PresentationPDF Available

Variable selection techniques

Authors:
  • Artificial Engineering

Abstract

Variable selection techniques are employed in statistical modelling process in order to reduce the number of predictor variables that are part of the model of the experiment under study.
!"#$%&'()*+
,#'-#./&(0&/&1%-23(4&1"3-56&7
!"#$%&'( )*(-7($'&7&3%&8(.9(:38'&#(8&(;-2'<-2
=">(!#38-8#%&(-3(='2861%-23(?3<-3&&'-3<@(#38'&#8<AB%"C7&@(#38'&#8&<-2'<-2C12D
!26'7&(-3(?E$&'-D&3%#/(>&7-<37(FG;H*IJK
$'2LC(M#'7(G#%%7723
N4O@(P29#/(Q37%-%6%&(2L(4&1"32/2<9
+4"&( 1"#$%&' (.&/23<7( %2( %"&(.22B (R0%#%-7%-1#/( >&7-<3( #38(:3#/97-7(2L(?E$&'-D&3%7S@
PCMC(G#723@(PCTC(;637% #38(UCMCO&77@($6./-7"&8( .9(U2"3(V-/&9(#38(0237@(WIIH(
G#9()X%"@(WI)Y
Z[&'[-&\
)*C)C( !2D$#'-3<(L-%%&8(D28&/7
)*CWC( ://]$277-./&]76.7&%(12D$#'-7237
)*CHC( 0%&$\-7&( 7&/&1%-23( D&%"287
)*CHC)C(T2'\#'8(7&/&1%-23
)*CHCWC(^#1B\#'8(&/-D-3#%-23
)*CHCHC(0%&$\-7&(-%&'#%-23
)*CJC(!2//-3&#'(&LL&1%7
Z[&'[-&\
4"& D#-3 #7$&1%7 #'&_
!'-%&'-# L2' %"& 7&/&1%-23 2L -D$2'%#3% $'&8-1%2' [#'-#./&7`
06.7&% 7&/&1%-23 D&%"287@ -31/68-3< %"& #77&77D&3% 2L #//
$277-./& '&<'&77-237 #38 7%&$\-7& 7&/&1%-23 $'21&86'&7`
!2//-3&#'-%9 &LL&1%7 23 7&/&1%-23 D&%"287C
!2D$#'-3<(L-%%&8(D28&/7
!"#$$%&$ ' ()$ * (
"+&" $* (,+&, *--- *(.+&. * /&
!,#$$%&$ ' ()$* (
"+&" $ * (,+&, *--- *(0+&0 * /&1
!"123%#-37 %"& 12D$/&%& 7&% 2L 2$'&8-1%2' [#'-#./&7 #38 -7
#776D&8 %2 #8&56#%&/9 '&/#%& %"& '&7$237& [#'-#./& %2 %"& $'&8-1%2'
[#'-#./&7C
!,1237-7%(2L(%"&(L-'7%(3$ 4 $2$$'&8-1%2'([#'-#./&7C
!2D$#'-3<(L-%%&8(D28&/7_(1'-%&'-#
1) F$statistic
5 ' !678!"9!,:
!6;"
4"-7 7%#%-7%-1 1#3 .& 1#/16/#%&8 L'2D %"& &''2' 76D7 2L 756#'&7 L2'
%"& %\2 D28&/7_
!67 !"!,'<<=>?<<=@
.?0 #38((!6;"'<<=@
A?.?"
