Question related to "apply model".

maverik

Hello! I generated a decision tree classifier that contains an attribute A in some branches.It predicts for a binary target attribute (0 or 1). During data cleaning, I removed all the examples that contain missing value for attribute A. I now try to use this list of removed examples as an independent testing set, as the classifier only needs A for certain cases to make a prediction. So I connect this testing set to "apply model", and connect the model to "apply model", then I get predictions for all of the examples.

So even if some predictions need value in A but it is missing, the classifier still predicts for 0 or 1. Could someone please explain why this is possible? My understanding is that you gonna need to trace down the tree to make the prediction. Then how can the classifier trace further down from A when the value is missing?

Also, for many examples, the confidence values for 0 and 1 are "?".

homburg

Hello maverik,

when you remove an attribute used by a previously generated model you will most likely get a default prediction whenever the missing attribute is needed. In your example you might get correct predicions in every case A is not needed and a default (first internal attribute) when the model refers to A. The confidence value for those predictions will be "?".

Cheers,
Helge