TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol.12, No.7, July 201
4, pp
. 5669 ~ 56
7
7
DOI: 10.115
9
1
/telkomni
ka.
v
12i7.572
3
5669
Re
cei
v
ed Fe
brua
ry 1, 201
3; Revi
se
d March 21, 201
4
;
Accepte
d
April 14, 201
4
Hypertension Expert System with C5.0 Algorithm and
Fuzzy Logic
I Gusti Made
Ngurah
Ardi
Yasa*
1
, I Ketut Gede Darma Putra
2
, Ni Md Ika Marini Mandenni
3
Dep
a
rtment of Information T
e
chno
log
y
,
Ud
ay
an
a
Un
i
v
e
r
si
ty
, Ba
l
i
,
In
do
nesi
a
Bukit Jimbar
an
, Badung, Ba
li, Indon
esi
a
,
T
e
lp. 0361-
78
535
33
*Corres
p
o
n
id
n
g
author, e-ma
i
l
:
ngura
h
.ard
i
y
a
s
a09
61@
gmai
l
.
com
1
, ikgdarmaputra
@gma
il.
com
2
,
ika_m
ade
@
y
a
hoo.com
3
A
b
st
r
a
ct
Expert System is a knowledge-bas
ed system
whic
h is
its
knowledge
is c
o
ming from
the ex
pert
s
itself. This syst
em is
expected to help
users t
o
take
best decisions
in solv
ing th
e pr
oblems they face. This
expert syste
m
can be
app
li
e
d
in var
i
ous fi
elds of
kn
ow
le
dges. Me
dica
l field is
one
ar
ea w
here ex
p
e
rt
system is
nee
ded i
n
cas
e
of
taking d
e
cis
i
o
n
, such as
d
i
a
gnos
ing
an
d treatment.Hyp
er
tensio
n is o
n
e
of
me
dic
a
l dis
e
a
s
e that can b
e
dia
g
n
o
sed fr
om
look
in
g at the patie
nt
’
s
physic
a
l ch
ara
c
teristics and
th
e
patie
nt
’
s
l
i
festyl
e. Besi
de
usi
n
g the
exp
e
rts k
now
led
g
e
in
di
agn
osin
g
proc
ess, w
e
als
o
c
o
mbi
ne
it w
i
th
C5.0
alg
o
rith
ms a
n
d
fu
zz
y
log
i
c to
get prec
ise r
e
s
u
lt. C5.0
alg
o
ri
thm is
use
d
to
create a
dec
isi
on tree
bas
ed
o
n
the ex
perts, w
h
ile
fu
zz
y
l
ogi
c used
to cate
gori
z
e
the
typ
e
of hy
perte
ns
ive d
i
seas
e th
at suffered
by
th
e
patient and
in
crease
the lev
e
l of
accur
a
cy
of the diagnosing system
.
T
he
accuracy
o
f
the co
mbin
at
io
n
betw
een c5.0 a
l
gorit
hm
and fu
zz
y
lo
gic is a
b
o
u
t 97.19%.
Ke
y
w
ords
:
ex
pert system
, hy
pertens
ion, C5.0 algorithm
s
, fu
z
z
y
logic
Copy
right
©
2014 In
stitu
t
e o
f
Ad
van
ced
En
g
i
n
eerin
g and
Scien
ce. All
rig
h
t
s reser
ve
d
.
1. Introduc
tion
As many as
1 billion people
worldwide
suffer from
hypertens
ion. I
n
the
United
States,
nearly 1 in 3
adults
(ap
p
ro
ximately 73 million peo
ple
)
have some
degree of hig
h
blood
pre
ssure.
Hyperte
nsi
o
n
is a
co
ntrib
u
ting facto
r
t
o
many
oth
e
r
di
sea
s
e
s
in
cludi
ng myo
c
ardial
infarcti
on
(MI), stro
ke,
heart failu
re, renal failu
re,
and retin
opat
hy, and is a leadin
g
ca
use
of death [1].
Awarene
ss, treatme
nt, an
d control
of
hyperten
s
io
n
are
subo
ptimal. Only two-thirds of
patients
with hyperten
s
io
n are aware
of their st
atus,
whi
c
h m
ean
s that a la
rge
segm
ent of t
he
popul
ation ha
s hyperte
nsi
o
n that is unreco
gni
z
ed an
d untreate
d
. Even in patients with kno
w
n
hyperten
s
io
n, som
e
a
r
e
no
t treated
for v
a
riou
s
re
ason
s;
in
cludi
ng p
h
ysici
an and
patient
u
nde
r-
recognitio
n
of
the impo
rtan
ce of tre
a
tme
n
t.To mi
nimize the impa
ct
of hyperten
s
i
on, the ne
ed
for
an expert sy
stem for early
diagn
osin
g p
a
tients wi
th h
y
perten
s
ion.