4"-7(-7(67&8(%2(%&7%(%"&("9$2%"&7&7_
B)#$a0C" ' D ' a.' E
[7
BF#:%( /&#7% 23& a0C" 'D'a.' EG H ' 3 * IG J G 2
!2D$#'-3<(L-%%&8(D28&/7_(1'-%&'-#
2)+Adjusted coefficients of+determination
0-31& %"& '&861&8 D28&/7 82 32% 123%#-3 %"& 7#D& 36D.&' 2L
$'&8-1%2' [#'-#./&7 #7 %"& 12D$/&%& D28&/@ #8b67%&8 12&LL-1-&3%7 2L
8&%&'D-3#%-23 7"26/8 .& 12D$#'&8_
7F
,' I K L K I
L K 3 K I M66;,
N66
T2' %"& 12D$/&%& D28&/@ 3-7 '&$/#1&8 .9 2#38 66;,.9 66;"C
QL %"& #8b67%&8 12&LL-1-&3%7 2L 8&%&'D-3#%-23 #'& #$$'2E-D#%&/9
&56#/ L2' %"& L6// #38 #'&861&8 D28&/@ %"& %\2 L-%7 #'& '&<#'8&8 #7
"#[-3< &56#/ $'&8-1%-[& #.-/-%9C
!2D$#'-3<(L-%%&8(D28&/7_(1'-%&'-#
3)+C$statistic
OP'66;,
!6;"
K 8L K QR:
\"&'& R$ ' $3$ * $IC
:3 #8&56#%& '&861&8 D28&/ -7 23& L2' \"-1" OP-7 #$$'2E-D#%&/9
&56#/ %2 RCV"&3 $'&8-1%2'7 L2' \"-1" (S$ ' $E$$#'& 12''&1%/9
&/-D-3#%&8 L'2D %"& D28&/@ 66;,&7%-D#%&7 8L$ K $3$ K $I:$T,$#38
!6;"&7%-D#%&7 T,C4"67@ OP-7 #$$'2E-D#%&/9 &56#/ %2 RC
V"&3 63-D$2'%#3% $'&8-1%2'7 '&D#-3 -3 #'&861&8 D28&/@ OP1#3
.& D61" /&77 %"#3 RC
://]$277-./&]76.7&%(12D$#'-7237
4"& D27% 12D$'&"&37-[& #77&77D&3% 2L 1#38-8#%& '&861&8 D28&/7
-7 #12D$#'-723 2L %"& 7&/&1%-23 1'-%&'-# F)K %2 FHK L2' #// $277-./&
76.7&%7 2L $'&8-1%2' [#'-#./&7C
^9 12D$#'-3< #// $277-./& 76.7&%7@ #3 -3[&7%-<#%2' 1#3_
8&%&'D-3& %"& .&7%+ '&861&8 D28&/F7K #112'8-3< %2 %"& #.2[&
1'-%&'-#`
-8&3%-L9 #/%&'3#%-[&7 %2 %"& .&7%+ 23&7C
*Note:There is no single reduced model that outperforms all others
because the selection criteria are all random variables and therefore
subject to variability.
://]$277-./&]76.7&%(12D$#'-7237_(#3(&E#D$/&
Chemical$yield data:detection of influential observations
4"& 76.7&% 7&/&1%-23 1'-%&'-# L2' #// $277-./& '&861&8 D28&/7 FL-'7% BK_
://]$277-./&]76.7&%(12D$#'-7237_(#3(&E#D$/&
Chemical$yield data:detection of influential observations
7F
,U EVWX
://]$277-./&]76.7&%(12D$#'-7237_(#3(&E#D$/&
Chemical$yield data:detection of influential observations
R 4 OP4 R * Q@(\"&'&(R$ ' $3$ * $I
FL-'7%$3$K
FL-'7%$3$K
://]$277-./&]76.7&%(12D$#'-7237_(#3(&E#D$/&
Chemical$yield data:detection of influential observations
0%&$\-7&(7&/&1%-23(D&%"287
QL(%"&(36D.&'(2L(1#38-8#%&($'&8-1%2'([#'-#./&7(-7($'2"-.-%-[&/9(/#'<&@(
