Indonesi
an
Journa
l
of El
ect
ri
cal Engineer
ing
an
d
Comp
ut
er
Scie
nce
Vo
l.
12
,
No.
3
,
Decem
ber
201
8
, p
p.
11
11
~
11
16
IS
S
N: 25
02
-
4752, DO
I: 10
.11
591/ijeecs
.v1
2
.i
3
.pp
11
11
-
11
16
1111
Journ
al h
om
e
page
:
http:
//
ia
es
core.c
om/j
ourn
als/i
ndex.
ph
p/ij
eecs
Optimi
zation
Learnin
g V
ec
t
or Qu
antizati
on
Using Gen
etic
Algorith
m for De
tection
of Di
abetics
Inggih Per
ma
na
1
, Nes
di Evr
il
ya
n
R
oz
an
d
a
2
, Fadhil
ah S
yafri
a
3
, Fe
bi N
ur S
alisah
4
1,2,4
Depa
rtment of Inf
orm
at
ion
Sy
stems
,
Fa
cul
t
y
of
Scie
n
ce a
nd
T
ec
hnolog
y
,
Univ
ersit
as
Sul
ta
n
S
yar
if
Kasim
(UIN
SU
SKA)
Ria
u,
P
eka
nbar
u
-
R
ia
u
,
28293
3
Depa
rtment
of
I
nform
at
ic
s E
ng
i
nee
ring
,
Fa
cul
t
y
of
Scie
n
ce a
nd
T
ec
hnolog
y
,
Univ
ersit
as
Sul
ta
n
S
yar
if
Kasim
(UIN
SU
SKA)
Ria
u,
P
eka
nbar
u
-
R
ia
u
,
28293
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Feb
11
, 201
8
Re
vised
Ma
y
30
, 2
01
8
Accepte
d
Aug
2
1
, 201
8
Thi
s
stud
y
p
rop
osed
the
m
et
ho
d
to
improve
th
e
result
of
Lear
ning
Vec
tor
Quanti
z
at
ion
(L
VQ
)
by
opti
m
i
zi
ng
the
weigh
t
vec
tors
using
a
gene
t
i
c
al
gorit
hm
(GA
)
to
det
e
ct
the
di
a
bet
i
cs.
Init
i
al
va
lue
of
indi
vidual
s
for
G
A
is
ta
ken
from
wei
ght
vec
tors
whi
ch
come
from
the
la
st
m
it
erati
ons
of
LV
Q
tra
ini
ng
resul
t.
The
r
esult
o
f
ex
per
iment
show
e
d
that
th
ere
is
a
significan
t
inc
re
ase
in
sen
siti
vity
le
ve
l,
h
oweve
r
the
r
e
is
a
signifi
c
ant
dec
re
ase
in
spec
ificity
le
v
el.
It
m
ea
ns
th
e
p
roposed
m
et
hod
succ
ess
in
improving
th
e
LVQ
abi
li
t
y
to
rec
ogni
ze
d
th
e
di
abe
t
ic
s,
but
i
t
lo
wers
the
abi
l
ity
of
LVQ
to
rec
ogni
ze t
he
pe
ople
un
aff
e
ct
ed
b
y
dia
b
et
es
.
Ke
yw
or
d
s
:
Diabeti
cs
GA
LVQ
Weig
ht v
e
ct
ors
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
:
Ing
gih
Perm
ana
,
Dep
a
rtm
ent o
f Info
rm
at
ion
System
s,
Faculty
o
f
Science
and
Tech
no
l
og
y,
Un
i
ver
sit
as
Su
l
ta
n
Syari
f
K
a
sim
(
UI
N SU
SKA)
Ri
a
u,
Peka
nb
a
r
u
-
Ri
a
u,
2829
3
.
Em
a
il
:
ing
gihp
erm
ana@u
in
-
s
us
ka
.ac.i
d
1.
INTROD
U
CTION
Diabetes
Me
ll
it
us
(D
M
)
is
a
disease
that
occu
rs
wh
e
n
th
e
pan
c
reas
can
not
to
secreti
on
en
ou
gh
insu
li
n
[
1].
D
M
m
ay
increase
the
risk
of
vessel
dam
age,
blindness
,
ki
dn
ey
disease
,
card
ia
c
diseas
e,
nerv
e
dam
age,
stroke,
birth
de
fect
s
[2
]
.
DM
is
on
e
of
the
m
a
j
or
healt
h
pr
oble
m
s
that
occu
r
in
I
ndonesi
a.
The
pr
e
valence
of
DM p
at
ie
nts (d
ia
betic
s)
in Indon
e
sia
is incre
asi
ng
e
ver
y y
e
ar.