Expert syste
m
s are com
p
uter
prog
ram
s
tha
t
attempt to a
c
hieve the le
vel of
performance in a way that is comparable to solve
probl
em
s wit
h
a human ex
pert in takin
g
a deci
s
io
n [2].
An Expert
system i
s
a
co
mputer
program t
hat
simul
a
tes the
jud
g
m
ent an
d be
havior
of a
human
bein
g
or a
n
o
r
ga
ni
zation th
at ha
s expe
rt kno
w
led
ge a
nd e
x
perien
c
e i
n
a pa
rticula
r
fi
eld.
Typically,
su
ch a
sy
stem
contain
s
a
kn
owle
dge
ba
se of
accum
u
l
a
ted exp
e
ri
e
n
ce
an
d
a
se
t o
f
rule
s for appl
ying the con
d
ition to each particular
situation that is de
scribe
d in the prog
ra
m.
Sophisti
c
ated
expert syste
m
s c
an be e
nhan
ce
d with
addition
s to the kno
w
le
dg
e base or to t
h
e
set of rule
s. In other word
s, it is a
soft
ware
ba
se
d system which
make
s
or ev
aluate
s
de
ci
si
ons
based o
n
rules e
s
tabli
s
h
e
d
within th
e
softwa
r
e [3].
Com
pared t
o
the kno
w
le
dge of h
u
ma
n
experts, the
Expert System Knowle
dge
has the adv
a
n
tage
s of permanent, ea
sy to transfer, e
a
sy
to edit, comp
atible, mode
rate and well suited for fault diagn
osi
s
[4].
In the d
o
mai
n
of h
ealth
ca
re, it i
s
imp
o
r
tant
that th
e
syste
m
i
s
a
c
curate
in
di
agno
sin
g
becau
se it de
als
with lives
of perso
ns
which
ca
n nev
er b
e
re
pla
c
e
d
if an e
rro
r
occurre
d
. Th
ere
are va
riou
s t
e
ch
niqu
es
o
n
implem
enti
ng the
expert system
an
d almo
st u
s
es
accu
rate
and
establi
s
h
ed a
l
gorithm
s. Most of these
use dat
a min
i
ng techni
que
s [5]. In
the first pha
se, the
medical b
a
ckgrou
nd
of di
sea
s
e
s
i
s
re
corded
thro
u
gh the
creati
on of p
e
rso
n
a
l intervie
w
with
docto
rs a
nd p
a
tients. In the second p
h
a
s
e, a set of
rul
e
s is
created
whe
r
e ea
ch
rule co
ntain
s
in
IF part that has the sym
p
toms an
d in T
H
EN pa
rt
that has the di
se
ase that shou
ld be reali
z
e
d
[6].
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 7, July 201
4: 5669 – 56
77
5670
Machi
ne lea
r
ning sy
stems can be u
s
e
d
to develop th
e kno
w
le
dge
bases u
s
e
d
b
y
expert
system
s. Given a set of clinic
al cases
that act as e
x
amples, a
machi
ne lea
r
ning syste
m
can
prod
uce a
systemati
c
de
scriptio
n of tho
s
e
clini
c
al
fea
t
ures that uni
quely charact
e
rise the
clini
c
al
con
d
ition
s
. This
kno
w
led
g
e
can
be exp
r
esse
d in the
form of sim
p
l
e
rule
s, or
often as
a de
cision
tree [7].
Hyperte
nsi
o
n
is a con
d
ition in whi
c
h
a person e
x
perien
c
e
s
a
n
increa
se i
n
blood
pre
s
sure abo
ve normal. Hyperten
s
ion i
s
estim
a
t
ed to cau
s
e 4.5
%
of the global dise
ase bu
rde
n
and i
s
a
s
pre
v
alent in man
y
developing
cou
n
trie
s
a
s
i
n
develo
ped
cou
n
trie
s. World
w
ide,
sev
en
million pre
m
ature de
aths have been attributed to
hyperten
s
io
n. In recent
decad
es, it has
become in
creasi
ngly cle
a
r
that the de
velopment of
stro
ke, isch
emic h
e
a
r
t disea
s
e, a
nd renal
failure have b
een attribute
d
by hypertension [8].
Many p
eople
have
hype
rtensi
on fo
r ye
ars
wit
hout e
v
en
kno
w
ing
it.
Acco
rdin
g torecent
estimate
s, on
e in four a
dult
s
in the Unite
d
St
ates have
hyperten
s
io
n
,
but, becau
se there a
r
e fe
w
symptom
s
, n
early o
nethird
of these pe
o
p
le do
n’t k
n
o
w
they h
a
ve i
t. That is
why
it is called the
“silent killer” [9].
