1

I am trying to use SPSS Modeler to test a decision tree model built in SPSS statistics, but I can't find any straightforward way to do it (only xml export, which I cannot import later). I also tried to re-build the model in Modeler using the same dataset and settings (CHAID, stopping rule minimum records in parent branch = 100 and 50 in child branch, etc.) but the results are completely different. I am using 3 input variables that in SPSS Statistics result in a 343 node tree, whereas in Modeler only one of them is included in a model of 3 nodes. Is there any way to import/export models or re-built them sharing settings?

eli-k
  • 10,898
  • 11
  • 40
  • 44
vabm
  • 285
  • 7
  • 16

2 Answers2

1

You can apply a Tree model directly in Statistics. The rules can be saved in a number of ways. I don't know of a direct way of using the xml from Statistics Trees directly in Modeler, though.

JKP
  • 5,419
  • 13
  • 5
  • Do you know the reason why the models might be different in Statistics and Modeler? I also tried it in RapidMiner and gives me the same results as Modeler. I assume might be because one of the variables has stronger predictive capabilities so the others are automatically disregarded by data mining software. But then, how can I get the same results in Statistics? It seems to use all variables regardless of their predictive power. – vabm Jan 03 '16 at 23:53
  • Just found my own answer, it was the number of intervals for scale independent variables. Statistics presets them at 10, which in my case gives better results than the presets of Modeler or Rapidminer – vabm Jan 04 '16 at 00:34
1

You can import Statistics's CHAID model in .xml as modeling nugget into Modeler via "Import PMML" choice [right click in model tab, lowest choice].

Vojtěch Skubanič
  • 141
  • 2