Am
ong
the
1980s,
the
pre
valence
of
dia
betic
s
in
pe
op
le
with
a
ge
ov
e
r
15
ye
ars
is
1,5%
t
o
2,3%
[3
]
.
I
n
2001,
in
urba
n
areas,
t
he
pr
e
va
le
nce
diabeti
cs
o
f
pe
op
le
age
d
bet
ween
25
-
64
ye
ars
old
is
5.7%
[
4].
In
20
13,
t
he
pr
e
valen
ce
of
dia
betic
s
in
urba
n,
rural
an
d
w
ho
l
e
I
ndonesi
a
are
6,8%
,
7%,
a
nd
6,9%
[3
]
.
From
twel
ve
m
i
llion
diabeti
cs
i
n
I
ndonesi
a,
th
e
re
was
69,6%
undia
gnos
e
d
since
the
disease
we
re
aff
ect
ed
.
This
m
eans
that
m
os
t
diabeti
cs
in
Ind
on
esi
a
reali
zed
the
disease
w
hen
it
is
seve
re.
T
he
reas
on
is
t
ha
t
the
DM
a
pp
e
ars
over
m
any
ye
ars
s
o
it
wa
s
not
reali
zed
by
th
e
su
f
fer
e
r.
Com
pu
te
r
te
ch
no
l
og
y
for
the
early
detect
ion
of
diab
et
ic
s
is
a
so
luti
on
t
o
r
esolve
t
he
pro
blem
s.
This
has
been
done
by
previ
ous
r
esearche
rs
by
dev
el
op
i
ng
the
var
ie
ty
of
al
gorithm
s,
su
c
h
as:
(1)
ch
an
gi
ng
t
he
al
gorithm
s
artif
ic
ia
l
i
m
m
un
e
recog
niti
on
syst
e
m
by
add
in
g
fuzzy
k
-
neare
st
neighb
or
[
5];
(2
)
c
om
bin
ing
the
al
gorithm
between
cent
rip
et
al
sp
e
d
up
par
ti
c
le
swar
m
op
ti
m
iz
at
ion
and
m
ul
ti
-
la
ye
r
perce
ptr
on
al
gorithm
[6
]
;
(3) by usi
ng
na
ive b
ay
es
and
decisi
on tree
m
et
hods
[7
]
.
This
researc
h
us
e
d
le
ar
ning
vecto
r
qu
a
ntiza
ti
on
(L
VQ)
m
et
ho
ds
f
or
ea
rly
diabeti
cs
de
te
ct
.
LV
Q
i
s
the
patte
r
n
cl
as
sifi
cat
ion
m
et
ho
d
w
her
e
the
e
ntire
outp
ut
unit
rep
rese
nts
t
he
certai
n
cl
asse
s
[
8].
LV
Q
we
re
the
fastest
an
d
ea
sie
st
app
li
ed
a
nd
intuit
ive
[9]
.
LV
Q
is
ch
ose
n
in
this
res
earch
beca
us
e
LV
Q
has
bee
n
the
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
12
, N
o.
3
,
Dece
m
ber
2
01
8
:
1
1
11
–
1
1
16
1112
su
ccess
t
o
a
ppli
ed
in
m
any
areas
of
resea
rch,
li
ke
i
n:
(
1)
i
d
e
ntific
at
ion
hand
-
w
riti
ng
[10];
(
2)
real
-
tim
e
m
on
it
or
ing [
11]
; and
(3)
a
rr
ay
an
al
ysi
s fro
m
a sen
s
or [1
2].
Howe
ver,
L
V
Q
has
a
la
c
k;
the
trai
ning
r
esult
of
weig
ht
vecto
rs
is
de
pende
d
on
the
par
am
et
ers
init
ia
li
zation
th
at
us
ed
in
L
V
Q.
A
w
ei
ght
ve
ct
or
s
is
im
po
rt
ant
in
L
V
Q
be
cause
this vect
or
w
il
l
determ
i
ned
the
data
cl
assifi
cat
ion
to
the
certa
in
cl
asses.
In
this
researc
h,
w
ei
gh
t
vecto
r
wi
ll
ref
eren
ces
to
determ
ine
so
m
eon
e
who
diag
no
se
as
the
diabeti
c
s
or
no
t
.
The
w
eakn
e
ss
of
L
V
Q
can
be
res
ol
ved
by
opti
m
ized
the
weig
ht
vecto
rs
us
e
a
ge
netic
al
gorithm
(G
A).
GA
is
a
sear
chin
g
al
gorith
m
that
insp
ire
d
from
the
nat
ur
al
ev
olu
ti
on
theo
ry.