2. Res
earc
h
Method
The type of
dise
ase that is create
d
as th
e ob
ject of re
se
arch is
hyp
e
rten
sion.
Hyperte
nsi
o
n
is
cla
s
sified i
n
to several
types:
P
r
e
Hyp
e
rten
sion, St
age I
Hype
rte
n
sio
n
a
nd St
age
II Hyperte
nsi
on [1]. Expe
rt’s
kno
w
le
dge
is o
b
taine
d
from literature
sou
r
ces
su
ch
as lib
ra
ry bo
oks
and inte
rni
s
t. Basic I
n
formations
ba
se
d on the
re
sults of cli
n
ica
l
examination
and la
boratory
tests. Thi
s
sy
stem’s
kno
w
l
edge
s is represe
n
tat
ed by usin
g the C5.
0
Algorithm a
nd Fuzzy Log
ic.
The
system
i
s
d
e
velop
ed i
n
to web
-
ba
se
d sy
stem
whi
c
h giving
th
e
users ea
se of
acce
ss.
T
h
e
output pro
d
u
c
ed by this system is the bel
ief of
the illness a
nd the st
age of the hyperten
s
io
n.
2.1. C.50
Algo
rithm
C5.0 mod
e
l d
one by splitting the sam
p
l
e
bas
ed on the field that provide
s
the
maximum
information gain. In C5.0 algorithm
s, the selectio
n
of the attrib
utes
pro
c
e
s
sed
using info
rmatio
n
gain. Attributes sele
cted h
euri
s
tically to find pure
s
t attribute
s
, whi
c
h
is giving the most net nod
e.
If one
of the
bran
ch
es of
a de
ci
sion
tre
e
de
rived
fro
m
on
e
cla
ss,
then thi
s
bra
n
ch
is
called
pure
[10]. Information gain i
s
the main
crite
r
ia that
is u
s
ed. So in sel
e
cting attri
b
u
t
es for splitting
obje
c
t
s
in so
me
cl
as
se
s,
we sho
u
ld se
lect
t
he
attri
b
utes that
pro
duce the b
e
st
informatio
n
gain
[11].
The si
ze
of the informatio
n gain i
s
u
s
e
d
to sele
ct th
e test attribut
es at ea
ch
n
ode in the
tree. This si
ze is use
d
to sele
ct the attribute
s
or no
des on the tree. Attribute with the high
est
informatio
n g
a
in value
will
be
sele
cted
as th
e pa
re
nt for the
next
node
[12]. Th
e form
ula to
get
the informatio
n gain value i
s
:
,
,…,
∑
(
1
)
S
is
a
set
t
h
at
co
nsi
s
t
s
o
f
s
sam
p
le
d
a
t
a
. kn
own
a
s
cla
s
s attrib
ute, m d
e
fin
e
s th
e
cla
s
s
e
s in it
,
C
i
(for i =
1, ...,
m), S
1
th
e numbe
r of sampl
e
s of S in class C
i
.
To cla
ssif
y
t
he
sampl
e
s th
at going to be
used,
Use the rul
e
(1
) a
bove to obta
i
n the req
u
ired inform
atio
n.
Acco
rdi
ng to rule (1
) ab
ove, p
i
is a pro
portion of the
output class, same as th
e cla
ss C
i
and
es
timated
by s
i
/s
. Attribut
e A has
spe
c
ific valu
es
{
a
1
, a
2
, ...
, a
v
}. Attribute A can
be u
s
ed
for
partition S into v subsets, {
S
1
, S
2
, .
..,
S
v
}
,
where S
j
co
ntains
sampl
e
s on S wo
rth
aj in A [11].
If A is chose
n
as the test
attribute (for ex
ample, the best attribu
t
e for split), then this
sub
s
et
will be
related
to th
e bra
n
ch of n
ode
set S. S
ij
is the n
u
mb
er of
sampl
e
s in cla
s
s Ci i
n
a
sub
s
et
S
j
. To get the value of a sub
s
et of t
he attribute A, then use formul
a (2
),
∑
…
,…,
(
2
)
…
is the numb
e
r of su
bsets
j
divided by the numbe
r of sample
s
in S. To get the value of
the gain, then
used the fo
rmula [11].
,
,…,
(
3
)
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Hyperte
nsi
o
n
Expert Syste
m
with C5.0 Algorithm
and Fuzzy… (I Gusti Ma
de Ngura
h
Ardi Ya
sa)
5671
Having the first bran
ch d
e
termin
ed, the next pr
ocess
is to do the seco
nd iteratio
n. In the
second iterati
on, the attributes
of the first branch will not be used
in
this process. The process of
determi
ning t
he se
co
nd branch is the same a
s
t
he p
r
eviou
s
meth
od of determi
ning bran
ch. The
iterative process will
conti
nue until the
data ca
n n
o
t be solve
d
anymore. From the vario
u
s
existing bran
che
s
will form
a sch
eme th
at will end in
an initial hypo
thesi
s
.
