Running a StatExplore on top of a Neural Network - H1=1, H1=2, H1=3?

[helenn]

Hi All,

I have built a neural network to predict a binary target with several class/interval predictors. To gain an understanding of the predictors on the model, I am running a StatExplore node out of my Neural Network node (see output below).

My question is, what are the H13, H12, H11 variables which have been created? I thought they were referring to the first/second/third nodes which are in the first layer, but this doesn't seem to make sense. They make up a significant percentage of the model, so I am looking to understand their influence. Also, is there another method for explaining a neural net (besides approximating it with a decision tree)?

Target	Variable	Importance	Worth	Analysis Variable	Label	plot
target_ind	H13	1	1.53E-05	1	Hidden: H1=3	NaN
target_ind	H12	2	1.16E-05	1	Hidden: H1=2	NaN
target_ind	var1	3	9.29E-06	1	var1	NaN
target_ind	H11	4	7.41E-06	1	Hidden: H1=1	NaN

Thanks so much!

[WendyCzika · Accepted solution]

Yes, those do correspond to the units/neurons in the hidden layer(s), so H11 is the first unit in the first hidden layer, and so on.  You can have those dropped from the data exported from the Neural Network node by setting the Hidden Units property (in the Score section) to No.  Or if you want them in there but not used by StatExplore, you can set Use=No for those variables from the Variables property of StatExplore.