Indonesian J
ournal of Ele
c
trical Engin
eering and
Computer Sci
e
nce
Vol. 1, No. 3,
March 20
16, pp. 607 ~ 6
1
8
DOI: 10.115
9
1
/ijeecs.v1.i3.pp60
7-6
1
8
607
Re
cei
v
ed
No
vem
ber 2, 20
15; Re
vised Janua
ry 2
6
, 20
16; Accepted
February 10,
2016
Belief-Rule-Based Intelligent Decision System to Select
Hospital Location
Md. Mahbub
ul Islam
1
*, T
a
njim Mahmud
2
, Mohammad Shahad
at Ho
ssain
3
1
Departme
n
t of Computer Sci
ence a
nd En
gi
neer
ing, Un
iver
sit
y
of Chitta
go
ng, Chittag
o
n
g
, Bangl
ad
esh
2
Department o
f
Computer Sci
ence a
nd En
gi
neer
ing, T
e
xtil
e Engi
ne
erin
g Coll
eg
e, Noak
hali
3
Departme
n
t of Computer Sci
ence a
nd En
gi
neer
ing, Un
iver
sit
y
of Chitta
go
ng, Chittag
o
n
g
, Bangl
ad
esh
e-mail: ma
hbu
bcse@c
u.ac.b
d
1
*, tanji
m
_cs
e
@yah
oo.co
m
2
, hossain_m
s@cu.ac.bd
3
A
b
st
r
a
ct
T
he g
e
n
e
ral
pub
lic
’
s
d
e
m
a
nd
of Ba
ng
la
desh
for s
a
fe
he
alth
is r
i
s
i
ng
pro
m
ptly
w
i
th the
improve
m
ent
o
f
the l
i
vin
g
sta
ndar
d.
How
e
v
e
r, the
all
o
cati
on
of li
mited
a
nd
unb
al
ance
d
med
i
cal
reso
u
r
ces
is deteri
o
ratin
g
the assuranc
e
of sa
fe health
of the peopl
e. T
herefore,
the
new
hospital
constructio
n
w
i
th
ration
al a
lloc
a
ti
on of res
ourc
e
s is i
m
min
ent
and s
i
gn
ifica
n
t.
T
he site sel
e
c
t
ion for esta
bl
i
s
hin
g
a
hosp
i
tal is
one
of the
cru
c
ial
pol
icy-rel
a
t
ed
decis
ions
taken
by
pla
n
n
e
rs an
d
pol
icy
mak
e
rs. T
he
p
r
ocess of
hos
p
i
tal
si
te
sel
e
cti
o
n
i
s
i
n
h
e
r
e
n
t
l
y
co
mp
l
i
c
a
t
e
d
beca
u
s
e o
f
thi
s
i
n
vo
l
v
e
s
ma
n
y
fa
cto
r
s to
b
e
me
a
s
u
r
e
d
and
eval
uate
d
. T
h
e
s
e factors
are
expr
essed
bo
th in
ob
je
ctive
an
d su
bjectiv
e
w
a
ys w
here
as a
hi
erarc
h
i
c
a
l
relati
onsh
i
p
ex
ists a
m
o
ng t
h
e factors. In
add
ition,
it
is
difficu
lt to
measur
e q
u
a
lita
t
ive factors
in
a
qua
ntitative
w
a
y, resu
lting
i
n
co
mp
le
ten
e
ss
in
data
a
nd
henc
e, u
n
cert
aint
y. Bes
i
des
it is
esse
ntia
l t
o
addr
ess the su
bject of u
n
cert
ainty by us
in
g
apt metho
dol
o
g
y; otherw
i
se, the d
e
cisi
on to
choos
e a su
ita
b
le
site w
ill bec
o
m
e ina
p
t. Therefore, this pa
per
de
mo
nstr
ates the ap
pl
icatio
n
of a
nove
l
met
hod n
a
m
e
d
be
l
i
ef
rule-b
ase
d
i
n
f
e
renc
e
meth
o
dol
ogy-RIMER
bas
e i
n
tell
i
g
e
n
t dec
isio
n sy
stem (IDS), w
h
ich
is c
apa
bl
e of
addr
essin
g
suit
abl
e site for h
o
s
pital by tak
i
n
g
accou
n
t of
lar
ge n
u
m
ber
of criteria, w
here t
here ex
ist factors
of both subj
ective an
d ob
jecti
v
e nature.
