Int
ern
at
i
onal
Journ
al of Ele
ctrical
an
d
Co
mput
er
En
gin
eeri
ng
(IJ
E
C
E)
Vo
l.
8
, No
.
6
,
Decem
ber
201
8
, p
p.
5425
~
5431
IS
S
N: 20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v8
i
6
.
pp
5425
-
54
31
5425
Journ
al h
om
e
page
:
http:
//
ia
es
core
.c
om/
journa
ls
/i
ndex.
ph
p/IJECE
Credit S
coring U
sing CA
RT Alg
orithm
and
Binary
Parti
cle
Swarm
Optimi
zation
Rez
a
Fir
sand
aya M
alik, He
rmawan
Facul
t
y
of
Com
pute
r
Sc
ie
nc
e,
Uni
ver
sita
s Sriwi
jaya
,
Indon
esia
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
A
ug
18
, 201
7
Re
vised
Ju
n
1
8
, 201
8
Accepte
d
J
ul
1
5
, 2
01
8
Credi
t
s
cor
ing
is
a
pro
ce
dur
e
th
a
t
exi
sts
in
eve
r
y
fina
nc
ia
l
insti
tution.
A
w
a
y
to
pre
dic
t
wheth
er
the
debt
or
wa
s
qual
ifi
ed
to
be
give
n
the
loa
n
or
not
and
has
bee
n
a
m
aj
o
r
concern
in
the
over
all
steps
of
t
he
loa
n
proc
ess.
Alm
ost
al
l
banks
and
othe
r
fina
nc
ia
l
inst
it
ut
ions
have
their
own
cre
dit
scorin
g
m
et
hods.
Now
aday
s,
data
m
ini
ng
appr
oa
c
h
has
bee
n
accepted
to
be
on
e
of
the
we
ll
-
known
m
et
hods.
Certainly
,
ac
cur
acy
was
al
so
a
m
aj
or
issue
in
th
is
appr
oa
ch
.
Thi
s
rese
arc
h
pr
oposed
a
h
y
br
id
m
et
hod
using
CART
al
g
or
it
h
m
and
Binar
y
Parti
cle
Sw
arm
Optimiza
ti
o
n.
P
erf
orm
anc
e
ind
i
ca
tors
that
ar
e
used
i
n
thi
s
rese
arc
h
ar
e
cl
a
ss
ifi
ca
ti
on
a
cc
ur
acy
,
err
or
ra
te,
sensiti
vity
,
spe
c
ifi
cit
y
,
an
d
pre
ci
sion
.
Expe
r
imenta
l
resul
ts
base
d
on
the
publi
c
dataset
sho
wed
tha
t
the
proposed
m
et
hod
ac
cur
a
c
y
is
78
%
and
87.
53
%
.
In
compare
to
seve
ral
popula
r
al
gor
it
h
m
s,
such
as
neu
ral
net
work
,
log
isti
c
reg
r
ession
and
support
vec
tor
m
ac
hin
e,
the
proposed
m
e
thod
show
ed a
n
outsta
nding
per
f
orm
anc
e.
Ke
yw
or
d:
CART
Credit Sc
or
i
ng
Data M
inin
g
PSO
Copyright
©
201
8
Instit
ut
e
o
f Ad
vanc
ed
Engi
n
ee
r
ing
and
S
cienc
e
.
Al
l
rights re
serv
ed
.
Corres
pond
in
g
Aut
h
or
:
Her
m
awan
,
Faculty
of Com
pu
te
r
Scie
nc
e,
Un
i
ver
sit
as
Sr
i
wij
ay
a.
South
Su
m
at
era
-
Ind
on
e
sia
Em
a
il
: her
m
aw
an@
m
dp
.ac
.id
1.
INTROD
U
CTION
Credit
sco
rin
g
is
a
par
ti
cular
j
ob
of
loa
n
li
fecyc
le
m
anage
m
ent
that
had
bee
n
a
big
chall
enge.
It
pr
e
dicts
wh
et
he
r
the
deb
t
or
is
qual
ifie
d
to
be
gi
ven
a
l
oan
or
no
t.
Cre
dit
sc
or
i
ng
is
te
rm
that
us
ed
to
de
s
cribe
form
al
m
et
ho
ds
us
e
d
f
or
cl
a
ssifiy
ing
a
pp
li
can
ts
f
or
c
red
i
t
into
good
cr
edit
or
bad
c
r
edit
cl
asses.
I
nd
ee
d,
wrong
pr
e
dicti
on
will
be
a
gr
eat
loss
to
ba
nk
s
a
nd
finan
ci
al
ins
ti
tuti
on
.
T
her
e
are
two
ty
pe
s
of
m
isc
la
ssific
at
i
on
patte
rn
w
hi
ch
is
cal
le
d
typ
e
I
an
d
ty
pe
II
er
ror
[1]
.
T
ype
I
error
oc
cur
s
wh
e
n
the
act
ually
good
cre
dit,
bu
t
la
te
r
was
no
t
acce
pted
a
nd
c
la
ssifie
d
as
ba
d
cre
dit
wh
ic
h
will
red
uce
t
he
insti
tuti
on
’s
pro
fit.
As
the
op
po
sit
e, ty
pe
I
I
e
rror
occurs
w
hen th
e act
ually
b
ad
cred
it
bu
t l
at
er
was
cl
assifi
e
d as g
ood
c
re
dit.
Th
us
,
it
will
br
ing
a
big
prob
le
m
and
seri
ou
s
dam
a
ge
to
the
i
ns
ti
tuti
on
[
1]
.
W
it
h
the
incr
easi
ng
i
m
po
rtance
of
cred
i
t
scor
i
ng
t
o
ba
nk
an
d
fina
ncial
insti
tuti
on
,
th
is
fiel
d
has
i
nvoke
d
intere
sts
to
m
any
research
e
r
to
wor
k
on
it
.
This
researc
h
area
has
been
cond
ucted
by
m
any
researc
he
rs
ov
e
r
ye
ar
s
with
s
o
m
any
m
et
ho
ds.
On
e
of
the
ver
y
popula
r
m
et
ho
d
is
the
data
m
ining
a
ppr
oac
h.
Data
m
ining
has
e
nt
ic
ed
a
gr
eat
im
portance
of
i
nterest
i
n
the infor
m
at
ion
industry in r
e
cent ye
ars
that fo
c
us
e
d
on the ex
tract
io
n
of h
i
dd
e
n
kn
ow
le
dge from
v
arious d
at
a
war
e
hous
e
,
da
ta
set
,
and
data
re
po
sit
ori
es
[2]
.
T
his
appr
oach
is
a
big
help
to
ba
nk
a
nd
ot
he
r
fina
ncial
instit
utions.
So
m
e
po
pula
r
m
et
ho
ds
that
ha
d
bee
n
use
d
by
so
m
e
researc
her
a
re
cl
assifi
cat
ion
an
d
re
gressi
on
tree
(CART
),
S
upport
Vecto
r
Ma
chine
(
SVM
),
A
rtific
ia
l
Neural
Net
work
(AN
N),
Mult
ivariat
e
Ad
a
ptive
Re
gr
essi
on
S
plines
(M
ARS)
[
3]
.
P
rev
i
ou
sly
,
researc
he
rs
ha
ve
us
ed
pri
vat
e
dataset
to
e
xplo
re
cre
dit
sc
or
i
ng.
