desired, to improve the appearance of the graph. This may be required when there is a large number of classes for the dependent variable, resulting in a large legend.

Histograms. Select the Histograms option button to produce a tree plot in which each node will contain a plot similar to a 2D histogram. In each plot, the observed class Ns for each node are represented by the height of vertical bars. In the ideal situation where all of the cases or objects sent to a node are perfectly classified, there would be a tall bar for the predicted class for the node and the bars for the remaining classes would be short, giving a quick visual indication of good classification.

A custom text legend, giving the correspondence between observed classes and the line colors and line types used to draw each bar in the node plot, will be produced in the upper left corner of the tree plot. This custom text can easily be moved, edited, or deleted, as desired, to improve the appearance of the graph. This may be required when there is a large number of classes for the dependent variable, resulting in a large legend.

No node plot. Select the No node plot option button to produce a tree plot without node plots.

Reset to defaults. Click the Reset to defaults button to reset all tree plot options to their default settings, which is set for all custom text label and legend options and Deviation plots for the Node plot type.