7%&$\-7&(7&/&1%-23(D&%"287(#'&($'&L&''&8C
4"&7&(D&%"287(7&56&3%-#//9(#88(2'(8&/&%&( $'&8-1%2'([#'-#./&7(%2(%"&(
$'&8-1%-23(&56#%-23@( <&3&'#//9( 23&(#%(#(%-D&C
4"'&&($2$6/#'(76.7&%(7&/&1%-23(D&%"287(\-//(.&(8-71677&8(-3(%"-7(
7&1%-23_(forward+selection@(backward+elimination@( #38(stepwise+
iterationC
4"&(D&%"287(#'&(-D$/&D&3%&8(67-3<(%"&(T]7%#%-7%-17C
T2'\#'8(7&/&1%-23(FT0K
='21&86'& L2' %"& L2'\#'8 7&/&1%-23 D&%"28_
)C T-% 27-3</&][#'-#./& '&<'&77-23 D28&/7@ &#1" "#[-3< 23& 2L %"&
$'&8-1%2' [#'-#./&7 -3 -%C!#/16/#%& %"& 2[&'#// D28&/ T]7%#%-7%-1
L2' &#1" 2L %"& 2L-%7C0&/&1% %"& L-% \-%" %"& /#'<&7% T]7%#%-7%-1C
WC c7-3< #7-<3-L-1#31& /&[&/ 2L Y ' EVQX@%&7% %"& 7-<3-L-1#31& 2L
%"& $'&8-1%2' [#'-#./&CQL -% -7 32% 7-<3-L-1#3%@ %&'D-3#%& %"&
$'21&86'& #38 1231/68& %"#% 323& 2L %"& $'&8-1%2'7 #'& 67&L6/
-3 $'&8-1%-3< %"& '&7$237&CQL -% -7 7-<3-L-1#3%@ '&%#-3 %"& $'&8-1%2'
[#'-#./&@ 7&% Z ' I #38 $'21&&8 %2 7%&$ HC
HC T-% 2 K Z '&861&8 D28&/7@ &#1" "#[-3< %"& Z$'&8-1%2' [#'-#./&7
L'2D %"& $'&[-267 7%#<& 2L %"& 7&/&1%-23 $'21&77 #38 23& 2L %"&
'&D#-3-3< $'&8-1%2'7C!#/16/#%& %"& 2[&'#// D28&/ T]7%#%-7%-1 L2'
&#1" 2L %"& L-%7C0&/&1% %"& L-% \-%" %"& /#'<&7% T]7%#%-7%-1C
T2'\#'8(7&/&1%-23(FT0K
='21&86'& L2' %"& L2'\#'8 7&/&1%-23 D&%"28_
JC c7-3< #7-<3-L-1#31& /&[&/ 2L Y ' EVQX@%&7% %"& 7-<3-L-1#31& 2L
%"& #88-%-23#/ $'&8-1%2' [#'-#./& 67-3< %"& T]7%#%-7%-1 \-%" %"&
D28&/ !"123%#-3-3< %"& Z * I $'&8-1%2'7 #38 %"& '&861&8
D28&/ !,123%#-3-3< %"& Z$'&8-1%2'7 L'2D %"& $'&[-267 7%#<& 2L
%"& 7&/&1%-23 $'21&77C4"& 8&<'&&7 2L L'&&82D L2' %"-7 T]7%#%-7%-1
#'& ["' I #38 [,' L K Z K QC
dC QL %"& T]7%#%-7%-1 -7 32% 7-<3-L-1#3%@ %&'D-3#%& %"& $'21&86'& #38
'&%#-3 %"& Z$'&8-1%2'7 L'2D %"& $'&[-267 7%#<&CQL -% -7 7-<3-L-1#3%@
#88 %"& #88-%-23#/ $'&8-1%2' %2 %"& Z$'&[-267/9 7&/&1%&8@
-31'&D&3% Z.9 )#38 '&%6'3 %2 7%&$ HC
T2'\#'8(7&/&1%-23(FT0K_(#3(&E#D$/&
Chemical$yield data:detection of influential observations
4"& L2'\#'8 7&/&1%-23 $'21&86'& -7 #$$/-&8_
^#1B\#'8(&/-D-3#%-23(F^?K
='21&86'& L2' %"& .#1B\#'8 &/-D-3#%-23 D&%"28_
)C T-% %"& 12D$/&%& D28&/ "#[-3< #// 2$'&8-1%2' [#'-#./&7 -3 -%C
0&/&1% %"& [#'-#./& "#[-3< %"& 7D#//&7% \2' T]7%#%-7%-1C4"& T]
7%#%-7%-1 -7 %"& 756#'& 2L %"& %]7%#%-7%-1 67&8 %2 %&7% %"&