The res
ult of th
e previ
ou
s
r
ese
arch sh
ow that
GA can
opti
m
i
ze the a
bili
ty
o
f
cl
assifi
ed
alg
or
it
hm
[
13
-
17]
.
LVQ
op
ti
m
iz
ation
us
i
ng
GA
has
bee
n
done
in
pr
e
vious
stu
dies.
The
opti
m
iz
at
ion
is
do
ne
by
fin
ding
the
init
ia
l
val
ue
of
L
V
Q
wei
gh
t
vect
or
us
i
ng
G
A
[
18
-
19]
.
Alth
ough
t
he
init
ia
l
value
of
the
wei
gh
t
vecto
r
aff
ect
s
the
L
V
Q
res
ult,
it
ca
n
not
guara
nte
e
that
LVQ
tr
ai
nin
g
pro
du
ce
s
the
op
ti
m
al
r
epr
ese
ntati
ve
vecto
r.
The
m
et
ho
d
of
fer
e
d
in
this
stud
y
do
the
oppo
sit
e,
LV
Q
t
r
ai
nin
g
do
ne
fir
st,
the
n
GA
is
us
e
d
to
opti
m
i
ze
the
vecto
r
re
prese
ntati
ve
of
t
he
LVQ
trai
ni
ng
resu
lt
s.
I
n
a
dd
i
ti
on
,
t
her
e
are
al
so
stu
dies
t
ha
t
op
ti
m
iz
e
L
VQ
by
fin
ding
dire
ct
ly
weigh
t
vect
or
s
us
i
ng
GA
[20].
I
n
s
uch
t
echn
i
qu
e
s,
the
LVQ
trai
ning
process
is
no
t
done
.
LVQ
is
only
use
d
in
the
cl
as
sific
at
ion
proc
ess
on
ly
.
T
he
m
et
ho
d
offe
re
d
in
this
stud
y
did
not
el
i
m
in
at
e
the
LVQ trai
ni
ng
process
since t
he
init
ia
l i
ndivi
du
al
s
u
se
d o
n GA de
rive
d fro
m
the LVQ trai
ning.
2.
RESEA
R
CH MET
HO
D
2.1.
Opt
im
iz
at
io
n
of we
ight vec
t
or of
LVQ
usi
ng
GA
The
opti
m
iz
e
of
L
VQ
c
om
pr
ise
four
sta
ge
s.
First
init
ia
lize
the
par
am
eter
value
s
of
L
VQ
a
nd
G
A
.
The
sec
ond
sta
ge
is
L
VQ
t
r
ai
nin
g.
T
he
thi
rd
sta
ge
is
the
est
a
blishm
ents
init
ia
l
ind
ivi
du
al
for
GA.
The
la
st
sta
ge
the
re is
optim
iz
at
ion
u
si
ng GA.
The i
ll
us
trat
io
n of t
he
pro
ces
s ca
n be
shown i
n
e
qua
ti
on
ure
1.
I
n
i
t
i
a
l
i
z
a
t
i
o
n
o
f
i
n
i
t
i
a
l
L
V
Q
w
e
i
g
h
t
v
e
c
t
o
r
s
,
L
V
Q
p
a
r
a
m
e
t
e
r
s
a
n
d
G
A
p
a
r
a
m
e
t
e
r
s
S
t
a
r
t
S
e
l
e
c
t
m
w
e
i
g
h
t
v
e
c
t
o
r
s
f
r
o
m
m
i
t
e
r
a
t
i
o
n
s
o
f
L
V
Q
t
r
a
i
n
i
n
g
a
s
t
h
e
i
n
i
t
i
a
l
i
n
d
i
v
i
d
u
a
l
o
f
G
A
L
V
Q
t
r
a
i
n
i
n
g
F
i
n
i
s
h
O
p
t
i
m
i
z
a
t
i
o
n
o
f
w
e
i
g
h
t
v
e
c
t
o
r
s
u
s
i
n
g
G
A
Figure
1. The
LVQ opti
m
iz
a
t
ion
proces
s
us
e
s GA
The
pa
ram
et
er
s
that
need
to
be
init
ia
li
zed
on
L
VQ
a
re
th
e
init
ia
l
value
of
the
wei
ght
vecto
rs,
the
nu
m
ber
of
it
er
at
ion
s,
the
le
a
r
ning
rate,
t
he
de
crem
ent
le
arni
ng
rate.
The
pa
ram
et
ers
that
need
to
be
init
ia
li
zed
on
GA
are
the
nu
m
ber
of
ge
ner
at
io
ns
,
the
nu
m
ber
of
indi
vid
uals
,
cross
ov
e
r
pro
ba
bili
t
y
(P
c)
an
d
m
u
ta
ti
on
pro
bab
il
it
y (P
m
).