Fo
r
e
xam
ple,
T.
S.
Lee,
Chi
u,
Ch
ou,
an
d
Lu
ha
ve
em
pl
oyed
CART
a
nd
m
ulti
v
ariat
e
adap
ti
ve
re
gressi
on
sp
li
nes
(MAR
S)
t
o
pr
i
vate
c
red
it
ca
rd
local
ba
nk
in
Tai
pe
i,
Tai
wa
n.
Ex
pe
rim
ental
sh
owed
that
c
om
par
ed
to
sever
al
al
go
rithm
s,
sti
l
l
CAR
T
and
M
ARS
hav
e
a
bette
r
overall
pe
rfor
m
ance
[
4]
.
A
no
t
her
e
xam
ple,
W.
Che
n
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
8
, N
o.
6
,
Dece
m
ber
2
01
8
:
5425
-
5431
5426
et
al
hav
e
pro
po
s
ed
hybr
i
d
m
et
ho
d
S
VM
+
CART
an
d
SV
M
+m
ulti
v
ariat
e
adap
ti
ve
regressio
n
spl
ine
s
(MARS
)
f
or
t
he
ir
pr
i
vate
dat
aset
bank
of
C
hin
a.
Thei
r
res
ults
showe
d
a
n
i
m
pr
ovem
ent
in
t
erm
of
accuracy
us
in
g
hybri
d
m
et
ho
d
[
5]
.
A
no
t
her
resea
rc
her
us
e
d
publi
c
datase
t
f
or
t
heir
e
xperim
ent.
J.
C
hen
us
e
d
ger
m
an
cred
it
dataset
and
Austral
ia
n
fr
om
Un
ive
rsity
of
Ca
li
fo
r
nia
(U
CI
)
re
po
si
tory.
He
propo
sed
a
hybri
d
m
et
hod
cal
le
d
SV
M
+
wh
it
eni
ng
s
pac
e.
His
m
et
ho
d
sh
owe
d
a
n
im
pr
ovem
ent
com
par
e
d
t
o
S
VM
[
6]
.
Se
ver
al
ap
proac
h
us
in
g
e
ns
em
bles
of
cl
assi
fier
has
bee
n
a
ppli
ed
t
o
c
red
it
sc
ori
ng,
su
c
h
a
s
ba
gg
i
ng,
boos
ti
ng,
rand
om
su
bs
pa
ce,
and
decorate
. T
he
base classi
fier conside
red in the ex
per
im
ental
stud
y al
ong wit
h
the en
s
e
m
bled
m
et
ho
ds
are:
l
og
ist
ic
re
gr
ess
ion
(L
og
R
),
m
ulti
la
ye
r
per
ce
ptr
on
(MLP
),
s
upport
vecto
r
m
achines
(
SVM
),
C4
.5
decis
ion
tr
e
e
(C4.5
)
an
d
cre
dal
decisi
on
tr
ee
(CDT).
Fro
m
the
resu
lt
,
cred
it
decisi
on
tree
as
the
bas
e
cl
assifi
er
ha
s
the
bette
r
resu
lt
,
wh
e
n
it
is
us
e
d
as
base
cl
a
ssifie
r,
i
n
a
en
sem
bled
sche
m
e
fo
r
cre
dit
scor
i
ng
assess
m
ent
[7]
.
Alm
os
t
all
research
e
r
w
orks
ha
ve
f
ocused
th
ei
r
researc
h
on
increasin
g
the
accuracy
of
cr
edit
scor
i
ng,
suc
h
a
s
Yao
Pi
ng,
Lu
Yonghe
ng
w
ho
pro
p
os
e
d
SV
M
+
Neig
hbor
hood
Roug
h
Set
and
c
om
par
ed
it
wit
h
LD
A,
Lo
gisti
c
regres
sion,
N
eu
ral
N
et
work.
Re
s
ult
show
s
that
t
he
ir
pro
pose
d
m
e
thod
gain
a
n
im
pr
ov
em
ent
in
te
rm
of
acc
ur
acy
[8]
.
So
m
e
research
ers
,
f
ocused
on
cat
c
hing
“b
ad”
cre
ditors
as
an
im
po
rtanc
e
perform
ance
issue
,
with
thei
r
propose
d
m
et
ho
d
Kernel
F
uzzifi
cat
ion
Pe
nalty
-
MC
OC
[
9]
.
O
ther
researc
he
r
s,
f
oc
us
e
d
thei
r
w
ork
on tim
e reducti
on for cre
dit sc
or
i
ng, such
as
Ba
ndhu & K
um
ar.
Thei
r
w
or
k based
on a
n appr
oach cal
le
d
S
VM
+
F
Sc
or
e
sam
pling
t
o
re
duce
d
com
pu
ta
ti
on
al
tim
e
fo
r
cre
dit
sco
rin
g
an
d
com
par
ed
it
w
it
h
SV
M
+
GA,
Ba
ck
Pr
opa
gatio
n
and
Gen
et
ic
Program
m
ing
.I
t
is
pr
ove
d
that
their
m
et
ho
d
is
co
m
petit
iv
e,
in
the
view
of
it
s
accuracy
a
s
w
el
l
as
the
pro
pose
d
m
et
ho
d
ha
s
a
le
ss
com
pu
ta
ti
on
al
ti
m
e
[10]
.
A
nothe
r
i
ssu
e
is
an
im
balance
dataset
s
that
be
ca
m
e
gr
eat
c
on
ce
r
n
by
H
onglian
g
He
et
al
,
that
they
f
oc
us
e
d
their
res
earch
on
a
dap
t
ion
of
diff
e
re
nt
i
m
b
al
ance
rati
os
a
nd
pro
posed
t
heir
novel
m
e
thod
to
ob
ta
i
n
supe
rio
r
pe
rfor
m
ance
an
d
high
rob
us
tness
[
11
]
.
In
this
pa
per,
we
pro
po
s
ed
hy
br
id
Cl
assifi
cat
ion
a
nd
Re
gressi
on
T
ree
(CART)
an
d
Bi
na
ry
Pa
rtic
le
Sw
arm
Op
ti
m
i
zat
ion
.
CART
is
well
kn
ow
n
sp
eci
fic
dec
isi
on
tree
al
go
rithm
.
It
is
us
ed
in
seve
ral
kinds
’
app
li
cat
io
n
of
data
m
ining
,
s
uch
as
web
m
ining,
ed
ucati
onal
m
ining
,
m
edical
m
ining
,
and
c
red
it
sc
or
i
ng.
Ma
ny
resea
rchers
hav
e
em
plo
ye
d
C
ART
i
n
their
in
vestigat
ion.
On
e
of
their
stu
dy
us
in
g
pr
i
vate
dataset
con
cl
ud
e
that
com
par
e
to
som
e
oth
er
popu
la
r
intel
li
gen
t
m
et
ho
ds
s
uch
as
SV
M
an
d
N
eur
al
Net
work,
CART
sh
ows
a
bette
r
perform
ance
in
cred
it
sco
rin
g
in
te
rm
of
AUC
m
easur
e
[
12
]
.
CART
has
be
en
adm
it
te
d
as
one
of
t
op
10
data
m
ining
al
gorithm
and
one
of
t
he
m
os
t
influ
e
ntial
data
m
ining
al
go
rithm
[13]
.