7-<3-L-1#31& 2L (S-3 %"& 12D$/&%& D28&/C
WC c7-3< #7-<3-L-1#31& /&[&/ 2L Y ' EVQX@%&7% %"& 7-<3-L-1#31& 2L
%"& $'&8-1%2' [#'-#./&CQL -% -7 7-<3-L-1#3%@ %&'D-3#%& %"&
$'21&86'& #38 '&%#-3 %"& $'&8-1%2' [#'-#./&7CQL -% -7 32%
7-<3-L-1#3%@7&% Z ' I #38 $'21&&8 %2 7%&$ HC
^#1B\#'8(&/-D-3#%-23(F^?K
='21&86'& L2' %"& .#1B\#'8 &/-D-3#%-23 D&%"28_
HC T-% %"& '&861&8 D28&/ "#[-3< %"& 2 K Z $'&8-1%2' [#'-#./&7
'&D#-3-3< L'2D %"& $'&[-267 7%&$C!#/16/#%& T]7%#%-7%-17 L2'
%&7%-3< %"& 7-<3-L-1#31& 2L &#1" 2L %"& '&D#-3-3< [#'-#./&7C
0&/&1% %"& $'&8-1%2' [#'-#./& "#[-3< %"& 7D#//&7% T]7%#%-7%-1C4"&
8&<'&&7 2L L'&&82D L2' %"-7 T]7%#%-7%-1 #'& ["' I #38 [,' L K
2 * Z K IC
JC c7-3< #7-<3-L-1#31& /&[&/ 2L Y ' EVQX@%&7% %"& 7-<3-L-1#31& 2L
%"& 7&/&1%&8 $'&8-1%2'CQL %"& T]7%#%-7%-1 -7 7-<3-L-1#3%@%&'D-3#%&
%"& $'21&86'& #38 '&%#-3 %"& '&D#-3-3< ] K Z $'&8-1%2'
[#'-#./&7 L'2D %"& $'&[-267 7%#<& 2L %"& 7&/&1%-23 $'21&77CQL -% -7
32% 7-<3-L-1#3%@8&/&%& %"& 7&/&1%&8 [#'-#./&@ -31'&D&3% Z.9 )
#38 '&%6'3 %2 7%&$ HC
^#1B\#'8(&/-D-3#%-23(F^?K_(#3(&E#D$/&
Chemical$yield data:detection of influential observations
4"& .#1B\#'8 &/-D-3#%-23 $'21&86'& -7 #$$/-&8_
^#1B\#'8(&/-D-3#%-23(F^?K_(#($277-./&($'2./&D
Q3 %"& T]7%#%-7%-1 %"& D&#3]756#'&8 &''2' -7 676#//9 '&1#/16/#%&8
#L%&' &#1" $'&8-1%2' [#'-#./& -7 &/-D-3#%&8CQL 7&[&'#/ $'&8-1%2'
[#'-#./&7 #'& &/-D-3#%&8@ %"&'& -7 #8#3<&' %"#% %"& D&#3]756#'&8
&''2' 126/8 .&12D& .-#7&8C^&1#67& 23/9 7D#// T]7%#%-7%-17 1#67&
$'&8-1%2'7 %2 .& &/-D-3#%&8@ %"& 76D7 2L 756#'&7 %"#% #'& #88&8 %2
%"& &''2' 76D 2L 756#'&7 1#3 1#67& %"& D&#3]756#'&8 &''2' %2
638&'&7%-D#%& %"& &''2' [#'-#31&CQL %"-7 -7 #1231&'3@ %"& D&#3]
756#'&8 &''2' L'2D %"& 2'-<-3#/ L-% %2 %"& 12D$/&%& 8#%# 7&% \-%" #//
%"& $'&8-1%2'7 -31/68&8 7"26/8 .& 67&8 -3 %"& 8&32D-3#%2' 2L %"& T]
7%#%-7%-1 &56#%-23_
5 ' !678!"9!,:
!6;"
0%&$\-7&(-%&'#%-23(F0QK
='21&86'& L2' %"& 7%&$\-7& -%&'#%-23 D&%"28_
)C Q3-%-#%& %"& 7&/&1%-23 $'21&86'& #7 -3 L2'\#'8 7&/&1%-23C
WC :L%&' %"#% #$'&8-1%2' [#'-#./& -7 #88&8@ #$$/9 7%&$ H#38 J2L
%"& .#1B\#'8 &/-D-3#%-23 $'21&86'& %2 %"& $'&8-1%2' [#'-#./&7
-3 %"& D28&/C
HC !23%-36& %"& 7&/&1%-23 $'21&86'& #7 -3 L2'\#'8 7&/&1%-23C:%
