Is it possible to compare two (or more) models of decision tree obtained with ctree on different data? In particular I would like a statistical test that compares two decision trees (builded with the same predictors' names) in R cran.
Asked
Active
Viewed 446 times
1
-
If you have a single dataset, you could use cross-validation after choosing a loss function. I'm a bit unclear what you want to achieve by comparing different models fitted to different data. Could you elaborate? – Stephan Kolassa Jun 07 '16 at 09:48
-
Thank you Stephan, I would like to compare two models that are built on two subset of a main dataset; cross-validation could be an options but my first idea was to compare the models structure without predicting on the dataset.. – Phalaen Jun 07 '16 at 10:02
-
Why do you want to use different subsets of your total data to build your models? – Stephan Kolassa Jun 07 '16 at 10:02
-
Because there is a temporal shift between the two subsets samplings.. – Phalaen Jun 07 '16 at 10:04
-
Thank you. That helps, although I don't have any ideas. Previous questions tagged both "cart" and "model-comparison" are not helpful, either. Good luck. – Stephan Kolassa Jun 07 '16 at 12:15
-
Why not simply include the temporal shift variable as one of the partitioning variables when growing a tree on the entire data set? Or grow the tree once without the temporal shift variable and once with and then check how different the outcome is - either in terms of clustering the data or in predicting the response. – Achim Zeileis Jun 07 '16 at 20:40