Ke
y
w
ords
:
MCDA, uncertainty, belief rule base, evid
ential reasoning,
inte
lligent decis
ion system
1. Introduc
tion
Ho
spital
s a
r
e on
e of
th
e mo
st im
p
o
rtant i
n
fra
s
tructu
ral
obj
e
c
ts. T
he i
n
crea
sing
popul
ation, espe
cially in d
e
veloping
cities, am
plifie
s the deman
d for ne
w hospi
tals. Whe
n
we
attempt to s
e
lec
t
suitable
s
i
te for hos
p
ital, it invo
lves multiple criterions
su
ch a
s
,
l
o
cat
i
o
n
,
saf
e
t
y
,
environ
ment, parking
sp
ace, Land
co
st, Risk, tran
sp
ortation
co
st and utility co
st etc. whi
c
h
are
quantitative
and q
ualitative in nat
ure
[20] [21]. Nume
ri
cal d
a
ta whi
c
h u
s
e
s
num
be
rs is
con
s
id
ere
d
a
s
qua
ntitative data and ca
n be mea
s
u
r
ed
with 10
0%
certai
nty [4]. On the co
ntrary,
qualitative d
a
ta is de
scriptive in
n
a
ture,
whi
c
h
define
s
so
me con
c
ept
s o
r
im
pre
c
ise
c
h
arac
teris
t
ics
or
quality of
things
[5].
Hence, th
is dat
a can’t d
e
scri
be a
thing
wit
h
certai
nty si
nce
it lacks the preci
s
io
n and
inherit
s ambi
guity,
ignora
n
ce, vague
n
e
ss. Con
s
eq
uently, it can be
argu
ed that qualitative da
ta involves unce
r
tainty
sin
c
e it is difficult to measu
r
e co
ncepts
or
cha
r
a
c
teri
stics o
r
q
uality o
f
a thing
with
100%
ce
rtai
nty. “Quality
of Lo
cation
” i
s
an
exampl
e
o
f
equivo
cal term sin
c
e it is
an exampl
e o
f
linguisti
c
term. Hen
c
e, it is difficult to e
x
tract its
correct
sema
ntics (m
eanin
g
). Ho
wever, this ca
n be evaluat
ed usi
ng so
me refe
rentia
l value su
ch
as
excelle
nt, good, avera
ge
and ba
d. Th
erefo
r
e, it
ca
n be seen th
at qualitative crite
r
ion
s
which
have b
een
co
nsid
ere
d
in
selectin
g h
o
spi
t
al locati
on i
n
volves lot
of
uncertaintie
s
and th
ey sho
u
ld
be tre
a
ted
wi
th app
rop
r
iat
e
metho
dolo
g
y is
RIMER,
whi
c
h i
s
co
nne
ct to Evidential rea
s
oni
ng
(ER) i
s
a mu
lti-crite
r
ia de
cision an
alysi
s
(M
CDA) me
thod [13] [14]. ER deals with pro
b
lem
s
,
con
s
i
s
ting of
both q
uantit
ative and
qu
alitative cr
ite
r
ia u
nde
r va
riou
s u
n
certa
i
nties
su
ch
as
incom
p
lete i
n
formatio
n, vague
ne
ss, a
m
biguity [7]. The ER ap
proa
ch, deve
l
oped ba
se
d
on
deci
s
io
n theo
ry in pa
rticul
a
r
utility theory
[1] [11],
artificial intelli
gen
ce in
pa
rticul
ar the th
eory
of
eviden
ce [9] [10]. It uses
a belief structure to
mod
e
l
a judgme
n
t with un
certa
i
nty. Qualitative
attribute such
as location o
r
safety nee
d
s
to be
evalu
a
ted usi
ng so
me lingui
stic
referential val
u
e
su
ch a
s
ex
cellent, avera
ge, goo
d an
d bad
etc [2
0] [21]. This requi
re
s hu
man jud
g
me
nt for
evaluating th
e attribute
s
based o
n
th
e mentio
ned
re
fe
rential v
a
lue. In thi
s
way, the i
ssue of
uncertainty can be ad
dressed a
nd mo
re accu
rate
a
nd ro
bu
st de
cisi
on can be
made. The b
e
lief
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 25
02-4
752
IJEECS
Vol.