I
n
c
ontrast
,
Bi
nar
y
Partic
le
Sw
arm
Op
t
i
m
iz
ation
(BP
SO
)
a
s
one
of
var
ia
nt
of
PSO
is
us
e
d
to
increase
ov
e
rall
perform
ance o
f
CART.
Partic
le
Sw
ar
m
Op
tim
iz
at
io
n
is
an
al
gorithm
,
a
kin
d
of
cal
culat
ion
m
e
thod
ba
sed
on
the
theo
ry
of
swar
m
intel
li
gen
ce, a
nd
a
kind of m
od
el
in
the f
ie
ld
of
swa
rm
intelli
gen
ce that
r
et
ai
ns
a gl
ob
al
searc
h
str
at
eg
y
base
d
on
po
pula
ti
on
of
swa
r
m
[14]
.
W
it
h
P
SO
,
t
he
pro
ble
m
is
so
lved
an
d
ad
dr
e
ssed
usi
ng
s
war
m
of
par
ti
cl
e
that
m
ov
e
ar
ound
the
s
wa
rm
,
lookin
g
f
or
the
b
est
possible
so
luti
on
[
15]
.
T
her
e
a
re
so
m
e
adv
a
ntage
s
of
us
in
g
PSO
s
uc
h
as,
it
do
es
not
nee
d
di
ff
e
ren
ti
at
io
n
unli
ke
m
any
tradit
ion
al
m
eth
od,
a
nd
it
has
the
abili
ty
to
escap
e
from
lo
cal
op
tim
i
m
u
m
.
Another
a
dvanta
ge
s
are
PS
O
ha
s
flexibili
ty
to
integrate
with
ot
her
opti
m
i
zat
io
n
te
chn
iq
ues
in
order
to
dev
el
op
c
om
plex
too
ls
and
it
can
be
us
e
d
f
or
th
e
obj
ect
ive
f
unct
ions
with
r
andom
natu
re,
sim
i
la
r
to
the
case
th
at
on
e
of
the
optim
iz
at
ion
var
ia
bles
is
rand
om
.
No
t
to
m
e
ntion
t
he
fact
tha
that
PSO
has
le
ss
s
ensiti
vity
to
th
e
obj
ect
iv
e
f
un
ct
ion
’s
nat
ur
e
,
wh
ic
h
m
eans
it
has
c
onve
xity
or
c
onti
nuit
y
[16]
.
Bi
nar
y
PS
O
i
s
var
ia
nt
of
Partic
le
Sw
a
r
m
Op
ti
m
iz
at
io
n.
It
is
a
nat
ur
e
i
ns
pi
red
a
lgorit
hm
,
as
well
as
m
et
aheu
risti
c
global
opti
m
izati
on
al
gorith
m
,
or
iginall
y
pro
posed
by
Ke
nn
e
dy
a
nd
E
be
rh
a
rt.
A
ty
pe
of
bio
-
insp
ire
d
opti
m
i
zat
ion
al
gorith
m
insipired
by
m
ov
e
m
ent
of
bi
rd
s
an
d
fis
h
fl
ock
wh
il
e
sear
chin
g
f
or
f
ood
[17]
.
PSO
s
ol
ution
s
wam
is
co
m
par
ed
t
o
the
bir
d
swar
m
,
the
bir
ds
’
m
ov
in
g
fro
m
on
e
place
to
ano
t
her
is
e
qu
al
to
the
de
velo
pm
e
nt
of
the
s
olu
ti
on
s
war
m
,
good
inf
or
m
at
ion
is
equ
al
to
the
m
os
t
op
tim
ist
so
luti
on,
a
nd
t
he
f
ood
resou
rce
is
equ
al
to
the
m
os
t
op
tim
ist
so
lu
ti
on
durin
g
the
wh
ole
co
urse
[18]
.T
his
m
et
ho
d
has
bee
n
use
d
to
sever
al
resea
rc
h
are
a.
It
is
use
d
to
cl
assify
high
dim
ension
al
ed
ucati
ona
l
data
with
go
od
pe
rfor
m
anc
e
res
ult
com
par
e to se
ve
ral al
gorithm
s.
Ot
her
resea
rc
her, em
bed
de
d t
his m
et
ho
d
wi
th S
VM to
a
nal
yz
e o
pi
nion m
i
ning
of
so
ci
al
m
edia.Their
stu
dy
s
howe
d
a
good
r
esult,
P
SO
af
fe
ct
s
the
acc
ur
ac
y
of
S
VM
after
the
hybri
dizat
ion
of
SV
M
-
PS
O
[
19]
,
[20]
.
Ba
se
d
on
li
te
ratu
re
stud
y,
t
his
m
e
tho
d
ca
n
al
so
b
e u
se
d
to
im
pr
ove
ov
e
rall
perform
ance
of CART al
gor
it
h
m
.
2.
RESEA
R
CH MET
HO
D
Figure 1
s
hows
f
lowc
har
t
of
prop
os
ed resear
ch
desi
gn. Th
e
fo
ll
owin
g
fl
owchar
t co
ns
ist
s
of
se
qu
e
nc
e
of
ste
ps
a
nd
m
et
hods
to
do
t
he
resea
rch.
It
exp
la
in
s
the
pr
ocess
of
co
nduc
ti
ng
this
ex
pe
rim
ental
research
in
m
or
e
detai
ls.
Re
searche
rs
w
il
l
fo
ll
ow
t
hes
e
ste
ps
wh
il
e
do
i
ng
resea
rch
to
en
sure
the
integrity
of
th
e
w
ho
le
researc
h proce
ss.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Credit
S
c
or
in
g Usin
g
C
ART
Algo
rit
hm
and
Bina
ry
P
ar
ti
cl
e S
w
ar
m
Opti
miza
ti
on
(
Rez
a Fi
rsan
da
y
a
M
alik
)
5427
S
T
A
R
T
D
a
t
a
C
o
l
l
e
c
t
i
o
n
:
G
e
r
m
a
n
.
d
a
t
a
-
n
u
m
e
r
i
c
A
u
s
t
r
a
l
i
a
n
d
a
t
a
s
e
t
C
l
a
s
s
i
f
i
c
a
t
i
o
n
T
a
s
k
:
1
.
C
A
R
T
a
l
g
o
r
i
t
h
m
2
.
C
A
R
T
+
B
P
S
O
a
l
g
o
r
i
t
h
m
V
a
l
i
d
a
t
i
o
n
a
n
d
E
v
a
l
u
a
t
i
o
n
:
1
.
1
0
F
o
l
d
V
a
l
i
d
a
t
i
o
n
2
.
C
o
n
f
u
s
i
o
n
M
a
t
r
i
x
L
i
t
e
r
a
t
u
r
e
R
e
v
i
e
w
:
C
r
e
d
i
t
S
c
o
r
i
n
g
A
n
a
l
y
z
e
R
e
s
u
l
t
:
1
.
C
o
m
p
a
r
e
i
n
t
e
r
n
a
l
r
e
s
u
l
t
2
.
C
o
m
p
a
r
e
t
o
o
t
h
e
r
m
e
t
h
o
d
E
N
D
P
e
r
f
o
r
m
a
n
c
e
M
e
a
s
u
r
e
m
e
n
t
:
1
.
