topic Oversampling method in program description in SAS Programming

Oversampling method in program description

kokodenden52 — Sun, 10 May 2020 07:42:35 GMT

Oversampling is introduced below as a method of building a model when the number of targets is extremely small.

The method by Enterprise Miner is introduced below,

Could it be done by writing program?

https://www.sas.com/offices/asiapacific/japan/service/technical/faq/list/body/em003.html

Sorry to trouble you, but please tell me.

Re: Oversampling method in program description

Ksharp — Sun, 10 May 2020 11:07:14 GMT

data have;
 set sashelp.heart;
run;
proc sort data=have;by status;run;
/*Check population ratio of Dead and Alive*/
proc freq data=have;
table status/missing;
run;

/*If you want oversample 1:1 for Dead and Alive*/
proc surveyselect data=have out=want sampsize=(1000 1000);
strata status;
run;

/*Check the oversample rate*/
proc freq data=want;
table status/missing;
run;

Re: Oversampling method in program description

kokodenden52 — Wed, 13 May 2020 16:33:57 GMT

Mr.Ksharp

Thank you for your answer.

You showed the program in detail.

It helps me very useful.

Once again, thanks a lot.

Re: Oversampling method in program description

PGStats — Wed, 13 May 2020 20:17:45 GMT

If you want oversampling, i.e. to sample more units than there are in the dataset, you will need to add option

method=urs

to the proc surveyselect statement of @Ksharp code above. URS means unrestricted random sampling, i.e. sampling with replacement. By default, proc surveyselect does SRS: simple random sampling which is sampling without replacement.