2.2. Fuzz
y
Logic
Basically, Fu
zzy L
ogi
c is a multivalue
d logi
c, that
allows inte
rm
ediate valu
es to be
defined b
e
tween
conventi
onal evalu
a
tions li
ke true
/false, yes/no
, high/low, et
c. Notio
n
s li
ke
rathe
r
tall or
very fast can
be formul
ate
d
mat
hemati
c
ally and pro
c
essed by co
mputers, in o
r
der
to apply a more hum
an-li
ke
way of thinki
ng in the pro
g
rammin
g
of compute
r
s [13]
.
2.2.1. Kno
w
l
e
dge
Repr
esen
ta
tion
Knowle
dge o
f
clinical sym
p
toms an
d la
borato
r
y re
su
lts of the patient is necessary for
the re
pre
s
e
n
tation of the
s
e fact
s alo
n
g
with a
n
expl
anation
of th
e que
stion
[1
4]. The q
u
e
s
tion
scheme
of th
e sy
stem i
s
o
b
tained
from
the de
ci
sion
tree
thro
ugh
C50
alg
o
rith
m. The
que
st
ions
scheme
will
be end
ed by
giving the result a
bout
whi
c
h type the hype
rten
sion is. La
bo
ratory
results
will
represented
by a fuzzy set with a
mem
bership function. So
me fuzzy
set
s
model
s
h
ow
n
be
lo
w
.
)
(
x
)
(
x
Figure 1. Membershi
p
Cu
rves of Fastin
g Blood
Sugar
Figure 2. Membershi
p
Cu
rves of Blood Sugar
Afte
r
2
H
o
ur
s F
a
s
t
in
g
)
(
x
)
(
x
Figure 3. Membershi
p
Cu
rves of Blood
Chol
este
rol
Figure 4. Membershi
p
Cu
rves of Low
Density
Lipop
rotein (L
DL)
)
(
x
)
(
x
Figure 5. Membershi
p
Cu
rves of High
Density
Lipop
rotein (HDL)
Figure 6. Membershi
p
Cu
rves of Trigli
serida
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 7, July 201
4: 5669 – 56
77
5672
Membe
r
ship functio
n
s of F
a
sting Bloo
d Sugar:
1,
100
105
5
,
100
105
145
5
,
145
150
1,
150
100
5
,
100
105
1,
105
120
125
5
,
120
125
120
5
,
120
125
1,
125
145
150
5
,
145
150
Membe
r
ship functio
n
s of th
e Blood Suga
r After Fastin
g 2 Hou
r
s:
1,
135
145
10
,
135
145
160
10
,
160
170
1,
170
185
195
1
0
,
185
195
135
10
,
135
145
1,
145
160
170
1
0
,
160
170
185
10
,
185
195
1,
195
Membe
r
ship functio
n
s of Blood Chole
s
te
rol:
1,
195
205
10
,
195
205
215
10
,
215
225
1,
225
235
245
10
,
235
245
195
10
,
195
205
1,
205
215
225
1
0
,
215
225
235
10
,
235
245
1,
245
Membe
r
ship functio
n
s of L
o
w Den
s
ity Lipoprotein (LDL)
1,
125
135
10
,
125
135
155
10
,
155
165
1,
165
185
195
10
,
185
195
125
10
,
125
135
1,
135
155
165
1
0
,
155
165
185
10
,
185
195
1,
195
Membe
r
ship functio
n
s of Hi
gh De
nsity Li
poprotein (HDL):
1,
40
45
5
,4
0
4
5
6
0
5
,
6
0
65
1,
65
75
80
5
,7
5
8
0
3
5
5
,3
5
4
5
1,
45
60
65
5
,6
0
6
5
0,
75
7
5
5
,7
5
8
0
1,
80
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Hyperte
nsi
o
n
Expert Syste
m
with C5.0 Algorithm
and Fuzzy… (I Gusti Ma
de Ngura
h
Ardi Ya
sa)
5673
Membership functions of Trigliserida:
1,
140
165
25
,
140
165
190
25
,
190
215
1,
215
490
515
2
5
,
490
515
140
25
,
140
165
1,
165
190
215
2
5
,
190
215
490
25
,
490
515
1,
515
2.2.2.
Implication and Composi
t
ion
Cal
c
ulation
of
the deg
ree o
f
fuzzy memb
ership for ea
ch symptom i
s
determi
ned
by the
value assig
n
e
d
by the user
[15].
