Decision Tree Builder


Read more about Data Workbench’s End-of-life announcement.

Decision trees are a predictive analytics visualization used to evaluate visitor characteristics and relationships. The Decision Tree Builder generates a decision tree visualization based on a specified positive case and a set of inputs.

A Decision Tree is a binary classifier with a set of rules (or filters) identifying visitors who satisfy specific rules based on a positive case. A decision tree sets rules to classify visitors who satisfy (or do not satisfy) this positive case. These rules generate a tree map to provide a level of confidence to meet these positive case results.

A Decision Tree is built by examining inputs at each level and choosing the one that provides a maximum gain of information at a specified split point. Split points for each variable-level generates two sets:

  • Values less than or equal to the split point, and
  • Values greater than the split point.

Use decision trees to

  • Perform meaningful analysis and interpretation in less time.
  • Employ automated segment generation.
  • Quickly make inferences from a model based on a large amount of data.

Toolbar and Menus

The toolbar includes buttons and menu commands for the Decision Tree, including features to set the Positive Case and add Input Listings.

Like other visualizations, the Element box lets you drag and drop Dimension and Elements, although you can also drag directly from the Finders pane.

For additional information, see Decision Tree Options.

Input Listing

This area displays the inputs into the tree model. They are color coded to match nodes in the Tree Display area.

Right-clicking on an input allows you to remove the input from the model and reset.

If you hover over a tree node, it will display the split conditions along the branch to that node and the prediction at that node with its confidence value.

Tree Display

This area displays the tree model with leaf nodes color-coded based on its prediction: green for a True prediction of the Positive Case, and red for a False prediction.

The split nodes are color coded to the inputs that match their selection condition. Hovering over a node displays information about the split and expands the inputs listing to display the split points along the branch and the distribution of the training set.

Nodes below a threshold are not displayed by default. Click on an expandable node (indicated by a + symbol) to explore a branch. Click on the root node to return to the full tree display.

On this page