目的に応じて手法を選ぶ 分類が有効な例 決定木 (Decision Trees)

11
)(Decision Trees)
4=G
%H;N8
I4
IM?H;N8
7!==MB:G1 FB:EG91
ANZ`0'Xb[I
P_YbY^I%
,G\b_FT]aR\b_I4
Sb_VaWbHN5<LI\b_I4
.3<LII
$@E?(&@E?)+96
!
=)#)(6#)+96CEC6
3,:&7F5(6*:68
A?ED6*:403F=B><.:(= <-)
A?ED"'7F+912-:(%)
CEC7F63#/;:
2
H;N8
QUF)#O2ANF>H7&$HKCDO
A
N8
6+"O*ANF>H7&$F6+-(H@D,G
O/J8
=B>B
G
…… …… …)
=B>A
=B>B
$"
&)(1(2$1+0,/-.2
#&'%!$(
$% $*
-6+*6)
98EM%-6+/6.%.2,(7.5,1
:8-6+*6)%SAMPSIO.HMEQ%
;8[]+0BAD [
]%Target-6+
/6.%#
BAD [
]
&6*4%$
'3,'
[
]!"
Target%#
O]W[T^QE2
XT[E
(&)
bad
target
binary
6
5>GBF'
D*
clage
input
interval
!H?U\^VYK]E"=
clno
input
interval
U\^V_M\PRU`YK]E
E
debtinc
input
interval
delinq
input
interval
*@C?IU\^VYK]E
derog
input
interval
.7$/?JA
3%ELSNZ
job
input
nominal
loan
input
interval
+E48;<
mortdue
input
interval
+EE#:
!9EM\PRUE?JA
ninq
input
interval
reason
input
binary
value
input
interval
+E8-
yoj
input
interval
+3E1
0GBF,
&**
'03*3)
-,[**' *]
[*'!][&**]
%$)*#*%+("
.,[**][&*]
54['03*3)]!,#.2
64[] 2(1+3
74['03*3)]"0&"[]
/$!-%2 2
(1+3
%,.'.&
")$+-
"(#*.$+-
'.& OBS!
"(#*.$+-
'.& OBS
!
'.&
'.& %,.
%,.!
%.1)1(2$+3
[]$+
&1+/
%.1.0!
#",0'
&1+/
#",0'!
-*
[]$+
0
( $+
22
KWZPZO=[!IR\
[!]IR<>+&3A9-@KWZ=!#(/LZRX<
A@,
#(=/ 3A@,
•CRETZFUY[N\
•<5@QZHYN(Row%\
•<5@QZHYN[Col%\
•=QZHYN[%\
aBADb<>+IZDJN
/ 3A@,
SVG$[]\>IZDJN=
#B 5,*
3
=<5@#(/89@,
2=MZI>=?.<"':0@,
•[BAD=1\=.7827/4
1
)3A6[a==>1b\
•[BAD=0\=.7_`^/%89
;
)3A6*
32
BAD=1
827
362
1189
BAD=0
263
4518
4771
1080
4880
5960
FPSJSI;!TFPSPQADMU
[FPSPQA]DM:<FPSPQA;AOL/$
;%(3RKOQBR
;#3@$4,
3?>,2;AOL<+FPS
;)<+GSDCEH@$4,
FPS;JSITroot nodeU;29,
;)@=1;PQA<+
;@$4,5;:/>PQA
<+5?&;';@$4,17
;PQA:$ 3?>CANQH;03
<+5;CANQH:->"GSD
CEH;OBS:4>,CANQH<
':3?8.68.>29:
,*
BAD==>1BAD==>0
/68382$9.4:
;<.4#%/68")&"#08.*!($*
)($*' $+-,571)(
*
!*-$-#.'+ "&/
0'+ "1&&!*-(,")%
"$! '&%#
&
(&[]["$]"$
$)*#! "('
&*%+interval node,
&*%+root node,
$)*
&*%OBS
&*%+leaf node,&*
%
# &%5.&# [$)(15-,6/3][$)(15-04*-]# [
][+2625'] "! # :Over fitting;.+
#&
290%*(/69. %'')#+# %
290%179.!'
.+
SAS"&290493.#$)*.
+
290.290#290$
290"528.290"528%
.
%&
:*;$)*-,+
Y
A
y1
y0
y2
y3
A
A
A
A
B
B
x2
x1
A
no
X>x1
yes
no
Y>y3
X
yes
yes
no
no
X>x2
A
Y>y0
A
21[%/$"/,&/]["/,']#/%/$ +
*./)/$+!0$-!(
31[#//][#/%/$][#/%/$]
yes
B
B
Y>y2
A,B yes
#/%/$
Y>y1
B
A
B
A
no
B
yes
no
B
A
B
$"$!
&%#[]
[]60%[]40%[ ]0%
'%[$"$!]
$60%
$60%
B
DT?GTF/7I<NR
DT?GTF0/$*%3 •U
V//(1/DT? •UV5>=LRE*MDQ/4'3(1.&(MDQ4)
013DT? •UC=EVMDQ.2$-!#,"#A6B8'3(1/-C=EDT? (,",!
/.[(,",!]*,&,!'-!)%
0.[,",!][(,",!]
1.[(,",!][$][+ )#,]
:RIPR9/413
SSC=EDT?>B
E.2)37H;JT<NR
/413
OR@K:RIQ4'3+%.'3
OR@K<TF413
L6M
[DI?A]>C.:=<EJA;GB9-47/ @K>/@K>
2%;GB10,-"7!
;GB9*7/ 2%92.@K>23%5
0&0,-$6 @K>3%,-,-"7
(83 @K>+'1)-2"F@H%58-7/"#(
/106 (89/"#
%(71
L6M9#!
@K>
@K>
N
'02
54!&%-1(#/+ 9
64[
]['02] '02 74*2)[
]['02",$.13T4] ['
02]