Table 1. Fu
zzy Rule
Rule
Num
b
er
Rule
Value
R000001
IF Parent
al history of suffering f
r
o
m
h
y
pe
rtension is
Y
ES THEN
CF = 0.90
R000002
IF Eat Foo
d
With High Salt Levels is often AND Str
e
ss is often THEN
CF = 0.70
R000003
IF Eat Foo
d
With High Salt Levels
is often AND Sleep Apnea is ofte
n THEN
CF = 0.75
R000014
IF Stress is often AND vertigo is often THEN
CF = 0.65
R000035
IF Smoking is ve
r
y
ofte
n AND Ve
r
t
igo is often THE
N
CF = 0.72
R000045
IF
Sistolic B
l
ood Pressure
Norm
al AND
Diastolic Blood Pressure
No
rmal THEN
CF = 0.50
R000057
IF
Sistolic B
l
ood Pressure
High A
ND
Diastolic Blo
od Pressure
Hig
h
THEN
CF = 0.90
R000068
IF
Sistolic B
l
ood Pressure
High T
h
reshold AND
Di
astolic Blood Pre
ssure
Ver
y
High
THEN
CF = 0.85
Based o
n
th
e degree of
membershi
p
, calcul
ate the implicatio
n function
with MIN
function [16]-[
18].
μ
(x) is the
degre
e
of membe
r
ship for x and w
i
is the result of implication.
,
(4)
The process
of comp
ositio
n is
mad
e
to obtain the value z
i
of ea
ch rule. The
certai
nty
value from ex
pert of ea
ch rule is value of
z
i
[14].
2.2.3. Defuzzific
ati
on
Defu
zzifi
catio
n
pro
c
e
s
s is done
usi
n
g
weig
hted a
v
erage
meth
od defu
z
zification by
cal
c
ulatin
g the averag
e value of z
i
[16-18].
∑
∑
(5)
wi is the
re
sult of implica
t
ion and
z
i
i
s
the re
sult
of com
p
o
s
ition
[14]. The
re
sults
of
defuzzificatio
n
demon
strate the value of the be
lief for the syndrome
experien
c
e
d
by patients.
2.2.4.
Cer
t
ainty
Factor Calcula
t
ion
The re
sult of defuzzificatio
n
pro
c
e
ss
will
be
use
d
to calcul
ate the value of belief for the
diagn
osi
s
. Firstly, will be calcul
ated cert
aint
y factor (CF)
seq
uenti
a
l as follo
ws [
16].
,
∗
(
6
)
CF(x,y) i
s
re
sult of
ce
rtai
nt
y factor se
quential,
CF
(x) is
re
sult
o
f
defuzzification a
n
d
CF(y
) is the
expert
ce
rtain
t
y value of e
a
ch
rul
e
.
In t
h
is
study,
CF
sequential m
u
ltiplies with t
he
weig
ht value
of each ph
ase of disea
s
e.
CF
s
equ
enti
a
l from
several rule
s g
ene
rated
com
b
in
ed
usin
g the followin
g
cal
c
ula
t
ion of the co
mbined
CF a
s
follows [17].
,
∗
(
7
)
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 7, July 201
4: 5669 – 56
77
5674
The re
sult
s of combin
ed
CF sugge
st the di
ag
no
sis of the disea
s
e to the sy
mptoms
experie
nced
by patients [1
4].
3. Resul
t
s
and
Discus
s
ion
The mai
n
obj
ective of this
resea
r
ch is to
create
a
n
ex
pert system whi
c
h can be
use
d
for
diagn
osin
g h
y
perten
s
ion.
The expe
ctat
ion of th
is
system is to
help u
s
e
r
s t
o
overcom
e
the
effects of hyp
e
rten
sion by telling
them e
a
rly about the
i
r con
d
ition.
3.1. Kno
w
l
e
dge
Acquisi
tion
The Kno
w
led
ge Acqui
sitio
n
role is tra
n
sferrin
g
kn
owl
edge that gai
ned from
com
p
leting a
variety of lite
r
ature a
n
d
ex
perts
kno
w
in
g into
a
kn
o
w
led
ge
ba
se.
Thi
s
kno
w
le
dge
ba
se
will
be
the mo
st imp
o
rtant
part
of
an
expe
rt system for
sto
r
ing
all th
e
knowl
edge
to
be u
s
e
d
a
s
a
stand
ard fo
r deci
s
io
n maki
ng.
3.2. Kno
w
l
e
dge
Repr
esen
ta
tion
The first thin
g that is don
e in rep
r
e
s
en
ting the kno
w
ledge i
s
start
ed from ge
ne
rating a
deci
s
io
n
tre
e
usin
g C5.0 al
gorithm. The
pro
c
e
s
s
of
fo
rming
a de
cision
tree start
from sele
ctin
g
the initial tree bran
ch
es.