1, No. 3, March 20
16 : 607 – 618
608
rule
-ba
s
e
d
inf
e
ren
c
e
meth
odolo
g
y-RIM
E
R [15] h
a
s
addresse
d
su
ch i
s
sue
by p
r
opo
sin
g
a
b
e
lief
stru
cture whi
c
h a
ssi
gn
s de
gree of beli
e
f in the
variou
s refere
ntial values of the attribute
s
.
In se
ction 2
will briefly
rep
r
e
s
ent be
lief rule ba
se infere
nce
methodol
ogy-RIMER.
Section 3
will
demonstrate the app
lication of BRB in hospital
site
selection
assess
ment problem.
Section 4 wi
ll represent the re
sults a
nd achievem
ent. Finally se
ction 5 wil
l
con
c
lude t
he
r
e
sear
ch.
2. Literature
Rev
i
e
w
In
RIMER, Belief
Rule Base (BRB)
c
an
capt
ure com
p
licate
d
nonlin
ea
r cau
s
al
relation
shi
p
s betwee
n
a
n
tece
dent at
tributes
and
con
s
eq
uent
s, whi
c
h a
r
e
not possibl
e in
traditional IF
-THEN
rule
s.
BRB is u
s
e
d
to model d
o
m
ain spe
c
ific kno
w
le
dge
u
nder
un
ce
rtai
nty,
and th
e ER
approa
ch i
s
employed
to
facilitate
infe
rence. Thi
s
section
intro
d
u
ce
s B
R
B a
s
a
kno
w
le
dge re
pre
s
entatio
n schema u
nde
r uncertai
n
ty
as well as inf
e
ren
c
e p
r
o
c
e
dure
s
of RIM
E
R.
2.1. Modeling Domain Kno
w
l
e
dge u
s
ing BRB
Belief
Rul
e
s are
th
e key constitue
n
ts of
a
BR
B,
whi
c
h in
clude
beli
e
f deg
ree.
Th
is i
s
the
extended fo
rm of tradition
al IF-THE
N rules. In a b
e
lief rule, ea
ch ante
c
ede
nt attribute takes
referential va
lues a
nd ea
ch possibl
e consequ
ent
is associ
ated
with belief d
egre
e
s [1
5]. The
kno
w
le
dge re
pre
s
entatio
n para
m
eters a
r
e rul
e
wei
g
h
t
s, attribute weights an
d b
e
lief degree
s in
con
s
e
que
nt a
ttribute, whi
c
h are n
o
t ava
ilable in
tra
d
itional IF
-THE
N rule
s. A be
lief rule
can
be
defined in the
following
wa
y.
:
∩
∩
…
……∩
,
,
,
,………
,
(1)
:
0
,
1
with a rule
we
ight
k
, attribute
weig
hts
k1
,
k2
,
k3
, ……,
KT
k
{1,……,
L
}
Whe
r
e
P
1
, P
2
, P
3
…P
Tk
rep
r
esent
the ante
c
e
dent attrib
utes in
the
kith rule.
1
,
…
…
,
,
1
,……,
represents
one of the referentia
l val
ues of the ith antecede
nt
attribute
P
i
i
n
the
k
th rule.
C
j
i
s
o
ne
of the
con
s
e
quent
refe
re
nce
value
s
o
f
the belief
rule.
