Decision Tree Analysis: Odd Tree


My decision tree analysis is odd.

I could have picked a bad attribute.

Any thoughts?

What's odd about it in particular?

1 Like

Hello Misha,
If you look at the terminal node when you traverse on the left branch from the root node, 58%20AM
Can I say that "Oh well, it does not matter what the quantities are, I will just claim customers will buy the Private Label ANYWAY"?

Hi @blackish952

I think pg 314 in introduction to statistical learning which describes why classification trees might have two nodes with the same value. When two terminal nodes are the same the tree is split in any case to increase the purity of the node. More generally you could say that items that have a cost_per_unit less than 13.5586 probably have a value of yes however when the costs per unit is less than 13.5586 and the costs are less than 87.99 and the quantiles are less than 12.5 we are way more certain its a yes

More specifically from pg 314

Why, then, is the split performed at all? The split is performed because it leads to increased node purity.

Hope this is useful

1 Like

Hi John,
When you traverse down the tree, the left side is true and the right side is false. Correct?

For instance, starting from the root node,
"Is the cost less than 182.62?"
If yes, go to the left. Else, go to the right.

Am I right?

I guess it depends on the package you use. You could find some blogs that use the package or the vignette for the package would give you that information.

I think by convention it as you mention above go left when the condition is met go right otherwise (i am open to correction)

I use tree()