M
e
t
r
i
c
s
2
.
R
O
C
C
u
r
v
e
Figure
1.
Pro
pose
d
R
esearc
h Desig
n
Re
search
bega
n
with
c
ollec
ti
ng
li
te
ratu
re
f
r
om
few
resource
s.
A
li
te
rature
search
c
ondu
ct
ed
befo
re
procee
ding
to
desig
n
ex
per
i
m
ent.
This
ste
p
pro
vid
es
fou
nd
at
io
nal
knowle
dge
ab
ou
t
the
researc
h
ar
ea,
the
desig
ns
,
in
strum
ents
us
ed,
th
e
procedure
a
nd
the
fin
d
in
gs
.
The
inf
orm
ation
disc
ov
e
red
duri
ng
this
ste
p
help
s
the
resea
rc
her
s
f
ully
underst
and
the
m
agn
it
ud
e
of
pro
blem
.
All
m
at
eria
ls
wer
e
capt
ured
a
nd
ext
racted
in
t
o
researc
h
m
appi
ng
.
L
at
er,
we
decide
d
to
us
e
public
dataset
.
Re
al
world
c
red
it
dataset
,
Ger
m
an.
d
at
a
-
num
eri
c
dataset
and
A
ust
rali
an
dataset
are
us
ed
as
a
n
obj
ect
to
our
researc
h.
Co
nsi
der
in
g
the
fa
ct
that
based
on
our
li
te
ratur
e
rev
ie
w,
th
os
e
datas
et
s
wer
e
gener
al
ly
us
ed
by
research
e
r
in
th
e
research
a
re
a.
The
Dataset
s
are
avail
able
fro
m
the
Un
ive
rsit
y
of
Ca
li
fo
r
nia
(U
C
I)
Re
po
sit
ory
of
m
achine
le
a
rn
i
ng
data
bas
es.Th
e
Ger
m
an.
data
-
num
eric
dataset
con
sist
s
of
24
pr
e
dictor
at
tri
bu
te
s
a
nd
1
ta
r
get
at
tribu
te
[21]
.
Total
num
ber
of
instances
are
1000.
T
her
e
are
700
instan
ces
are
la
beled
as
cred
it
w
or
t
hy,
and
300
instanc
es
are
la
bed
e
d
as
no
t
cred
it
w
or
t
hy.
Au
st
rali
an
data
set
con
sist
of
14
predict
or
at
tr
ibu
te
s
a
nd
1
ta
rg
et
at
trib
ute.T
her
e
a
re
total
ly
690
instances
i
n
A
us
tral
ia
n
datas
et
,
co
ns
ist
s
of
307
i
ns
ta
nces
are
la
bel
d
cr
editwo
rthy,
a
nd
383
in
sta
nc
es
ar
e
la
beled
as
no
t
cred
it
w
or
t
hy
[21]
.
Ta
ble
1
f
ur
t
her
desc
ribe
s
detai
ls
of
these
dataset
s.
Th
e
wo
r
k
of
rese
arch
is
con
ti
nue
d
by
cond
ucted
the
cl
assifi
cat
ion
ta
sk
with
C
AR
T
al
gorithm
a
nd
t
he
pro
pos
ed
m
et
ho
d
(C
ART
+
BPSO
).
The
ex
per
im
ental
pro
cedures
will
be
car
ried
out
in
this
phase
.
T
he
n
10
-
f
ol
d
valid
at
ion
a
nd
c
onf
us
io
n
m
at
rix
are
use
d
to
trai
n
ou
r
cred
it
sco
rin
g
m
od
el
.
So
m
e
m
et
rics
are
us
e
d
to
m
eas
ur
e
perform
ance
of
cl
assifi
er.
Me
tric
s
for
e
valua
ti
ng
cl
assifi
er
pe
rfor
m
ance
ar
e
accuracy,
e
rror
rate,
sen
sit
ivit
y,
sp
eci
fici
ty
,
and
pr
eci
sio
n.O
veral
l
per
f
or
m
ance
is
sh
owe
d
in
Re
cei
ver
O
pe
rati
ng
C
har
act
erist
ic
(ROC)
Curve
an
d
a
re
a
unde
r
curve (
A
UC)
of
ROC
[
22]
,
[
23]
.
At
la
st, our
exp
e
rim
ental
resu
l
t
is
analy
ze
d
a
nd
c
om
par
e
d
to
t
he
oth
er
s
i
m
i
la
r
m
et
ho
d o
f data
m
ining
.
Table
1.
Detai
ls o
f data
set
s
Dataset
No
.
attribu
t
No
.
Ins
tan
ces
Ger
m
an
.data
-
n
u
m
eric
25
1000
Au
stralian
15
690
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
8
, N
o.
6
,
Dece
m
ber
2
01
8
:
5425
-
5431
5428
R
a
n
d
o
m
l
y
i
n
i
t
i
a
l
i
z
e
p
o
p
u
l
a
t
i
o
n
p
o
s
i
t
i
o
n
s
a
n
d
v
e
l
o
c
i
t
i
e
s
E
v
a
l
u
a
t
e
F
i
t
n
e
s
s
o
f
P
a
r
t
i
c
l
e
u
s
i
n
g
C
A
R
T
A
l
g
o
r
i
t
h
m
I
f
p
a
r
t
i
c
l
e
f
i
t
n
e
s
s
>
p
a
r
t
i
c
l
e
b
e
s
t
f
i
t
n
e
s
s
U
p
d
a
t
e
b
e
s
t
p
a
r
t
i
c
l
e
I
f
p
a
r
t
i
c
l
e
f
i
t
n
e
s
s
>
g
l
o
b
a
l
b
e
s
t
f
i
t
n
e
s
s
U
p
d
a
t
e
g
l
o
b
a
l
b
e
s
t
Y
e
s
T
e
r
m
i
n
a
t
i
o
n
?
N
o
U
p
d
a
t
e
p
a
r
t
i
c
l
e
v
e
l
o
c
i
t
y
U
p
d
a
t
e
p
a
r
t
i
c
l
e
p
o
s
i
t
i
o
n
O
p
t
i
m
i
z
e
d
P
a
r
a
m
e
t
e
r
s
a
n
d
f
e
a
t
u
r
e
s
u
b
s
e
t
Figure
2
.
Pro
pose
d
Me
th
od
Bi
nar
y
PSO
a
ppr
oach
is
us
e
d
as
featu
re
se
le
ct
ion
m
e
tho
d
to
sel
ect
best
su
bs
et
that
pr
oduce
best
perform
ance.
BPSO
is
an
ex
te
nd
e
d
al
gorith
m
of
Partic
le
S
war
m
Op
tim
izati
on
that
op
e
r
at
es
on
bin
a
ry
searc
h
sp
ace.
Each
pa
rtic
le
represe
nts
posit
ion
in
bi
nar
y
s
pa
ce
a
nd
pa
rtic
le
’s
posi
ti
on
ca
n
ta
ke
on
the
bi
nar
y
va
lue
0
or
1.
Fi
gure
2
s
hows
th
e flowc
har
t
of
pro
po
se
d
m
et
ho
d.
It b
e
gin
s
with r
a
nd
om
l
y i
niti
alize
par
ti
cl
e. Popul
at
ion
of
par
ti
cl
es
ar
e
create
d,
an
d
each
par
ti
cl
e
is
correla
te
d
with
ge
ne
rated
so
luti
on.