The
init
ia
l
ind
i
viduals
f
or
GA
in
this
resea
rc
h
we
re
ta
ke
n
f
r
om
m
la
st
i
te
rati
on
of
L
VQ
t
r
ai
nin
g.
T
he
chrom
os
om
e
f
ro
m
the
ind
ivi
du
al
s
of
GA
in
this
researc
h
was
re
pr
e
sente
d
in
the
real
f
orm
.
Fo
r
m
or
e
detai
ls
sh
ow
Fig
ur
e
2.
This stu
dy u
se
s stoch
ast
ic
uni
ver
sal
sam
pling
(
S
US) as a se
le
ct
ion
m
et
ho
d. SUS is a m
e
t
hod
that
has
a
bias
is
0
a
nd
has
th
e
com
plexity
is
O
(N)
[
21
]
.
Ba
sic
al
ly
,
SU
S
is
a
r
oull
et
e
wh
eel
with
N
po
i
nters.
N
is
th
e
nu
m
ber
of
in
di
vid
uals
sel
ec
te
d.
The
first
i
nd
i
vidual
is
a
rand
om
value
betwee
n
1
a
nd
1
/
N
.
T
he
ne
xt
ind
ivi
du
al
is
1
/
N
from
the
previ
ou
s
in
div
i
dual
.
Fi
gure
3
is
an
il
lust
rati
on
of
how
SUS
works.
I
n
the
pi
ct
ur
e
is
sel
ect
ed
five
ind
ivid
uals
so
that
there
are
five
pointer
s
and
the
di
sta
nce
between
in
divi
du
al
s
is
0.2.
Ba
se
d
on the
sel
ect
ion
resu
lt
, I
1,
I1,
I2, I4
a
nd
I5 we
re s
el
ect
ed
.
The
c
rosso
ve
r
te
chn
i
que
tha
t
us
e
i
n
this
r
esearch
is
li
ne
-
cr
os
s
ov
e
r,
an
d
the
m
utati
on
is
do
ne
by
add
i
ng a sm
al
l
rand
om
v
al
ue.
Fo
rm
ula of li
ne
-
cr
os
s
ov
e
r
c
a
n be see
n
i
n
E
quat
ion 5
.
=
1
+
(
2
−
1
)
(1)
In
the
Eq
uatio
n
1,
CGe
ni
is
i
-
th
ge
n
of
a
c
hi
ld,
PG
e
n1i
is
i
-
th
gen
of
a
fi
rs
t
par
e
nt,
P
gen2
i
is
i
-
th
ge
n
of a sec
ond pa
r
ent, λ
is a
rand
om
v
al
ue
bet
w
een
0 un
ti
l
1.
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c
En
g
&
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Op
ti
miz
atio
n
L
ear
ning Vect
or Q
uantizati
on
Using
Genet
ic
Algo
rit
hm f
or
Detect
ion
…
(
I
nggih Per
man
a
)
1113
Figure
2. Early
ind
i
vidual G
A
creati
on
I
1
I
2
I
3
I
4
I
5
0
0
.
1
0
.
2
0
.
3
0
.
4
0
.
5
0
.
6
0
.
7
0
.
8
0
.
9
1
P
o
i
n
t
e
r
1
(
R
a
n
d
o
m
)
P
o
i
n
t
e
r
2
P
o
i
n
t
e
r
3
P
o
i
n
t
e
r
4
P
o
i
n
t
e
r
5
0
.
2
0
.
2
0
.
2
0
.
2
Figure
3. Illust
rati
on of
how SUS
works
2.2.
Fitness
Func
tion
Fit
ness values
in this
researc
h was
com
pu
te
d wit
h Eq
uatio
n 2.
=
(
1
−
Ac
c
ura
c
y
)
+
(
1
−
S
e
n
sit
ivity
)
+
(
1
−
S
pe
sif
ic
i
ty
)
3
(2)
Accuracy
, s
e
nsi
ti
vity
an
d
s
pec
ific
it
y are got b
y Eq
uation 3
, E
quat
ion 4
and
Equati
on
5.
=
+
+
+
+
(3)
=
+
(4)
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
12
, N
o.
3
,
Dece
m
ber
2
01
8
:
1
1
11
–
1
1
16
1114
=
+
(5)
Wh
e
re,
tr
ue
po
sit
ive
(TP)
is
the
num
ber
of
DM
patie
nts
cl
assifi
ed
as
DM
,
false
neg
at
ive
(F
N
)
is
the
num
ber
of
non
-
DM
pat
ie
nts
cl
assifi
ed
as
DM,
true
ne
gative (TN)
is
the
num
ber
o
f
non
-
DM p
at
ie
nts
cl
assifi
ed
a
s
non
-
DM p
at
ie
nts, f
al
se posit
ive (F
P)
is t
he nu
m
ber
of
DM
patie
nts classi
fied
a
s non
-
DM p
at
i
ents.
