topic Re: Calculation of the Probabilities from Scores of Decision Tree in SAS Data Science

Calculation of the Probabilities from Scores of Decision Tree

mohammad__101 — Sun, 21 Aug 2016 14:31:40 GMT

Dear All,

After I have created a Decesion Tree Model, kindlly i need your help to advice me how can I convert the Score I optained from the Variable P_target_1 in to Probabilites . Please note that I haven't used Deceison Node . Is that avaiable ?

All I have right now is the Score of Prediction and a Flag for the real event (1/0)

P.S : I have reviesed the below link (https://communities.sas.com/t5/SAS-Data-Mining/Adjusted-Prior-in-Enterprise-Miner-7-1-Please-help-Thank-You/td-p/152065) for a simmilar enquiry but couldn't find a solution to my particular case.

Thanks and Best Regards,

Mohammed ElSofany

Data Scientist .

Re: Calculation of the Probabilities from Scores of Decision Tree

Reeza — Sun, 21 Aug 2016 14:47:56 GMT

What do you mean by probability? Probability of being a 0/1? I don't think that's how a decision tree process works. It's a rule based algorithm, not probabilistic.

Re: Calculation of the Probabilities from Scores of Decision Tree

mohammad__101 — Sun, 21 Aug 2016 15:03:44 GMT

Dear Reeza,

Thanks for your reply . Normaly DT is not Probabilistic . But what if I want to extract from the Score produced a Probability of the event to happen (to be 1 not 0 ) .

Thanks

Re: Calculation of the Probabilities from Scores of Decision Tree

Reeza — Sun, 21 Aug 2016 15:08:34 GMT

I'm trying to think how that would work and it would be equivalent to the prior probability of 0/1 is my guess.

If you need a probability why not fit a different model type, ie logistic regression?

If you fit multiple trees with differing rules then perhaps an average of outcomes could be determined to extract a probability. So a simulation type analysis.

Re: Calculation of the Probabilities from Scores of Decision Tree

Reeza — Sun, 21 Aug 2016 15:31:16 GMT

I think your looking for the conditional probability of the end points/nodes. Does that sound correct?

Similar to this?

http://petrowiki.org/Decision_tree_analysis

Re: Calculation of the Probabilities from Scores of Decision Tree

mohammad__101 — Sun, 21 Aug 2016 15:48:05 GMT

As a matter of fact that's simmilar to what I'm reading now : )

https://www.microsoft.com/en-us/research/wp-content/uploads/2016/02/decisionForests_MSR_TR_2011_114.pdf

Thanks !

Mohammed ElSofany

Re: Calculation of the Probabilities from Scores of Decision Tree

mohammad__101 — Sun, 21 Aug 2016 17:13:47 GMT

In a matter of fact I still can't get it from the both links .

All I am looking for is to get the Probability using the Score and using overall Probabilty of event occure gained from trainning data or cross validation data.

VBR

Re: Calculation of the Probabilities from Scores of Decision Tree

Reeza — Sun, 21 Aug 2016 18:02:59 GMT

Is your issue technical, ie how to calculate a specific value, or methodological - how do I calculate the probabilities.

Please take the time to explain your question in as much detail as possible. Especially if its a calculation question.

Re: Calculation of the Probabilities from Scores of Decision Tree

rayIII — Sun, 21 Aug 2016 20:09:31 GMT

For a decision tree built using a nominal or binary target, the P_ variables are the posterior probabilities (given a set of input values, what is the predicted probability of a particular outcome) for each possible outcome.

So you probably already have what you are looking for.

Here's a detailed reference:

Meaning of score output variables

Re: Calculation of the Probabilities from Scores of Decision Tree

mohammad__101 — Mon, 22 Aug 2016 12:33:31 GMT

Thanks Rally for your reply ,

Is this can be applied even if I didn't use Decesion Node and I have used Oversampling ?

Regards,

Re: Calculation of the Probabilities from Scores of Decision Tree

mohammad__101 — Mon, 22 Aug 2016 12:49:10 GMT

Dear Reeza,

Thank you for all the message very much!

Yes It's a calculation ,

I have built a Decesion Trees Model with Binary Classification.

To build this model I have used Oversampling approach .

I have't used any Decesion Nodes ( Before Model or After Model )

I want to get the Probability of occurnace of the event from the P_target_1

P_target_1 is not probability it's Model Score .

I need to have the Probability .

I hope I was clear enough,

Thanks and best regards,

Mohammed ElSofany