All
par
ti
cl
e’s
fitn
ess
is
evaluate
d.
Thi
s
exp
e
rim
ental
stud
y
us
ed
C
ART
cl
assifi
cat
ion
accu
racy
as
the
fitness
f
un
ct
io
n.
Ba
se
d
on
the
resu
lt
,
the
ne
xt
ste
p
is
to
ev
al
uate
par
ti
cl
e’s
pbest
an
d
gb
e
st.
Fo
ll
ow
e
d
by
update
pa
rtic
le
velocit
y
an
d
sigm
oid
functi
on.
Co
ns
tr
uction
phase
le
t
par
ti
cl
es
m
ov
e
to
an
oth
e
r
pote
ntial
so
luti
on
base
d
on
i
ts
own
exp
e
rience
a
nd
that
of
nei
ghbor
.
The
l
oop
e
nded
with
a
sto
pp
i
ng
crit
eria
in
te
r
m
inati
on
phas
e
that
pr
e
determ
ined bef
or
e
[
24
]
[
25]
.
3.
RESU
LT
S
A
ND AN
ALYSIS
Ex
per
im
ental
resu
lt
is
com
par
ed
i
n
tw
o
pha
se
or
pa
rt.
Firs
t
ph
ase
,
an
int
ern
al
e
xp
e
rim
e
ntal
res
ult
is
com
par
ed
eac
h
ot
her.
Pe
rform
ance
of
c
re
di
t
scor
in
g
us
in
g
CART
al
gor
it
h
m
is
co
m
par
ed
t
o
cre
dit
s
cor
i
ng
us
in
g
CART+
PSO
al
gorit
hm.
Seco
nd
ph
as
e,
we
com
par
e
d
pro
posed
m
et
hod
to
sim
ilar
researc
h.
T
able
2
sh
ows
the
first
ph
a
se c
om
par
ison res
ult.
Table
2
.
C
om
par
iso
n
Re
s
ult o
f
CART
and C
ART+B
PS
O
Metr
ic
Ger
m
an
.data
-
n
u
m
eric
d
ataset
Au
stralian
datas
et
CAR
T
CAR
T+BPSO
CAR
T
CAR
T+BPSO
Accurac
y
(
%)
7
5
.2
78
8
5
.36
8
7
.53
Er
ror
r
ate
(
%
)
2
4
.8
22
1
4
.64
1
2
.47
Sen
sitiv
ity
(
%
)
8
9
.1
9
1
.71
8
4
.04
8
6
.97
Sp
ecif
icity
(
%
)
4
2
.7
46
8
6
.42
8
7
.99
Precisio
n
(
%)
7
8
.4
7
9
.85
8
4
.04
8
5
.30
AUC
0
.71
9
6
0
.73
9
2
8
7
.71
0
.90
3
4
No
.
o
f
Attr
ib
u
te us
ed
24
11
14
6
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Credit
S
c
or
in
g Usin
g
C
ART
Algo
rit
hm
and
Bina
ry
P
ar
ti
cl
e S
w
ar
m
Opti
miza
ti
on
(
Rez
a Fi
rsan
da
y
a
M
alik
)
5429
Table
3
.
C
om
par
iso
n resu
lt
to
o
the
r researc
hs
No
.
Metho
d
s [
Ger
m
an
.
d
ata
-
n
u
m
eric
data
set]
Accurac
y
%
1
Su
p
p
o
rt
Vector M
achi
n
e (
SV
M)
7
5
.98
2
SVM
+
W
h
iten
in
g
T
rans
f
o
r
m
atio
n
(
W
T
)
7
6
.88
3
Linear
Disri
m
in
an
t Analysis
6
6
.60
4
Log
istic Reg
ressio
n
7
2
.40
5
Neu
ral
N
etwo
rk
7
5
.20
6
SVM
+ Neigh
b
o
rho
o
d
Ro
u
g
h
Set
7
6
.60
7
Multi
-
Crite
ria
Op
tim
i
zatio
n
Class
if
ier
(M
COC
)
73
8
Kernel Fu
zzif
icati
o
n
Penalty
–
MC
O
C
7
3
.40
9
SVM+
Genetic
Al
g
o
rith
m
7
6
.84
0
Back
Prop
ag
atio
n
7
6
.69
1
Gen
etic Pr
o
g
ra
m
m
in
g
7
7
,26
2
Decorate
+ log
R (e
n
se
m
b
le
)
7
7
.40
3
Bag
g
in
g
+
SVM
(
e
n
se
m
b
le
)
7
6
.60
4
CAR
T
+ B
PSO
(P
rop
o
sed
M
eth
o
d
)
78
Figure
3
.
Acc
uracy
co
m
par
iso
n
c
har
t
Table
2
sho
w
s
the
ov
er
al
l
perform
ance
of
pro
po
se
d
m
et
hod
(BPS
O+
CART)
com
par
ed
to
ba
se
m
et
ho
d
(CAR
T).
It
is
cl
ear
that
there
is
rem
ark
able
im
pro
vem
ent
in
the
pro
pose
d
m
et
ho
d.
Per
f
orm
ance
sh
ows
an
incre
ase
in
te
rm
of
accuracy,
the
a
ccur
acy
is
rais
ed
f
r
om
75
.
2%
to
78
%
with
Ger
m
an.
data
-
num
eric
dataset
and
85.
36%
to
87.53
%
with
Austral
ia
n
dataset
.
In
te
rm
of
error
r
at
e,
propose
d
m
et
ho
d
s
hows
a
bette
r
perform
ance.
Anothe
r
in
dicat
or
of
i
m
pr
ove
m
ent,
the
area
unde
r
curve
(
AU
C
)
of
our
pro
posed
m
et
ho
d
valu
e
is
0.
73
92
with
Ger
m
an.
data
-
nu
m
eric
dataset
and
0.9
034
with
A
us
tral
ia
n
dataset
,
w
hi
ch
are
hi
gh
e
r
than
the
base
le
arn
e
r
m
et
hod.
Ex
pe
rim
ental
resu
lt
a
lso
shows
that
featur
e
sel
ect
ion
do
es
a
ff
ect
ov
e
rall
perfor
m
ance.
Feat
ur
e sele
ct
ion
is an
im
po
rtance task to
im
pro
ve
th
e p
re
di
ct
ion
accu
racy of
the h
y
br
i
d
m
od
el
. Classi
fi
cat
ion
pro
blem
s
gen
erall
y
inv
ol
ve
a
nu
m
ber
of
fea
tures
or
at
tri
bute
.
Howe
ver,
not
al
l
of
t
hese
featur
e
s
are
e
qual
ly
i
m
po
rtant
for
c
la
ssific
at
ion
ta
sk
.
So
m
e
of
th
ese
featu
res
a
r
e
not
rele
van
t
and
re
duda
n.
O
ur
pro
po
se
d
m
et
ho
d
search
f
or
the
m
os
t
i
m
po
rtan
ce
featur
es
f
rom
the
search
sp
ace
(all
featu
res).