2.3.
Dataset
Dataset
wh
e
re
us
e
d
in
this
res
earch
is
Pim
a
In
dia
ns
Data
bas
e.
This
dataset
can
dow
nlo
a
d
on
web
sit
e
reposit
ory
le
arn
i
ng
m
achine
of
the
Un
i
ve
rsity
o
f
Ca
li
fo
r
ni
a
Irvine
(h
tt
ps
:/
/a
rc
hiv
e
.ics.uci.e
du/m
l/d
at
aset
s/Pi
m
a+
Indians+
Diabe
te
s).
T
he
pur
po
s
e
of
t
his
set
of
data
i
s
f
or
cl
assifi
ed
the
Pi
m
a
Indians
people
t
hat
af
f
ect
ed
by
DM
or
not
acc
ordi
ng
to
pe
rs
on
al
data
a
nd
heat
h
c
hec
k.
This
set
of
dat
a
wer
e
sta
g
e
from
76
8
data
w
her
e
268
peopl
e
aff
ect
ed
by
DM
an
d
500
pe
op
le
un
a
ff
ect
ed
by
DM.
The
re
are
7
at
tribu
te
in
t
his
set
of
data.
Ther
e
is
the
nu
m
ber
of
preg
nan
cy
,
plasm
a
glu
c
os
e
co
nce
ntrate
,
diastoli
c b
l
ood p
ressure,
the t
hick
ness o
f
tric
eps
s
kin f
old
s
, body
we
ig
ht, DM ge
neal
og
i
cal
h
ist
or
y a
nd
age.
2.4.
Experim
en
ta
l
Setu
p
The
e
valuati
on
was
perf
or
m
e
d
us
in
g
10
-
f
ol
d
cr
os
s
-
validat
ion
.
To
ap
ply
the
e
valuati
on,
then
t
he
da
ta
set
is
di
vid
e
d
i
nto
te
n
gro
ups
.
Ni
ne
gro
ups
will
be
us
e
d
a
s
trai
ni
ng
data
w
hile
an
oth
e
r
will
be
us
e
d
as
te
st
data.
For
each
com
bin
at
ion
of
par
am
et
ers
ten
ex
per
im
ents
wer
e
pe
rfo
rm
e
d
so
that
al
l
gr
oups
we
re
onc
e
te
st
data.
Af
te
r
all
exp
e
rim
ents w
ere calc
ulate
d t
he
a
ver
a
ge
acc
ur
acy
,
sensi
ti
vity
an
d specific
i
ty
.
3.
RESU
LT
S
AND A
N
ALYSIS
Figure
4
is
the
per
f
or
m
ance
com
par
ison
between
L
VQ
a
nd
LV
QGA
us
i
ng
trai
ni
ng
data.
The
fig
ure
sh
ows
G
A
incr
eases
accu
racy
of
L
V
Q
by
10.27%
(
(
73.87
-
66.
99)/6
6.9
9)
.
S
ensiti
vity
le
vel
of
L
V
Q
inc
re
ased
162.5
7%
(
(
59.
00
-
22.47)
/
22.47)
but
sp
eci
fic
it
y
le
vel
of
L
VQ
dec
reased
9.92%
(
(90.8
5
-
81.
84)/9
0.8
5).
Ba
sed
on
this
com
par
iso
n
it
can
be
con
cl
ude
d,
in
the
trai
ning
data,
G
A
incr
eases
LVQ
ac
cur
acy
thr
ough
hi
gh
increase
of se
nsi
ti
vity
level even t
houg
h
s
pe
ci
fici
ty
level decrease.
Figure
5
is
the
pe
rfo
rm
ance
com
par
ison
betwee
n
L
V
Q
an
d
L
V
Q
G
A
us
i
ng
te
sti
ng
data.
GA
decr
ease
s
accu
racy
of
L
VQ
by
1.17%
((7
1.0
3
-
70.
20)/7
1.03).
Howe
ver
,
there
is
a
sign
ific
ant
inc
rea
se
of
sensiti
vity
le
vel
of
LV
Q
by
85
.
86%
(
(54.94
-
29.56)
/
29.56).
Sp
eci
fici
ty
le
vel
of
LV
Q
dec
reased
16.
04%
((93.43
-
78.
44)/
93.43
).
