Narrowing Inputs - Prediction Model w/ Poisson distributed target

[teg_76]

Hi,

I'm new to use Enterprise Miner, and Predictive Modeling in general, but have a stats background.  I have a dataset with about 100 inputs and a target variable with a discrete Poisson distribution.  I'd like to know:

1) Can I use a decision tree model in EM to help me narrow down inputs for a Poisson Model?

2) Can EM model a Poisson distribution, without using the "Rate Making" node (I don't have this node)?

Please let me know what additional information you need to help me answer my questions.

Thank you!

[M_Maldonado]

Hi,

What version of Enterprise Miner do you have?
What are you trying to model? Just asking to confirm your discrete target is nominal.

Thanks!

[teg_76]

Thank you Miguel. I'm using EM 13.1.

My discrete target is ordinal, with a minimum of 1.

[WendyCzika]

Hi there...

1) Yes, decision trees make no assumptions about the distribution of your target and are very robust, so you can certainly use that for variable selection

2) That depends... what version of Enterprise Miner do you have?

[teg_76]

Thank you Wendy for your reply. I'm using EM 13.1

[WendyCzika · Accepted solution]

Oh great, then you can use the HP GLM node (on the HPDM tab) - set the level of your target as interval (even though it is integer-valued) and by default, the HP GLM node will use the Poisson probability distribution with a log link function.  

[teg_76]

Great...thank you Wendy!