1
,……,
,
1
,
……,
is one of th
e belief degrees to whi
c
h
the con
s
eq
ue
nt refere
nce
value
C
j
is be
lieved to be true. If
∑
1
the kth rule is said t
o
be compl
e
te; otherwi
se, it is
incom
p
lete.
T
k
is the
total n
u
mbe
r
of
ant
ece
dent
attrib
utes
used i
n
kth
rule
L
i
s
t
he n
u
mbe
r
of all
belief rules in
the rule ba
se
.
N
is the nu
mber of all po
ssi
ble con
s
eq
uent in the rul
e
base.
For example
a belief
rule t
o
assess accessib
ility term
for hospital
can
be
written in the
followin
g
way
.
,
0.0
0
,
,
1.
00
,
,
0,00
(2)
Whe
r
e
{(Excellent, 0.00
), (Go
od, 1.0
0
)
, (A
verage,
0.00)}
is a
b
e
lief di
stributi
on for
acce
ssi
bility consequ
ent, stating t
hat the degree of beli
e
f associate
d
with Excellen
t
is 0%, 100
%
with G
ood
a
nd 0%
with
Average. In
this
belief
rul
e
, the total d
egre
e
of
beli
e
f is
(0
+1
+0) =1,
hen
ce that the asse
ssm
en
t is complete.
2.2. BRB Inferenc
e usin
g ER
The ER
app
roa
c
h [7] [18
]
develope
d
to handl
e m
u
ltiple attribu
t
e deci
s
io
n
analysi
s
(MADA) prob
lem having
b
o
th qualitativ
e and
qua
ntit
ative attributes. Diffe
rent
from tra
d
ition
a
l
MADA app
ro
ach
e
s, ER p
r
esents MA
DA problem
b
y
using a d
e
ci
sion mat
r
i
x
, or a belief
Evaluation Warning : The document was created with Spire.PDF for Python.
IJEECS
ISSN:
2502-4
752
Belief-Rule-B
ased Intelligent Deci
sion
System
to Select Hospital Location
(Md. M
ahbu
bul Isla
m
)
609
expre
ssi
on
matrix, in which e
a
ch a
ttribute
of an alternative
describ
ed
by a distrib
u
tion
asse
ssm
ent
usin
g a b
e
li
ef stru
cture. The infe
ren
c
e p
r
o
c
ed
ures in B
R
B i
n
feren
c
e
system
con
s
i
s
ts
of variou
s
com
p
o
nents such
a
s
inp
u
t tra
n
sf
ormatio
n
, rul
e
a
c
tivation
weig
ht cal
c
ul
ation,
rule u
pdate
mech
ani
sm, followed by th
e agg
reg
a
tion
of the rule
s o
f
a BRB by u
s
ing E
R
[15] [16]
[18] [24]. The input tran
sformation of
a value of an a
n
tece
dent attrib
ute P
i
con
s
ist
s
of distri
buti
n
g
the value into
belief de
gree
s of different
referentia
l val
ues
of that a
n
tece
dent. T
h
is i
s
eq
uival
ent
to transfo
rmi
ng an input into a distribu
tion on re
ferential values
of an antece
dent attribute
by
usin
g their
correspon
ding
belief deg
re
es [14] [
24]. The ith value
of an ante
c
edent attrib
ute at
instant
p
o
int in
time can equivalently be
tra
n
sf
o
r
m
ed
into a
di
stribution over
the
refere
ntial
values, defin
ed for the attribute
by usin
g their belief
degree
s.
The input val
ue of
P
i
,
which is the
i
th antece
dent attribute of a rul
e
, along with
its belief
degree
i
is shown belo
w
by equation (3). The beli
e
f degre
e
i
to the input valu
e is assig
ned
by
the expert in this re
se
arch.
H
(
P
i
,
i
) =
{(
A
ij
,
ij
)
, j
=
1
,…j
i
}
, i = 1,……,T
k
(3)
Here H is u
s
e
d
to sho
w
the
assessme
nt of the belief degre
e
assig
n
ed to the inpu
t value
of the
ante
c
e
dent attri
bute
.