CART
+
BPSO
m
et
ho
d
us
ed
60.00
0%
62.00
0%
64.00
0%
66.00
0%
68.00
0%
70.00
0%
72.00
0%
74.00
0%
76.00
0%
78.00
0%
Supp
ort V
ector
Machi
ne
SVM
+ W
hiten
i
n
g Tran
sfor
mati
on
Li
near
Disr
i
mi
nant A
na
l
y
sis
Log
i
sti
c
R
eg
res
sio
n
Neu
ral
Ne
twor
k
SVM
+ Neig
hbor
hood Ro
ugh Se
t
Mul
ti-Cr
i
t
eria
Opti
miz
ation C
l
a
s
si
fie
r
Kern
el
F
u
zzifi
cat
i
on
Pen
alty
- MC
OC
SVM+
G
e
netic
Alg
orit
hm
BackP
rop
agat
i
on
Ge
net
i
c
Prog
ram
m
in
g
De
corate
+ log
R
Bagg
i
n
g +
S
V
M
Clas
sif
i
c
ation
and R
egres
si
on Tree
CA
RT +
Par
ticle
Swa
rm Op
tim
i
za
tion
A
ccurac
y
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
8
, N
o.
6
,
Dece
m
ber
2
01
8
:
5425
-
5431
5430
on
ly
11
from
24
at
trib
utes
and
6
from
14
at
tribu
te
s.
T
he
pr
op
os
e
d
m
e
t
hod
ch
oose
th
e
best
at
tribu
te
that
con
t
rib
ute
the
m
os
t
to
increase
ov
e
rall
per
f
or
m
ance.
N
ot
to
fo
r
get,
the
aver
a
ge
ex
ecuti
on
ti
m
e
of
ou
r
pro
po
se
d
m
et
h
od
is
ab
out
te
n
m
inu
te
s.
Ter
m
of
execu
ti
on
or
com
pu
ta
ti
onal
tim
e
is
the
nex
t
bi
g
chall
e
ng
e
to
our
resear
ch
,
si
nce
s
pee
d
has
a
great
im
po
rtance
i
n
the
21st centu
ry.
T
he
le
ss
com
pu
ta
ti
on
al
tim
e
m
eans
m
or
e
eff
ic
ie
nt a
nd m
or
e
b
e
nef
it
t
o
t
he ban
k
a
nd in
du
st
ry.
The
n
we
m
easur
e
d
an
d
c
om
par
ed
our
ex
pe
r
i
m
ent
resu
lt
w
it
h
ano
t
her
si
m
il
ar
m
e
tho
d
and
researc
h.
Figure
3
s
how
s
that
c
om
par
e
to
se
ver
al
wel
l
-
know
n
arti
fic
ia
l
intel
li
gen
t
and
p
opula
r
al
gorithm
,
our
pr
opos
e
d
m
et
ho
d
s
hows
an
outst
and
i
ng
res
ult
with
78
%
acc
ur
ac
y.
Accuracy
le
vel
wh
ic
h
is
hig
he
r
tha
n
Neural
Netw
ork
al
gorithm
, G
eneti
c algori
thm
an
d S
upport
Vecto
r M
achine.
4.
CONCL
US
I
O
N
In
this
cre
dit
scor
i
ng
re
sear
ch,
we
e
xp
l
ore
an
appro
ac
h
to
increase
th
e
per
f
or
m
ance
of
our
ba
se
le
arn
er
al
gorithm
.
CART
al
go
rithm
is
cho
ose
d
as
a
base
le
arn
e
r,
since
it
is
on
e
of
the
best
al
gorithm
s
that
is
m
os
tly
us
ed
f
or
t
he
cl
assifi
cat
ion
ta
s
k.
B
inary
P
a
rtic
le
Sw
arm
Op
ti
m
iz
at
ion
is
a
dopted
t
o
inc
rea
se
the
perform
ance
of
CART
al
gorit
hm
.
The
propo
sed
m
et
ho
d
is vali
dated
with
r
eal
pu
blic
cre
dit
dataset
. The
res
ult
sh
ows
a
n
over
al
l
i
m
pr
ov
em
e
nt
of
ou
r
ex
perim
ent.
Ba
sed
on
seve
ral
ind
ic
at
or
s,
the
pro
pose
d
m
e
th
od
s
hows
a
bette
r per
form
ance, suc
h
as
a
ccur
acy
,
er
ror r
at
e, sen
sit
ivit
y,
sp
eci
fici
ty
and
preci
sion.
Com
par
ed
to
ano
t
her
resear
ch,
our
pr
opose
d
m
et
ho
d
al
s
o
sho
ws
a
n
ou
tperfo
rm
resu
lt
with
78
%
accuracy,
22
%
err
or
rate
w
it
h
Ger
m
an.
dat
a
-
num
eric
dataset
and
85.
36
%
accuracy,
14.
64
%
er
ror
r
at
e
with
Au
st
rali
an
dat
aset
.
Be
tt
er
classificat
ion
rat
e
than
an
oth
e
r
popu
la
r
cl
assi
ficat
ion
al
gorit
hm
su
ch
as
su
pport
vecto
r
m
achine,
ne
ur
al
net
w
ork,
a
nd
ge
netic
al
go
rithm
.
It
al
so
con
cl
uded
the
fact
tha
t
featur
e
sel
ect
ion
a
s
pr
e
processi
ng
ste
p of
data m
i
ning c
ou
l
d
in
cr
ease pe
rfor
m
ance.
Nex
t
bi
g
c
halle
ng
e
is
to
inc
r
ease
the
sp
ee
d
of
e
xecu
ti
on
of
the
pro
pose
d
m
od
el
,
due
to
the
lo
ng
execu
ti
on
ti
m
e
.
Since
s
pee
d
ha
s
bec
om
e
a
prob
le
m
,
further
researc
h
will
be
fo
c
us
i
ng
t
o
i
ncr
ea
se
the
s
pe
ed
of
execu
ti
on ti
m
e
. Futu
re s
tu
dies
m
a
y use a
no
t
he
r
feat
ur
e
selec
ti
on
m
et
ho
d as
par
t
of f
it
nes
s fun
ct
io
n
B
PS
O.
ACKN
OWLE
DGE
MENTS
This
paper
is
pa
rt
of
resea
rch
w
ork
f
or
Ma
ste
r
of
Inf
or
m
at
ic
s,
Faculty
of
Com
pu
te
r
Scie
nce
,
Un
i
ver
sit
as
Sr
i
wij
ay
a.
REFERE
NCE
S
[1]
X.
-
L.
Li,
“
An
O
ver
vie
w
of
Personal
Credi
t
Scor
ing:
Techni
ques
and
Future
W
ork,
”
Int.
J.
Intell
.
Sci
.
,
vol.
02,
no
.
24,
pp
.
182
–
190
,
2012.
[2]
S.
Hus
sain,
N.
A.
Daha
n,
F.
M.
Ba
-
al
wi
,
and
N.
Riba
ta,
“
Educ
ational
Dat
a
Mining
and
Anal
ysis
of
Student
s
’
Aca
demic
Perfor
m
anc
e
Us
ing
W
EKA,”
In
t. J
.
E
l
ec
tr.
Comput.
E
ng.
,
vol. 9, no. 2, pp. 447
–
459,
20
18.
[3]
S.
M.
Sadat
r
asoul,
M.
Ghol
amian,
M.
Si
ami,
a
nd
Z.