This b
r
an
ch i
s
obtain
ed from the value
of the attribute that has
th
e
highest gai
n
of all the
existing attributes. In this
case we
use 15 attributes
that
will be used i
n
gene
rating
a
deci
s
io
n tre
e
. The
de
cisi
o
n
tree
ma
de
up three
-
pron
ged
bra
n
che
s
, that in
eve
r
y
bran
ch late
r i
s
the re
sult of
the questio
n
in the previou
s
bra
n
ch.
Figure 7. De
cisi
on Tree
After determi
ning th
e first bra
n
ch, the
pro
c
e
s
s conti
nue
s to th
e
next iteration
.
In the
second iterati
on, the first bran
ch attributes will not
be used in
this process. The process o
f
determi
ning t
he second b
r
an
ch i
s
the
same
as th
e previo
us
method to d
e
termin
e the
first
branch. Iterat
i
ve process
will cont
inue
until the data can not be
solved again.
From the vari
ous
existing bran
che
s
will form
a sch
eme th
at
will ende
d up by giving an initial hypo
thesi
s
.
Figure 8. Con
s
ultation
User Interface
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Hyperte
nsi
o
n
Expert Syste
m
with C5.0 Algorithm
and Fuzzy… (I Gusti Ma
de Ngura
h
Ardi Ya
sa)
5675
The fu
zzifi
cat
i
on process i
n
the dia
gno
sing
hy
perte
n
s
ion i
s
divid
e
d
into 2 p
h
a
s
es. Th
e
first ph
ase be
gins
with p
r
o
c
e
ssi
ng the
d
a
ta proxim
ity of the de
ci
sio
n
tree
with th
e data from t
he
kno
w
le
dge b
a
se that ha
s unde
rgo
ne th
e fuzzificati
o
n
pro
c
e
ss. T
he data we
re
fed by the u
s
er
will b
e
mat
c
hed
with
hypotheses that
have b
een
made
by fuzzy logi
c. Th
e
next process i
s
pro
c
e
ssi
ng t
he re
sult of
a labo
rato
ry data t
hat
may be o
w
n
ed by the u
s
er.
Data fro
m
a
laboratory re
sults in numb
e
r
s
will be gro
uped into
sev
e
ral cl
asse
s.
Figure 9. Co
nsultatio
n
User Interfa
c
e
With Fu
zzy L
ogic
Results from
each of these classes
will
be pr
ocessed by implicating classes from every
existing laboratory results.
3.3. Sy
stem
Performance
The pe
rform
a
nce of the
expert sy
stem i
s
obtai
n
ed fro
m
the com
p
a
r
iso
n
of the result
s of
the diagno
si
s made by a real expert wit
h
the di
agno
se
s given by the expert sy
stem. Base
d on
the stru
ctu
r
e
of the syst
em
above, this
expert sy
ste
m
wa
s te
ste
d
in many different
ca
se
s. One
example of consultation a
nd diag
no
sis
of
the system
is given in Figure 1
0
.
Figure 10. Result
s of Con
s
ultation
Total sam
p
le
of case
s u
s
ed to determ
i
ne t
he accu
racy of the data is mea
s
u
r
ing the
accuration
p
e
rcentag
e by
com
p
a
r
ing
900 d
a
ta
sa
mples. The data will
be
pro
c
e
s
sed u
s
ing
algorith
m
c5.
0
which will
g
enerating a
d
e
ci
sion
tree.
De
cisi
on
tree
will be used as a kno
w
le
d
ge
base sy
stem. Grievan
c
e
s
t
hat perceive
d
by use
r
s
will
be matched
with a d
e
ci
sio
n
tree, that will
give a con
c
lu
sion a
bout th
e illness suffered by us
ers.
The pe
rform
a
nce of the ex
pert sy
stem with
C5.0 alg
o
rith
m can be
see
n
in Table 2.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 7, July 201
4: 5669 – 56
77
5676
Table 2. Expert Systems
Perform
a
n
c
e
with C5.0 al
g
o
rithm
Scheme Case
T
o
tal Cases
Same Case Wi
th
Scheme Correct
Wrong
Accuracy
(%
)
1 900
50
44
6
88
2 900
39
37
2
95
3 900
48
45
3
93.75
4 900
49
48
1
98
5 900
64
60
4
93.75
6 900
60
57
3
95
7 900
57
57
0
100
8 900
38
37
1
97.22
9 900
65
65
0
100
10 900
40
38
2
95
Average of the di
fference result
95.57
Expert Syste
m
Perfo
r
man
c
e
with a
co
mbination
of
C5.0 a
nd fu
zzy logi
c al
gori
t
hm ha
s a
b
e
tte
r
a
c
c
u
r
a
c
y
ra
te
th
an
th
e
us
e o
f
two me
th
od
s se
pa
r
a
te
ly. T
h
e
c
o
mb
in
a
t
io
n
c
a
n impr
o
v
e
th
e
accuracy
of d
i
agno
se
s
pre
d
icted
by the
expert
sy
ste
m
. The
expe
rt system
pe
rforma
nce
with
a
combi
nation
of C5.0 algo
ri
thms and fu
zzy logic
can b
e
see
n
in Tab
l
e 3.