&#1" 7%#<& 2L %"& $'21&77@ -L #$'&8-1%2' [#'-#./& -7 #88&8@ '&%6'3
%2 7%&$ WC 4&' D - 3# % & %"& $'21&86'& \"&3 32 $'&8-1%2' [#'-#./&7
#'& #88&8 #% #7%#<& 2L L2'\#'8 7&/&1%-23C
0%&$\-7&(-%&'#%-23(F0QK_(#3(&E#D$/&
Chemical$yield data:detection of influential observations
#88&8
32%(8&/&%&8
32%(8&/&%&8
#88&8
#88&8
32%(#88&8
!2//-3&#'(&LL&1%7
G#39 8&7-<3&8 &E$&'-D&3%7@ 32%#./9 12D$/&%& L#1%2'-#/7@ 9-&/8
-8&3%-1#/ 76.7&%7 '&<#'8/&77 2L %"& 7&/&1%-23 D&%"28 67&8C
Z3& $'2$&'%9 2L 8#%# 7&%7 %"#% -7 B32\3 2L%&3 %2 $'2861& '&76/%7
8&$&38-3< 23 %"& 76.7&% 7&/&1%-23 D&%"28 -7 12//-3&#'-%9
F'&8638#319K #D23< $'&8-1%2' [#'-#./&7C
!2//-3&#'(&LL&1%7_(#3(&E#D$/&
P&<'&77-23(L-%(%2(%"&(!'92<&3-1]T/2\D&%&'( >#%#
!2//-3&#'(&LL&1%7_(#3(&E#D$/&
T2'\#'8(7&/&1%-23(7&/&1%7(7-E($'&8-1%2'([#'-#./&7@(#//(%"27&(/-7%&8(
-3(%"&(%#./&(&E1&$%(4?G=(#38(4QG?C
^#1B\#'8(&/-D-3#%-23( #/72(7&/&1%7( 7-E($'&8-1%2'7@( 8&/&%-3<( >?e0(
#38(4QG?C
0%&$\-7&( -%&'#%-23(#<'&&7(\-%"(L2'\#'8(7&/&1%-23C
:3 &E#D-3#%-23 2L #// $277-./& '&861&8 D28&/7 -38-1#%&7 %"#% .2%"
2L %"&7& 7-E][#'-#./& 76.7&%7 #'& #11&$%#./&@ %"& T0 76.7&% "#[-3<
7,$ ' $EV^^X$#38 O_$ ' $`VEa@#38 %"& ^? 76.7&% "#[-3< %"& 7#D&
7,\-%" O_$ ' $`VI^CV"#% -7 8-7%6'.-3< #.26% %"&7& '&76/%7 -7 %"#%
%"& %]7%#%-7%-17 76<<&7% %"#% .2%" >?e0 #38 4?G= 1#3 .& 8&/&%&8 .6%
3&-%"&' 2L %"& #.2[& 76.7&%7 8&/&%&7 .2%"C4"& '&#723 L2' %"&7&
#$$#'&3%/9 123%'#8-1%2'9 '&76/%7 -7 %"#% DENS and TEMP are highly
collinearC4"& 12''&/#%-23 12&LL-1-&3% F=&#'723f7 'K .&%\&&3 %"&7&
%\2 $'&8-1%2'7 -7 gIC***C
!2//-3&#'(&LL&1%7_(#77&77D&3%
,#'-#31&(-3L/#%-23(L#1%2'7(F,QTK_(
bc5
S$ ' $ 8I$ K $7S
,: K I
\"&'& 7S
,-7 %"& 12&LL-1-&3% 2L 8&%&'D-3#%-23 2L %"& '&<'&77-23 2L +S
23 %"& 1237%#3% %&'D #38 %"& 2%"&' $'&8-1%2' [#'-#./&7 -3 %"&
D28&/C
:$'&8-1%2' [#'-#./& %"#% -7 "-<"/9 12//-3&#' \-%" %"& 1237%#3% %&'D
2' 23& 2' D2'& 2%"&' $'&8-1%2'7 \-// "#[& #7S
,[#/6& 1/27& %2 )C 4"-7
-3 %6'3 \-// $'2861& #/#'<& [#'-#31& -3L/#%-23 L#1%2'C
Q3 %"& $'&[-267 &E#D$/&@ %"& 12//-3&#'-%9 .&%\&&3 >?e0 #38 4?G=
23 %"&-' [#'-#31& -3L/#%-23 L#1%2'7 \#7 2[&' )IIIC
ResearchGate has not been able to resolve any citations for this publication.
ResearchGate has not been able to resolve any references for this publication.