Ba
se
d
on
t
his
com
pari
so
n
it
can
be
con
cl
ud
e
d,
in
t
he
te
sti
ng
data
,
G
A
dec
rease
LVQ
accuracy t
hro
ugh a
high
decr
e
ase o
f
s
pecifici
ty
level eve
n
th
ough se
ns
it
ivit
y l
evel increa
s
e.
This
stu
dy
use
sensiti
vity
t
o
m
easur
e
capab
il
it
y
of
al
gorithm
s
fo
r
re
cognize
dia
bet
ic
s
wh
e
rea
s
sp
eci
fici
ty
to
m
easur
e
capa
bi
li
t
y
of
al
gorithm
s
fo
r
rec
ognize
non
-
dia
be
ti
cs.
Ther
e
fore
,
base
on
c
omparis
ons
of
perform
ance
betwee
n
LV
Q
an
d
LV
Q
G
A
it
can
be
con
cl
ud
e
d
that
GA
im
pr
ov
e
LVQ
ca
pab
il
it
y
in
recog
nizing di
abeti
cs,
bu
t l
ower
LVQ ca
pa
bili
ty
in
recog
ni
zi
ng
non
-
dia
be
ti
cs.
Figure
4. Com
par
is
ons
of
pe
r
form
ance b
et
w
een L
V
Q
a
nd
LVQG
A
in
trai
ning
data
Ac
c
ur
ac
y
Sens
i
ti
v
i
t
y
Spes
i
fi
c
i
t
y
LVQ
66.9
9
22.4
7
90.8
5
LVQ
G
A
73.8
7
59.0
0
81.8
4
0.00
20.0
0
40.0
0
60.0
0
80.0
0
100
.00
Perc
ent
(%
)
C
omparison of
Performance B
et
w
een
LV
Q
and
LV
Q
G
A Us
i
ng Traini
ng
Da
ta
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c
En
g
&
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Op
ti
miz
atio
n
L
ear
ning Vect
or Q
uantizati
on
Using
Genet
ic
Algo
rit
hm f
or
Detect
ion
…
(
I
nggih Per
man
a
)
1115
Figure
5. Com
par
is
ons
of
pe
r
form
ance b
et
w
een L
V
Q
a
nd
LVQG
A
in
tes
ti
ng
data
4.
CONCL
US
I
O
N
In
the
t
rainin
g
data,
the
al
go
rithm
of
fer
e
d
im
pro
ve
s
the
acc
ur
acy
of
L
VQ
al
gorithm
in
the
detect
ion
of
dia
betic
s.
But
in
the
te
sti
ng
data,
LV
Q
G
A
al
gorithm
de
creases
LV
Q
a
ccur
acy
.
T
his
m
eans
there
ha
s
been
ov
e
r
-
fitt
ing
on
the
trai
nin
g
da
ta
trai
n
pr
oce
ss
so
that
the
m
od
el
gen
erat
ed
by
LV
QGA
is
too
exclusive
to
the
trai
ning
data.
Wh
et
her
on
tra
ining
data
or
on
te
sti
ng
data,
GA
inc
reases
LVQ
se
ns
it
ivit
y
le
vel,
bu
t
there
is
al
so
a
sign
ific
a
nt
dec
rease
in
t
he
le
vel
of
s
pecifici
ty
.
This
m
ean
s
the
m
et
ho
d
offer
e
d
inc
reas
es
LV
Q'
s
abili
ty
to
recog
nize
people
aff
ect
ed
by
diabetes,
bu
t
LVQ
'
s
abili
t
y
to
reco
gniz
e
people
who
are
no
t
aff
e
c
te
d
by
diabetes.
REFERE
NCE
S
[1]
Alber
ti
KG
MM
,
Zi
m
m
et
PF
.
“
Defi
nition,
D
ia
gn
osis
and
C
l
assifi
ca
t
ion
of
D
i
abet
es
M
el
l
it
us
and
I
ts
C
om
pli
cations.
Part
1:
D
iagnos
is
and
C
la
ss
ifica
t
ion
of
D
ia
b
et
es
M
el
li
tus
”
.
Provi
sional
R
epor
t
of
a
W
HO
C
onsulta
ti
on
.
Diab
et
i
c
medic
in
e
.
1998;
15
(7)
:
539
-
553
.
[2]
Forbes JM
,
Cooper
ME.
“
Mec
ha
nism
s of
D
ia
bet
i
c
C
om
pli
c
at
ions
”
.
Phy
sio
logi
ca
l
R
evie
ws
.
2013;
93
(1)
:
137
-
188
.
[3]
“
InfoDA
TIN
”
.
P
usat
Data
dan
In
formas
i
Keme
nter
ian
Kesehetan Republ
i
k
Indon
e
sia
.
