Report Item: Decision Tree

Decision Trees are a type of machine learning tool that analyzes historical data, looking at past candidates/students who have been successful, and then makes rules to describe the most and the least successful. By applying these rules to current candidates/students, we can get some idea of whether they are likely to succeed. How it works:

• Select the number of Instances,
• Select the variables to use,
• Select one or more Role(s) to focus on, if necessary; and
• Identify which Statuses represent “passing” and “failing”.
• The Decision Tree splits the data (into yes, branches) based on what would best divide the outcome into two groups; and then repeats that splitting process to attempt to display a predicted outcome.

The optimal data set needs to have at least 2 variables and 200 total data points from all the Instances. That number increases by 50 for each variable added.

The Decision Tree settings are the same for the Custom Person and Custom Group Report with the exception that you specify a person for the report to reference in the former.

General Options

• Columns … Specify how many columns (1-4) the output takes up on the report.

Decision Tree Options

• Size … Choose Small, Medium (default), or Large from the dropdown.
• R Script to use … The Tree script is the most popular, but other scripts may be made available. The Tree script is the most popular, but other scripts may be made available. Contact your GideonSoft administrator to do so.
• Current Configurations … The last paragraph of the instructions provides dynamic feedback about the size of the dataset and if the dataset is Invalid (will not run), Not Optimal, or Optimal.
• # of Historical Measurement Periods to Analyze … The number of past instances to be included in the analysis.
• “Passing” Status List … Choose the target statuses representing the desirable outcomes (e.g. Graduated).
• “Failing” Status List … Choose the target statuses representing the undesirable outcomes (e.g. Failed).
• Use people from the following roles … (Optional) Use dropdown to select one or more Role(s) to include in the Decision Tree. Leaving it blank will include anyone that has a valid score for the selected variable(s).
• Variables … Select one or more variable(s) to reference.
• Click Save.

Decision Tree Interpretation

• Starting at the top, each splitting point (node, branch) represents a rule to guide decisions for classifying the data.
• Each split will have a variable name and the value at which the split is occurring.
• The thickness of the lines represents the number of people in that path.
• The boxes at the end (leaves) contain three rows of information about the categorization used following the related path.
  • Row 1 … Predicted Classification
    • 1 = Specified “Passing” outcome
    • 0 = Specified “Failing” outcome
  • Row 2 … Accuracy, represented by the percentage of individuals in this “leaf” that are “Passing” (1) in the data.
  • Row 3 … The percentage of the total sample in the “leaf”.

Note: Use the Decision Tree with caution because the results may not generalize across samples.

= = = = =

Version(s): GideonSoft 2021 Release 2 and later