How can I visualize decision tree leaves in tabular form effectively? I have a decision tree from a classification model that is 6 levels deep and has about 30 different leaf nodes.
In a table, I want to sort each leaf node by training probability, and capture cumulative statistics for the test data set on the other columns. Right now, it looks something like this, but for longer decision paths:




Decision Path
Cumulative Population %
Capture Rate




X1>=1|X2<0
5%
15%


X1 < 1|X2>10
10%
25%


X1>=1|X2>=0|X3>5
15%
32%




where capture rate is the % of the positives captured over the cumulative population.
Additionally, this view becomes more difficult to read when there are about a dozen or so X values with different names, with meanings that require a separate dictionary/key to describe, e.g.,




Variable
Description




X1
Number of pets owned


X2
Net worth (in thousands USD)


X3
Number of children




Ideally I would like to put the resulting table in a powerpoint slide, with the top 10 leafs in tabular form, but using the table seems cluttered, and the tree visualization I have is too large, even if I were to color-code the different branches by ranking.
 A: Here's an idea I mocked up quickly in Excel:

I've simplified the predictor names to one or two words - you can always give the detailed explanation on another slide, in the slide notes or in a small-print footnote.
I've used either <n to denote a predictor value less than n, or n+ to denote a predictor value greater than or equal to n, which I think are both compact and easily understood.
Finally I've used colours to separate the cumulative population and capture rate columns and a colour scale to highlight the increasing cumulative population and capture rate values. I'm sure you could do more with colour or shading to suit your preferences.
I think you could fit ten rows of this table on a PowerPoint slide and have it reasonably legible. It's not a way of presenting a tree classifier that makes the model all that easy to understand in my opinion, but if you want to be able to show something that satisfies your audience that you have the data behind your classifier, it seems like it would fit the brief.