2014.
[4]
“
La
pora
n
Survei
Keseha
ta
n
Rum
ah
Ta
ngga
2001
”
.
Badan
Pe
n
el
itian
dan
Pe
mbangunan
Kesehat
an,
Keme
ntria
n
Kesehat
an
Re
pu
bli
k
Indone
sia
.
2
002.
[5]
Chikh
MA
,
Saidi
M,
Sett
outi
N.
“
Diagnosis
of
D
ia
betes
D
isea
ses
U
sing
an
A
rti
fic
i
al
I
m
m
une
R
ec
ognit
ion
S
y
st
em2
(AIRS
2)
with
F
uzzy
K
-
N
e
are
st
N
ei
ghbor
”
.
Journal
of
M
edical
S
y
stems
.
2012;
36(
5)
:
2721
-
2729.
[6]
Behe
shti
Z,
Sha
m
suddin
SM
H,
Behe
shti
E,
Yu
hani
z
SS
.
“
Enha
nce
m
ent
of
A
rti
ficia
l
N
eur
al
N
et
work
L
ea
rni
ng
U
sing
C
ent
rip
etal
A
c
cele
ra
te
d
P
art
ic
l
e
S
warm
O
pti
m
iz
at
io
n
fo
r
M
edi
c
al
D
iseas
es
D
ia
gnosis
”
.
Soft
Computing
.
2014;
18(11)
:
22
53
-
2270.
[7]
I
y
e
r
A,
Je
y
alat
ha
S,
Sum
baly
R.
2015.
“
Diagnosis
of
D
ia
bet
es
U
sing
C
l
assific
a
ti
on
M
in
ing
T
e
chni
ques
”
.
Inte
rnational
Jo
urnal
of
Data
Mi
ning
&
Knowle
d
ge
Manage
m
ent
Proce
ss
.
2015;
5
(1):
1
-
14.
[8]
Fausett
L.
“
Fundam
ent
al
s
of
Neura
l
Networks
:
Archi
t
ec
tur
es,
Alg
orit
hm
s
and
App
li
c
at
ions
”
.
New
Jerse
y
:
Pren
ti
c
e
-
Hall
.
1994:
187.
[9]
Ham
m
er
B.
“
Two
or
T
hre
e
T
hinks
t
hat
W
e
D
o
N
ot
K
now
A
b
out
Le
arn
ing
Vec
tor
Quantiza
t
io
n
but
W
e
S
houl
d
C
onsider
”
.
MIW
OCI Works
hop
-
2013
.
2013:
6
-
12
.
[10]
Ho
TK.
“
Rec
o
gnit
ion
of
H
an
dwritt
en
D
ig
it
s
b
y
C
om
bini
ng
I
ndepe
nd
ent
L
ea
rn
ing
V
ecto
r
Q
uant
izat
ions
”
.
Proce
ed
ings
of
T
he
Sec
ond
Inter
nati
onal
Conf
e
renc
e
on
Docu
m
ent
Analysis
a
nd
Recogni
t
ion
.
Tsukuba
Sci
en
ce
Cit
y
.
1993:
818
-
821.
[11]
Mou
y
X,
B
ahou
ra
M,
Sim
ard
Y.
“
Autom
at
ic
R
e
cogni
ti
on
of
F
in
and
B
lue
W
hale
C
alls
for
R
eal
-
T
ime
M
onit
orin
g
in
T
h
e
St
.
L
awre
nce
”
.
The
Journ
al
of
the A
coustical
So
ci
e
ty of
A
meric
a
.
200
9;
12
6(6)
:
2918
-
2928
.
[12]
Ciosek
P,
W
rób
le
ws
ki
W
.
“
Th
e
A
naly
sis
of
S
e
nsor
A
rra
y
D
ata
with
V
ar
ious
P
at
t
ern
R
e
cogniti
on
T
e
chni
ques
”
.
Sensors
and
Ac
t
uators B
:
Chemi
cal
.
2006;
114(1
)
:
85
-
93.
[13]
Kare
gowda
AG
,
Manjuna
th
AS
,
Ja
y
ar
am
MA
.
“
Applic
at
ion
o
f
Gene
ti
c
A
lgorithm
Optimize
d
Neura
l
Network
Connec
ti
on
W
eights
for
Medic
a
l
Diagnosis
of
Pim
a
India
ns
Diab
et
es
”
.
Int
ernati
o
nal
Journal
on
Soft
Computing
.
2011;
2(2):
15
-
2
3.
[14]
Meli
n
P,
Herre
r
a
V,
Rom
ero
D,
Valdez
F,
Casti
ll
o
O.
