Initializing help system before first use

Profiling Decision Trees

When you profile a decision tree, you choose one or more variables to follow through as the input data is processed.
You will need to decide on one or more profile variables - either via the Profile Variables option on the main user interface, or via the modal dialog that pops up if necessary prior to a tree count.
Important Xpress Insight only supports the profiling of non-categorical variables (i.e. numeric variables).

The Profile View allows you to analyze and review the profile variables you defined for the tree as the input records are processed by the tree's nodes. In this view, all the decision nodes and their split conditions appear in tabular format where the tree hierarchy is preserved.

You can view the statistics for each profiled variable. These statistics can be modified by hovering over any column, clicking the drop-down arrow that becomes visible, selecting Columns and clicking the check box next to each statistic you wish to display. You can also display different statistics for each profile variable.

The figure shows the Profile view being configured in this way.

Profile View

Profile View

Covered Topics
  • Determining Best Split with Target-Driven Decision Trees
    When you already know what the target assignment is and want to edit the tree so it matches this target, you can use the target variable and the best split functions.
  • Best Split Algorithm—Gini Impurity Measure
    When a decision tree is defined with a target variable and the Best Split algorithm is applied, the algorithm aims to partition the data so that the resulting group of records at the new node minimizes impurity. A node with high impurity has a high population of several different values of the target variable because the parent split has not segmented the data effectively.
  • Choosing Profiling Statistics
    In general, variables and data represent groups of things about categorical or discrete characteristics, or they represent measurements on a continuous scale.
  • Viewing Leaf Node Statistics
    The Leaf View shows statistics and treatments for all leaf nodes in the decision tree.
Parent Topic Manipulating Decision Trees