Ha
ji
m
oha
m
m
adi
,
“
Credi
t
scoring
in
ban
ks
and
fina
nc
ial
insti
tutions
via
d
at
a
m
ini
ng
techn
ique
s:
A
l
it
er
at
ur
e
rev
ie
w,
”
J
.
AI
Data
Mini
ngJou
rnal
AI
Data
M
i
n.
,
vo
l.
1
,
no
.
2,
pp.
119
–
129
,
20
13.
[4]
T.
S.
L
ee,
C.
C
.
Chiu,
Y.
C
.
Ch
ou,
and
C.
J.
Lu
,
“
Mining
th
e
cu
stom
er
cre
dit
us
ing
class
ifi
catio
n
and
reg
ression
tre
e
and
m
ult
ivaria
t
e
ad
aptive re
gre
ss
ion
spline
s,
”
Comput.
Sta
t.
Data
Anal.
,
vol
.
50,
no
.
4
,
pp
.
11
13
–
1130,
2006
.
[5]
W
.
Chen,
C.
Ma,
and
L.
Ma,
“
Mining
the
customer
cre
di
t
using
hy
brid
support
ve
ct
or
m
ac
hine
t
echnique
,
”
Ex
p
ert
Syst.
Appl.
,
vol.
36,
no
.
4
,
pp
.
76
11
–
7616,
2009
.
[6]
J.
Chen,
“
A
Method
of
Im
proving
Credi
t
Ev
al
u
at
ion
with
Supp
ort
Vec
tor
Mac
hine
s,”
2015
11t
h
I
nt.
Conf.
Nat
.
Comput.
,
pp
.
61
5
–
619,
2015
.
[7]
J.
Abell
án
and
J.
G.
Caste
ll
ano
,
“
A
compara
ti
ve
stud
y
on
base
cla
ss
ifi
ers
in
ense
mbl
e
m
et
hods
for
cre
di
t
scoring,”
Ex
pert
Syst
.
App
l.
,
vol
.
73
,
pp
.
1
–
10,
2017
.
[8]
Y.
Ping
and
L
.
Yongheng,
“
Nei
ghborhood
rough
set
and
SV
M
base
d
h
y
brid
cr
e
dit
scoring
cl
assi
fie
r,”
E
xpe
rt
Sys
t.
Appl
.
,
vo
l. 38, n
o.
9
,
pp
.
11300
–
11304,
2011
.
[9]
Z.
Zha
ng
,
G.
Gao,
and
Y.
Shi,
“
Credi
t
risk
eva
luation
using
m
ult
i
-
cri
t
eria
opti
m
iz
ation
class
ifi
er
with
ker
n
e
l
,
fuz
zi
f
ic
a
ti
on
and
penalt
y
fa
c
tors,
”
Eur.
J
.
Oper
.
R
es.
,
vol. 237, no. 1, pp. 335
–
348,
2014.
[10]
A.
Bandhu
and
M.
Kum
ar,
“
Com
puta
ti
onal
t
ime
red
u
ct
ion
for
cre
di
t
scoring
:
An
int
egr
at
ed
a
pproa
ch
b
ase
d
o
n
support
vector
m
ac
hin
e and
stra
tified
sam
pli
ng
m
et
hod,
”
Ex
pert
S
yst.
Appl.
,
vol.
3
9,
no
.
8
,
pp
.
677
4
–
6781,
2012
.
[11]
H.
He,
W
.
Zh
an
g,
and
S.
Zha
ng
,
“
A
novel
ense
m
ble
m
et
hod
for
cre
di
t
scoring
:
ada
pt
ion
of
diff
ere
nt
imbala
n
ce
rat
ios Hongl
ia
ng
,
”
Ex
p
ert
S
yst. Appl.
,
2018.
[12]
F.
N.
Koutanaei
,
H.
Saje
d
i,
and
M.
Khanba
ba
ei,
“
A
hy
br
id
da
t
a
m
ini
ng
m
odel
of
fea
tur
e
sel
ectio
n
al
gori
thms
and
ense
m
ble
l
ea
rn
in
g
class
ifi
ers
for
cre
di
t
scoring
,
”
J.
Retail.
Consu
m.
Serv
.
,
vol
.
27
,
pp.
11
–
23,
2015
.
[13]
X.
W
u,
V.
Kum
ar,
Q.
J
.
Ross
,
J.
Ghos
h,
Q.
Yan
g,
H.
Motoda
,
G.
J.
McLachl
a
n,
A.
Ng,
B
.
L
i
u,
P.
S.
Yu,
Z.
H.
Zhou,
M
.
Ste
inb
ac
h,
D.
J.
Hand
,
and
D.
Stei
nb
erg
,
Top 10
algorit
h
ms
in
data
min
in
g
,
vol
.
14
,
no
.
1
.
2008.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Credit
S
c
or
in
g Usin
g
C
ART
Algo
rit
hm
and
Bina
ry
P
ar
ti
cl
e S
w
ar
m
Opti
miza
ti
on
(
Rez
a Fi
rsan
da
y
a
M
alik
)
5431
[14]
J.
He
and
H.
Gu
o,
“
A
Modifie
d
Parti
cle
Sw
arm
Optimiza
ti
o
n
Al
gorit
hm
,
”
Int
.
J
.
El
e
ct
r.
Comput
.
Eng.
,
vo
l.
11
,
no
.
10,
pp
.
6209
–
62
15,
201
3.
[15]
W
.
A.
Shukur
and
K.
K.
Jabba
r,
“
Inform
at
ion
Hiding
using
LSB
Te
chn
ique
base
d
on
Deve
lope
d
PS
O
Algorit
hm
,
”
Int.
J. Ele
c
tr.
Co
mput.
Eng
.
,
vol
.
8,
no
.
2
,
pp
.
115
6
–
1168,
2018
.
[16]
H.
Shahinz
ad
eh,
S.
M.
Nasr
-
az
ad
ani
,
and
N.
Jann
esa
ri,
“
Appli
ca
t
i
ons
of
Parti
cle
Sw
arm
Optimiza
tion
Algorit
hm
to
Solving
the
Ec
o
nom
ic
Loa
d
Dis
pat
ch
of
Units
in
Pow
er
S
y
stems
with
Valve
-
Poin
t
Eff
e
ct
s,
”
Int
.
J
.
El
e
ct
r.
Comput
.
Eng.
,
vol
.
4
,
no
.
6,
pp
.
858
–
867
,
2014.
[17]
R.
F.
Ma
li
k,
T
.
A.
Rahman,
R
.
Ngah,
S.
Za
i
ton,
and
M.
Hash
im,
“
The
New
Mul
ti
point
Rela
y
s
Selecti
on
in
OLS
R
using Par
ticle Sw
arm Opti
m
iz
at
i
on,
”
TEL
KOMN
IKA
,
vo
l. 10, no. 2, pp. 343
–
352,
2012.
[18]
Q.
Bai,
“
Anal
y
si
s of
Particl
e
Sw
a
rm
Optimiza
ti
on
Algorit
hm
,
”
20
10.
[19]
A.
A.
Yah
y
a
a
nd
A.
Os
m
an,
“
Cla
ss
ifi
ca
t
ion
of
Hi
gh
Dim
ensiona
l
Educat
io
nal
Data
using
Parti
cle
Sw
arm
Cla
ss
ifi
c
at
ion
,
”
pp.
34
–
41
,
2014
.
[20]
A.