In the
abov
e eq
uation
A
if
(ith val
ue) i
s
the
j
th
referential value of
the
input
P
i
.
ij
is the belief de
gree to the re
ferential valu
e
A
if
with
ij
0.
∑
1
1
,…,
,
and
j
i
is the n
u
mbe
r
of the referential val
ues.
For
example,
the input 0.82 for A
c
cessi
bility
is equivalently transf
ormed to {(E
x
cellent,
0.81),
(Go
od,
0.19),
(Average, 0.0
0
)}.
The in
put
val
ue
of an ant
ece
dent attribute
i
s
colle
cted
from the expe
rt in terms of lingui
stic valu
es such
as ‘E
xcellent’, ‘Go
od’, ‘Average’
and ‘Bad’. This
lingui
stic valu
e is then
a
ssi
gned
de
gre
e
of belief
i
by takin
g
a
c
cou
n
t of expe
rt j
udgme
n
t. Thi
s
assign
ed d
e
g
ree
of b
e
lie
f is then
di
stributed i
n
te
rms of beli
e
f deg
ree
ij
of the differe
n
t
referential va
lues
A
if
[Excellent, Go
od,
Averag
e, Ba
d] of
the ant
ece
dent attribute.
The
ab
ove
pro
c
ed
ure of input tran
sformation is ela
bor
ate
d
by eq
uation
s
(4 an
d 5) given bel
ow.
Ho
wever,
wh
en a ho
spital
is located 1.
1 km
of the
place, it can
be both ex
cel
l
ent and
averag
e. Ho
wever,
it is i
m
porta
nt for
us to
kno
w
,
with
what
de
gree
of
belief
it is excell
en
t and
with what de
gree of belief
it is average
. This
phen
o
m
enon
can b
e
cal
c
ulated
with the following
formula.
i
n
i
n
i
n
n
n
i
n
h
i
h
h
h
,
,
1
,
1
1
,
1
,
,
(
4
)
i
n
i
n
h
h
h
if
,
1
,
(
5
)
Here, the de
gree
of belief
i
n
,
is asso
ciate
d
with the ev
aluation g
r
a
d
e
‘avera
ge’ while
i
n
,
1
is asso
ciate
d
with the up
per level ev
al
uation grade i
.
e. excellent. The value of h
n+1
is the
value related
to excell
ent,
whi
c
h i
s
co
nsidere
d
a
s
1km i.e. the l
o
cation of th
e h
o
spital.
The
value
of
1
n
h
is related
to averag
e, whi
c
h is 1.5
km.
He
nce, applying eq
uat
ion (2
) the di
stributio
n of
the de
gre
e
of
belief
with resp
ect to
1.3
Km of
the l
o
cation
of the
hos
pital can
be a
s
se
ssed
by
usin
g equ
atio
n (2) a
nd the
result is given
below:
{(Excell
ent, 0.4), (Go
od, 0.6), (Avera
ge, 0), (Bad, 0)}
W
h
en
th
e
k
t
h
rul
e
i
s
a
c
ti
vated, the
weight of
activ
a
tion of th
e
k
th rule,
W
k
W
k
is cal
c
ulate
d
b
y
usin
g the flowing form
ul
a [17] [18] [24].
∑
∏
∑
∏
∏
and
,
…
(6)
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 25
02-4
752
IJEECS
Vol.
1, No. 3, March 20
16 : 607 – 618
610
Whe
r
e
is the relative
weight of
P
i
used i
n
the
k
th
rule
, which is
cal
c
ulate
d
by di
viding weight
of
P
i
with m
a
ximum wei
ght
of all the
ant
ece
dent attr
i
b
utes
of the
kth rul
e
. By doi
ng so, the val
u
e
of
be
com
e
s no
rmali
z
e,
meanin
g
that
the ra
nge
of
its value
sh
ould b
e
bet
ween 0
and
1
.
