How to Use the Decision Tree Visualization Tool

PredictionWorks has created a Decision Tree Visualizer, named VisualTree, in order for you to understand which customer segments will and will not respond to a campaign. Are they your best customers, your weekday shoppers, or your bargain hunters? To ensure maximum accuracy each Decision Tree has been selected based on exhaustive cross-validation testing.

Navigational Window: Click and Drag within the smaller window in the lower right hand corner (Navigational Window) to view an area within the main window. By clicking and dragging with the mouse, a box will be drawn. The size of the drawn box will determine the displayed area and the level of zoom.

Node Results:

• Line 1: Breakpoint - Variable where decision point is defined
• Line 2: Number of Records within the node
• Line 3: Benefit Per Customer - Determines the profit per customer based on the Cost-Benefit Matrix if the selected node is selected for the campaign.
• Line 4: Total Benefit - The total profit if the selected node is used for the campaign.
• Success Results and Colored-Bar: Shows the percentage of the population that falls into each bin. The colored-bar shows the graphical breakdown of the population.

Leaf-node Shadow and Branch-thickness: Each node's shadow is a 3-D representation of the number of customers that fell within that node. The larger the shadow, the larger the population of that node. Similarly, the thickness of each branch represents the number of customers that met the criteria necessary to fall into that path. The thicker the branch, the larger the population passing through that branch.

Mouseovers: The Results of any leaf-node can be seen by hovering the pointer over any selected leaf-node. This is especially useful when fully zoomed out and when getting an overall sense of the model results.

Selecting Nodes: Simply click any node to select the node and all of its respective child nodes. Hold down the Ctrl key to select multiple nodes. De-selecting Nodes: Simply click the node again or select a new node. Export to the Server - To create a generated list of the population within the selected nodes, simple select the proper nodes and click the Export button. This population can then be used to create a marketing campaign.

Hints for Interpreting the Decision Tree:

• Pay more attention to the choices made nearest to the parent node of the tree. These are the most critical to decide a customer’s segment membership.
• Select the most profitable nodes for campaign generation. This activity is similar to pruning the less profitable nodes from the tree.
• Use the color bar to quickly discern which nodes are positive.