S.
H.
Basar
i,
B.
Hus
sin,
I
.
G.
P.
Ananta,
and
J.
Ze
n
ia
rj
a,
“
Opinion
m
ini
ng
of
m
ovie
r
evi
ew
usin
g
h
y
b
rid
m
et
hod
of
support
v
ec
to
r
m
ac
hine a
nd
p
art
i
cl
e
sw
arm opt
im
iz
ation,
”
Proce
dia
Eng
.
,
vol.
53,
pp
.
453
–
462
,
2013.
[21]
M.
Lichm
an, “{U
CI}
Mac
hine L
ea
rning
Reposit
o
r
y
,
”
2013
.
[Onli
ne]
.
Available:
h
tt
p://ar
chi
ve
.
i
cs.
uci
.
edu/
m
l.
[22]
H.
Jiawe
i
,
M
.
K
amber,
J.
Han,
M.
Kam
ber
,
and J
.
Pei,
Data
Mining: Conc
ep
ts an
d
Techni
q
u
es
.
20
12.
[23]
F.
Gorunescu,
“Cl
assificat
ion
Perform
anc
e
Ev
aluati
on
,
”
in
Data
Mini
ng.
Concepts,
Mode
ls
and
Techni
ques
,
vol
.
12,
2011
,
pp
.
31
9
–
330.
[24]
S.
-
W
.
Li
n
,
K.
-
C
.
Ying,
S.
-
C.
Ch
en,
and
Z.
-
J.
L
e
e,
“
Particl
e
sw
ar
m
opti
m
iz
ation
for
par
ame
te
r
d
e
te
rm
ina
t
ion
and
fea
tur
e
se
le
c
ti
on
of
support
vecto
r
m
ac
hine
s
,
”
Expert
Syst
.
Appl
.
,
vol.
35
,
no
.
4
,
pp
.
1817
–
1824
,
20
08.
[25]
S.
Talukder
,
“
Mathe
m
atical
Mod
el
li
ng
and
Appli
ca
t
ions
of
Part
icle
Sw
arm
Optimization,”
Schoo
l
of
Eng
ine
e
ring
Ble
kinge Insti
tute
of Te
chnol
o
g
y
,
2011.
BIB
LIOGR
A
PHY
OF A
UT
HORS
Rez
a
Firsand
a
y
a
Mali
k
was
born
in
Padang,
W
est
Sum
at
era
in
19
76.
He
recei
v
ed
his
senior
high
school
in
SM
AN
70
Bulunga
n,
Jaka
rt
a
(1991
-
1994).
He
gr
a
duat
ed
f
rom
Instit
ut
Sa
ins
dan
Te
knologi
Nasio
nal
(ISTN
),
Jaka
rta
,
as
S.T
(Ba
ch
el
or
of
Engi
ne
ering)
in
2000
and
obta
in
ed
M.T
(Master
of
Te
ch
nique
)
from
Instit
ut
Te
kno
logi
Bandung
in
200
3.
He
rec
e
ive
d
t
he
PhD
degr
ee
from
Univer
siti
Te
knologi
Malay
s
ia
(UTM)
in
2011,
where
he
inv
esti
gated
Routi
n
g
Optimiza
ti
o
n
Sc
hem
e
in
W
ire
le
s
s Mesh Net
works
using Par
ticle
Sw
arm Opti
m
iz
at
ion.
He
joi
ned
Fa
cult
y
of
Com
puter
Scie
nce,
Univ
ersit
as
Sriwijaya
(UN
SR
I)
as
a
Le
c
ture
r
in
Dec
ember
2010.
He
al
so
appoi
nte
d
as
m
ember
of
Com
m
unic
ation
Network
and
Secur
i
t
y
(COM
NETS)
Resea
rch
L
abor
a
t
o
r
y
in
Fa
cul
t
y
o
f
Com
pute
r
Scie
nce,
Univer
sit
a
s
Sriwijay
a.
During
complet
i
ng
Ph.D
stud
y
i
n
W
ire
le
ss
Com
m
unic
at
ion
C
en
tre
(W
CC)
(200
4
-
2006),
he
invol
ved
in
W
ire
le
ss
Campus
Project
–
Design
and
Deplo
y
m
e
nt
of
Hot
-
spot
IEE
E
802
.
11g
W
ire
le
ss
LAN,
col
la
bor
at
io
n
bet
wee
n
W
CC,
UTM
and
In
dustr
y
.
He
wor
ked
cl
osel
y
as
rese
arc
h
er
in
Malay
s
ia
gov
er
nm
ent
age
nc
ie
s
such
as
Minist
r
y
of
Sci
ence,
Te
chno
log
y
and
Innova
ti
on
(MO
STI)
and
Min
istr
y
of
Highe
r
Edu
ca
t
ion
(MO
HE) M
al
a
y
s
ia.
He
appoi
nt
ed
as
a
Co
-
Chie
f
Ed
it
or
in
Com
EngApp
-
Journal.
Th
us,
as
m
ember
of
Instit
ute
of
El
e
ct
ri
ca
l
and
Elec
tron
ic
s
Eng
ineers
(IE
E
E),
m
oshara
ka
for
res
ea
r
ch
and
stud
ie
s
(
m
oshara
ka.
ne
t)
and
As
socia
t
ion
of
Inform
at
ic
s
a
nd
Com
pute
r
Coll
eg
e
(AP
TIKOM
).
His
expe
ri
e
nce
in
journa
l
m
ana
gement
as
a
rev
ie
wer
in
T
EL
KO
MN
IKA
Journal,
Journa
l
of
Network
a
nd
Com
pute
r
Applic
a
ti
ons
(JN
CA)
and
seve
ral
Int
ern
a
ti
ona
l
Confer
e
nce
s
and
al
so
as
Journal
Ed
it
o
r
in
Com
pute
r
a
nd
Engi
ne
eri
n
g
Applic
a
ti
ons
(
Com
EngApp
)
and
Insti
tut
e
of
Advanc
e
d
Engi
ne
eri
ng
and
Scie
nc
e
(IAES
).
In
UN
SR
I,
hi
s
cur
ren
t
r
ese
ar
ch
intere
sts
in
clude
computer
net
works
and
so
ft
computing
.
He
al
so
assigned
as
Hea
d
of
Serv
ic
e
and
Appli
cation
W
orking
Group i
n
Indon
e
sia
5G Forum
.
Herm
awa
nis
a
m
aste
r
student
at
the
f
ac
u
lty
of
computer
sci
enc
e
Univ
ersity
of
Sriwij
a
y
a
,
Pale
m
bang,
Sout
h
Sum
at
era
.
Cur
ren
tly
working
a
s
a
le
ct
ur
e
a
t
inf
orm
at
ion
s
y
stem
depa
rtment
at
STMIK
GI
MDP,
Pale
m
bang,
South
Sum
at
era
.
Pass
iona
te
about
la
te
st
t
ec
hnolog
y
,
d
eve
lop
ing
informati
on
s
y
st
em,
ana
l
y
a
z
e
s
y
stem.
His
rese
ar
ch
int
er
est
area
in
so
ftwa
re
enginee
ring
,
data
m
ini
ng,
da
ta sci
e
nti
st,
databa
se
a
nd
informati
on
sy
stem.
Evaluation Warning : The document was created with Spire.PDF for Python.