How to segment variable based on correlation matrix?

[MikeTurner]

Hi All,

I need to do some segmentation on variables based on correlation matrix. The  correlation matrix is shown as below. If correlation coefficients of some variables in correlation matrix are larger than 0.75 or smaller than -0.75, they will be classified into one subgroup and be stored in one dataset, e.g. dataset_1. Then the variables in dataset_1 will be removed from correlation matrix. Then, if correlation coefficients of some variables in remaining correlation matrix are larger than 0.75 or smaller than -0.75,  they will be classified into another subgroup and be stored in another dataset, e.g. dataset_2. Then the variables in dataset_2 will be removed from correlation matrix....................And repeat until no variables have correlation coefficients of above 0.75 or below -0.75. The remaining variables will be stored in one datsets, dataset_n.

_NAME_	V1	V2	V3	V4	V5	V6	V7	V8	V9	V10	V11
V1	1	0.795283	0.648228	0.702434	0.814356	0.898034	0.72141	0.545573	0.410562	0.67573	0.79505
V2	0.795282885	1	0.785185	0.852621	0.830391	0.75202	0.817556	0.682524	0.517509	0.8671	0.995331
V3	0.648227521	0.785185	1	0.711466	0.690316	0.60089	0.716786	0.552231	0.50897	0.695645	0.824341
V4	0.702433756	0.852621	0.711466	1	0.846026	0.662565	0.748602	0.701647	0.457522	0.757802	0.853448
V5	0.81435639	0.830391	0.690316	0.846026	1	0.817881	0.772293	0.523626	0.398964	0.778101	0.829205
V6	0.898034025	0.75202	0.60089	0.662565	0.817881	1	0.707077	0.457641	0.329565	0.679927	0.750266
V7	0.721409554	0.817556	0.716786	0.748602	0.772293	0.707077	1	0.604067	0.517373	0.788003	0.819051
V8	0.545573147	0.682524	0.552231	0.701647	0.523626	0.457641	0.604067	1	0.585093	0.584004	0.681971
V9	0.410562159	0.517509	0.50897	0.457522	0.398964	0.329565	0.517373	0.585093	1	0.485006	0.522883
V10	0.675730408	0.8671	0.695645	0.757802	0.778101	0.679927	0.788003	0.584004	0.485006	1	0.860323
V11	0.795050139	0.995331	0.824341	0.853448	0.829205	0.750266	0.819051	0.681971	0.522883	0.860323	1

Is there one solution to do this?

Thanks in advance

MT.

[Ksharp · Accepted solution]

Are you doing some analysis Like decision tree ?

I pick up 0.2 as correlation coefficient benchmark .

data x;

array x{*} a1-a20;

do j=1 to 100;

do i=1 to dim(x);

x{i}=ranuni(1234);

end;

output;

end;

drop i j;

run;

%let corr=0.2;

proc sql noprint;

select name into : list separated by ' '

  from dictionary.columns

   where libname='WORK' and memname='X';

quit;

%macro decision;

%let i=1;

%do %while(1);

proc corr data=x outp=person(where=(_TYPE_='CORR')) noprint;

var &list ;

run;

data temp;

set person;

array x{*} a: ;

length var $ 40;

do i=1 to dim(x);

  if x{i} eq 1 then leave;

   else if abs(x{i}) ge &corr then do;

                                  corr=x{i};

                                  var=_name_; output;

                                  var=vname(x{i}); output;

                                 end;

end;

keep corr var;

run;

proc sql ;

create table data_&i as select distinct var from temp;

%if &sqlobs = 0 %then %do;

create table data_&i as select _name_ as var from person; quit;

%return;

%end;

select name into : list separated by ' '

  from dictionary.columns

   where libname='WORK' and memname='PERSON' and name not in (select distinct var from temp) and  name not in ('_NAME_' '_TYPE_');

quit;

%let i=%eval(&i+1) ;

%end;

%mend decision;

%decision

Ksharp

消息编辑者为：xia keshan