Table 3. Perf
orma
nce with
a Combin
ation of C5.0 Algorithm
s an
d Fuzzy Logi
c
Case
S
y
stem Diagnosis
F
u
zz
y
(%
)
S
y
stem Diagnosis
Fuzz
y
& C50
(%
)
Expert
Diagnosis (%)
The difference
res
u
l
t
(%)
Fuz
z
y
The difference
r
e
sult F
u
zz
y
&
C5.0 (%)
1 59
58
55
6.78
5.17
2 69
71
72
4.17
1.39
3 38
37
35
7.89
5.41
4 77
74
75
2.6
1.33
5 61
61
64
6.25
4.69
6 66
62
64
4.47
3.12
7 69
71
70
2.78
1.41
8 67
64
65
2.99
1.54
9 74
71
72
4
1.39
10 71
76
74
4.05
2.63
Average of the di
fference result
4.60
2.81
4. Conclu
sion
Expert syste
m
for dia
gno
sing
Hype
rte
n
sio
n
ha
s b
een d
e
velop
ed into a
web-b
a
sed
platform to
receive fe
ed
in a form
of physi
ca
l
symptoms
and
laboratory t
e
st result
s. the
que
stion
s
systematic th
at a
r
ise
s
i
s
a d
e
cision
tree
de
ri
ved from
p
r
o
c
e
ssi
ng th
e required
data
by
usin
g the
C5.0 alg
o
rithm.
Fuzzy rule
s relating
ea
ch dise
ase symptoms usi
ng expert ce
rtai
nty.
This sy
stem provides an
output of hypertension
illness possibilit
y that su
ffered by users. The
combi
nation
of C50
with f
u
zzy logi
c alg
o
rithm
in
crea
se the
expert
syst
em hyp
o
t
hese
s
a
c
curacy.
Fr
om s
t
atis
tical Per
f
or
manc
e
acc
u
r
a
c
y
of c
5
.0 is 95.7% and Fuzzy logic
95.4%. Tes
t
r
e
s
u
lts
with a
combi
nation
of c5.0 alg
o
rithm
with fu
zzy
lo
gic sy
st
e
m
s
how
s
th
at th
e devel
ope
d
expert
sy
st
em h
a
s 9
7
.
19%
ac
cu
ra
cy
of
the simil
a
rity to a real expert.
Referen
ces
[1]
Jeffery
Martin.
Hy
p
e
rtens
ion
Guidel
ines: R
e
visitin
g
the J
NC 7 Recom
m
end
ations.
T
he Jour
nal
o
f
Lanc
aster Gen
e
ral H
o
spita
l
. 2008; 3(3): 9
1
-9
7.
[2]
Josep
h
, Ha
ns-
D
. W
eb-Base
d
Expert S
y
ste
m
for
Class
ific
ation of
Indust
r
ial and Com
m
ercial
W
a
st
e
Products.
Jour
nal of E
m
erg
i
n
g
T
r
ends in C
o
mp
utin
g an
d Informati
on Sci
e
nces
. 201
1; 2(
6): 357-2
62.
[3]
Josep
h
in
e, Je
ya
ba
lara
ja. E
x
pert S
y
stem a
nd Kn
o
w
l
e
d
g
e
Mana
geme
n
t
for Soft
w
a
re
Devel
o
p
e
r i
n
Soft
w
a
re Companies. International
Jour
na
l of
Informatio
n
and Co
mmu
n
ic
ation
T
e
c
h
n
o
lo
gy
Res
earch
.
201
2; 2(3): 243
-247.
[4]
F
eng Yo
ng
xi
n
,
Yang T
ao.Stud
y
of F
ault
Dia
g
nos
is Method for W
i
nd T
u
rbine
with Decis
i
o
n
Classific
a
tio
n
Algorit
hms an
d Expert S
y
s
t
em.
T
E
LKOMNIKA Indon
esia
n J
our
nal
of Electric
a
l
Engi
neer
in
g
. 2012; 10(
5): 905
-910.
[5]
Rome
o Mark,
Jae
w
a
n
Le
e
.
Healthc
a
re
Ex
p
e
rt S
y
ste
m
base
d
o
n
Group
Coo
p
e
ratio
n
Mo
del.