“
Gene
t
i
c
Optimiza
t
ion
of
Neura
l
Netw
orks
for
Perso
n
Rec
ognition
Bas
ed
on
th
e
Iris
”
.
TEL
KOMNIKA,
Tele
communi
ca
ti
on
Computing
El
e
ct
ronics
and
Control
.
2012
;
10(2):
309
-
320
.
Ac
c
ur
ac
y
Sens
i
ti
v
i
t
y
Spes
i
fi
c
i
t
y
LVQ
71.0
3
29.5
6
93.4
3
LVQ
G
A
70.2
0
54.9
4
78.4
4
0.00
10.0
0
20.0
0
30.0
0
40.0
0
50.0
0
60.0
0
70.0
0
80.0
0
90.0
0
100
.00
Perc
ent
(%
)
C
omparison of
Performance B
et
w
een
LV
Q
and
LV
Q
G
A Us
i
ng Testing
Da
ta
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
12
, N
o.
3
,
Dece
m
ber
2
01
8
:
1
1
11
–
1
1
16
1116
[15]
Li
u
S,
Tai
H,
Ding
Q,
Li
D,
Xu
L,
W
ei
Y.
“
A
Hy
brid
App
roa
ch
of
Support
Vec
tor
Regr
es
sion
with
Gene
t
ic
Algorit
hm
Opti
m
iz
at
ion
for
Aq
uac
ul
ture
W
a
te
r
Quali
t
y
Pr
edi
c
tion
”
.
Ma
the
mat
ical
and
Comput
er
Mode
l
li
ng
.
201
3;
58(3):
458
-
465
.
[16]
Singh
S,
Gill
J.
“
Te
m
pora
l
W
e
at
her
Prediction
using
Bac
k
Propaga
t
ion
base
d
Gene
t
ic
Algor
i
thm
Te
chn
ique
”
.
Inte
rnational
Jo
urnal
of
In
te
l
l
ig
e
nt
Syst
ems and
A
ppli
cations
.
201
4;
6
(12)
:
55
-
61
.
[17]
W
ei
Koh
J,
Ta
n
TS,
EnChu
ah
Z,
Soh
SS
,
Arif
M,
Le
ong
K.
“
Gene
tic
Algori
t
hm
Optimize
d
Bac
k
Propag
at
io
n
Neura
l
Network
for
Knee
Os
te
o
art
hri
ti
s
Cla
ss
ifica
t
ion
”
.
Re
searc
h
Journal
of
Ap
pli
ed
S
ci
en
ce
s,
Engi
ne
ering
an
d
Technol
ogy
.
201
4;
8(16):
1787
-
1
793.
[18]
Sen
O,
Zhe
ngx
i
ang
S,
Jianhu
a
W
,
Degui
C.
“
Applic
a
ti
on
of
LVQ
Neura
l
N
et
works
Com
bine
d
with
Gen
etic
Algorit
hm
in
Pow
er
Quali
t
y
S
ig
nal
s Cla
ss
ifica
t
i
on
”
.
Proce
ed
ing
s of
The
Inte
rnat
ional
Conf
ere
nc
e
on
Pow
er
Syst
em
Technol
ogy
.
Ku
nm
ing.
2002:
49
1
-
495.
[19]
W
ang
JM
,
W
en
YQ
.
“
Applic
at
i
on
of
Gene
ti
c
LVQ
Neura
l
Network
in
Credit
Anal
y
sis
of
Pow
er
Custom
er
”
.
Proce
ed
ings o
f
Fourth
Int
ernational
Conf
ere
nce on
Natural
Computati
on
.
2008:
305
-
309.
[20]
Chen
N,
Rib
ei
ro
B,
Vie
ira
AS
,
Duart
e
J,
N
eve
s
JC.
“
Hy
br
id
Ge
net
i
c
Algorit
hm
and
Learni
ng
V
e
ct
or
Quant
izati
o
n
Modeli
ng
for
CostS
ensit
ive
Ba
nkruptcy
Pred
iction
”
.
Proce
ed
i
ngs
of
The
Sec
ond
Inte
rnation
al
Confe
renc
e
on
Mac
hine Learni
ng
and
Computi
ng.
Banga
lore
.
2
010:
213
-
217.
[21]
Bake
r
JE.
“
Reduc
ing
Bia
s
and
In
eff
iciency
in
Th
e
Sele
c
ti
on
Algo
rit
hm
”
.
Proceedi
ngs
of
The
Sec
ond
Inte
rnational
Confe
renc
e
on
Gene
tic
A
lgorit
h
ms
.
1987:
14
-
21
.
Evaluation Warning : The document was created with Spire.PDF for Python.