topic result score node in SAS Data Science

result score node

yanlu — Sat, 20 May 2017 17:27:45 GMT

Hi,

I have a question about how can change the results computed inside the score : imagine I am an doctor and I want to identify the person who are sick. In the score node, I identify my target is 10% but I only identify 5% of sick people as prediction output. What I would like to do is identify more people healthy as sick and increase this percentage of sick who are really diasgnosed. I don't know how I should proceed to do so in sas.

ty for considering my question .

Re: result score node

Reeza — Sat, 20 May 2017 17:29:00 GMT

You change your scores by chaingimg your models. Determine which model you're using and figure out how to tweak that.

Once you determine your model, if you need further assistance in how to make changes please post back.

Re: result score node

yanlu — Sat, 20 May 2017 19:48:01 GMT

First of all , ty for your answer

I am not really familliar with sas but I can do the basic

The model is one option in sas . It's in the model package ?

ty for your help

Re: result score node

Reeza — Sat, 20 May 2017 20:11:48 GMT

But what kind of model? NN, Decision Tree, Linear Regression Model, Logistic regression model, some form of Ensemble Model?

Honestly, given what you're saying here, I think you should first walk through some SAS Enterprise Miner courses or books because I don't sense a whole of understanding for data mining or statistical methodology.

Re: result score

yanlu — Sat, 20 May 2017 21:23:19 GMT

Well I have check all my models I have done decision tree , linear regression , forest , neural network, maximum regression with the parameter misclassification error minimization

I check the misclassification error table , I don't have any improvement with the different model :'(
One is slightly better with at the level of the classification table concerning my problem, progress of 10 clients.

I was just thinking they were maybe other parameter, statistical wich allow to invers somehow the analysis.

My problem is I am trying to identify client who buy during one period I identify only 10% of the client who buy something so I have wrongly identied 90% of the clients who buy. This 90% who buy effectively are predicted as not buyer false negative

100% of my customer who buy is egual to 1700 more or less. My whole sample is 15 000 clients. So I have 13 300 clients in this period who doesn't buy anything. But I would like to revert this analysis somehow 😛

I think that the problem is maybe a lack of linear correlation between my class imputs , can I do anything to improve this ?
Have you one other explaination about why, this phenomenom appears

thank you for your answer !!!

Re: result score

Reeza — Sat, 20 May 2017 22:13:32 GMT

I'm going to guess that you're defining something wrong. In a case where 90% are not buying, my guess may be that you're actually modeling the opposite event. Double check how your variables are defined.

You would get a 'better' model by simply guessing that everyone did not buy at this rate, so with 15000 records and good input variables I would assume a much better mode.

Re: result score

yanlu — Sat, 20 May 2017 23:14:32 GMT

Hi,

I do identify 150 clients who buy

I should have identified 1500 that actualy the problem

My target is set as binary, I can maybe consider it has nominal class it only contains 0 and 1.

I am going to check again all the variables

thank for your answer !

Re: result score

yanlu — Sat, 20 May 2017 23:52:44 GMT

Is there a way to understand the rapport of two column, I have the sales of each month in my analysis.

the model considers each column separetely for the different month. Maybe if I could tell him this is one variable, my model would be better.

thank for any answer !

Re: result score

Reeza — Sat, 20 May 2017 23:42:14 GMT

@yanlu wrote:

Is there a way for sas to understand the rapport of two column

I don't know what that means.

A time variable such as this should be in a long format, not wide format. Then make sure it's set as a time variable.

Time Value

Jan2015 100

Jan2016 200

Or some other structure may work, it really depends on your problem.

FYI - SAS isn't a him and neither are many of the contributers on here, including myself. If a pronoun isn't required I would suggest not using one.

Re: result score

yanlu — Sun, 21 May 2017 00:09:00 GMT

Sorry for this, I really appologize

Concretely, I have my ID on excel january, februari , mars in each column for example

50 , 100 , 80

How can I passed to the long format

january-february-mart

january 50-february 100- mars 80

I really appreciated the help you have provided

Re: result score

Reeza — Sun, 21 May 2017 03:08:31 GMT

There should be a tranpose task maybe under utilities?

If not, you can use a SAS code node with PROC TRANSPOSE to flip it to a wide dataset.