Internatio
na
l Journ
a
l of Softw
are Eng
i
n
eeri
n
g and Its Appl
i
c
ation.
20
08; 2
(
1): 105-1
16.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Hyperte
nsi
o
n
Expert Syste
m
with C5.0 Algorithm
and Fuzzy… (I Gusti Ma
de Ngura
h
Ardi Ya
sa)
5677
[6]
Santosh
Kum
a
r, Dipti
Prav
a. An E
x
pert
S
y
stem
f
o
r Di
agn
osis
of H
u
man D
i
se
ases
. Internatio
na
l
Journ
a
l of Co
mputer App
lic
ations
. 201
0; 1(1
3
): 71-73.
[7]
Prasad
l, Krish
na. An Ap
pro
a
ch to Dev
e
l
o
p
Expert S
y
stems in Me
dic
a
l Di
ag
nosis
Using M
a
chi
n
e
Lear
nin
g
Alg
o
rithms (Asthma)
and A Performance Stud
y.
Internatio
na
l Journ
a
l on Soft
Computi
n
g
(IJSC)
. 2011; 2
(
1): 26-33.
[8]
S
y
er
Re
e T
ee, Xi
n Y
un T
eoh.
T
he Prevale
n
c
e
of
H
y
p
e
rte
n
si
on
and
Its Associate
d
R
i
sk F
a
ctors In T
w
o
Rural Communities In
Pena
ng
, Mala
y
s
ia.
IeJSME.
2010: 4(
2): 27-40.
[9]
Christin
e L
a
in
e ed. L
i
vin
g
W
i
th H
y
perte
n
s
ion.
Americ
a
n
Col
l
e
ge of
Ph
y
s
icia
ns a
nd Micro
lif
e
Corp
oratio
n. Report num
ber: 4. 2004.
[10]
Prof. Nilima P
a
til, Prof Rekh
a.
Compar
ison
of C5.0 & CART
Classification al
gorit
hms usin
g pru
n
i
n
g
techni
qu
e
. Internatio
nal Jo
urn
a
l of Engi
ne
er
i
ng Res
earch &
T
e
chnol
ogy (IJERT
)
. 2012; 1
(
4): 1-5.
[11]
Erna
w
a
t
i
. Pre
d
iksi Status
Keaktifan Stu
d
i Mah
a
sis
w
a
Den
gan A
l
g
o
ritma C5.
0
Dan K-N
ear
es
t
Neig
hb
or T
hesis. Bogor:
Postgrad
uate IPB. 200
8.
[12]
Mehme
d
Kant
ardzic. D
a
ta M
i
nin
g
: Co
nce
p
ts, Models, Met
hods, a
nd A
l
g
o
rithms.
IEEE Press, A joh
n
Wiley & Sons, Inc, Public
atio
n
. 2003: 1
77-1
7
8
.
[13]
LA Z
adeh. Mak
i
ng com
puters
think lik
e pe
opl
e.
IEEE. Spectrum
.1
99
8; 21(8
)
26-32.
[14]
Darma P
u
tra.
F
u
zz
y
E
x
pert S
y
stem for
T
r
opical Infecti
ous
Diseas
e
b
y
C
e
rtaint
y F
a
ctor.
TELKOMNIKA
Indon
esi
an Jou
r
nal of Electric
al Eng
i
ne
eri
n
g
.
2012; 1
0
(4): 8
25-8
36.
[15]
Putu Manik Pri
hatin
i, Katut Gede D
a
rma Pu
tra.
Expert kno
w
l
e
d
ge b
a
se
d s
y
stem
w
i
th u
n
c
ertain
it
y
f
o
r
T
r
opical Infections Di
eseas
ed
Diag
nosis
. IJSCSI
. 2012; 3(4)
: 157-16
3.
[16]
Sivan
and
am S
N
, Sumathi S,
Dee
pa SN. I
n
troducti
on to
F
u
zz
y
L
o
g
i
c u
s
ing MAT
L
AB. Ne
w
Y
o
rk:
Sprin
ger. 20
07
.
[17]
Neg
nevitsk
y M
i
cha
e
l. Artifici
al
Intelli
ge
nce A
Gu
ide
to Intel
lige
n
t S
y
stems
Secon
d
E
d
itio
n. Engl
an
d
:
Pearso
n Educ
ation. 20
02
[18]
Konar Amit. A
r
tificial Intel
l
i
g
e
n
ce a
nd So
ft Comp
uting
B
e
havi
o
ral and Cog
n
itive
M
o
d
e
lli
ng of
the
Huma
n Brain.
W
a
shin
gton D
C
: CRC Press. 2000.
Evaluation Warning : The document was created with Spire.PDF for Python.