∏
is the
co
mbi
ned m
a
tchi
ng
deg
ree,
whi
c
h i
s
cal
c
ulat
ed by u
s
in
g
multiplicative
aggregatio
n functio
n
.
Whe
n
the
k
th rule as given
in.(1) is a
c
tivated,
the inco
mpletene
ss o
f
the conse
q
u
ent of
a rule
ca
n al
so
re
sult f
r
o
m
its
ante
c
e
dents du
e to
lack of d
a
ta
. An in
compl
e
te inp
u
t for an
attribute
will lead to a
n
in
complete o
u
tp
ut in ea
ch
of t
he rul
e
s in
which th
e attrib
ute is u
s
e
d
. T
h
e
origin
al beli
e
f deg
ree
in the
i
th co
nsequ
ent
C
i
of the
k
th rul
e
is
up
dated b
a
sed
on the
actu
al
input inform
ation as [15] [1
7] [18] [24].
∑
,
∑
∈
∑
,
(7)
Whe
r
e
,
1,
0
1
,
…
Her
e
is the o
r
iginal b
e
lief degree an
d
is the upd
ated
belief degree
.
Due to the incompl
e
te input for ‘Accessibility’,
the belief degree
of the co
nnected rul
e
s
needs to
be modified t
o
sho
w
the in
compl
e
tene
ss by usin
g (7
)
.
9
,...
1
;
3
,
2
,
1
,
8
.
0
*
2
6
.
1
k
i
ik
ik
ik
Therefore
1
0
3
1
i
ik
for all
rule
s th
at are a
s
so
ci
ated with ‘
C
o
s
t’. Usi
ng the
sub
rule
ba
se,
the assessm
ent result for ‘Acce
ssibility’ is obtained using ID
S
syst
em as Accessibility:
{(Excell
ent, 0.66), (G
ood, 0
.
23), (Avera
g
e
, 0.02), (Bad
, 0.00), (Un
k
n
o
wn, 0.09
)}
Whe
r
e, “Un
k
nown” in th
e
above resul
t
means th
at the output i
s
al
so in
com
p
lete input.
ER
approa
ch is u
s
ed to ag
gre
gate a
ll the packet ante
c
e
dents of the
L
rule
s to obtain the deg
re
e of
belief of
ea
ch refere
ntial
values of th
e
co
nsequ
ent
attribute
by takin
g
a
c
cou
n
t
of give
n i
n
put
values
P
i
of antece
dent attribute
s
. This
aggregatio
n c
an be ca
rri
ed
out either usi
ng re
cursive
or
analytical ap
proa
ch. In this re
sea
r
ch a
nalytical
app
roach [19] has bee
n con
s
i
dere
d
sin
c
e i
t
is
comp
utationa
lly efficient than re
cu
rsive
approa
ch [14
]
[20] [21], beca
u
se an
alytical ap
pro
a
ch
deal with all
para
m
eter
su
ch a
s
rul
e
weight, a
ttribut
e weig
ht, belief degre
e
, utility etc. For this
why there is no chan
ce
of absen
ce
of any para
m
eter. The
concl
u
si
on
O(Y)
,
con
s
i
s
t
i
n
g
of
referential va
lues of the
con
s
e
que
nt
attribute
,
i
s
gene
rated.
Equation
(8
) as given bel
ow
illustrate
s the
above phe
no
menon. :
0
(
Y
) =
S
(
P
i
) =
{(
C
j
,
j
),
j
=
1
, …,
N
}
(
8
)
W
h
er
e
j
de
notes the
bel
ief deg
ree
a
s
so
ciated
with
one
of th
e
consequ
ent re
feren
c
e
valu
es
su
ch as
C
j
. The
j
is cal
c
ulatin
g by a
nalytical fo
rm
at of the E
R
algo
rithm [3]
as ill
ust
r
ate
d
in
equatio
n (9
).
∏
1
∑
∏
1
∑
1
∏
1
With
1
1
1
(9)
Evaluation Warning : The document was created with Spire.PDF for Python.
IJEECS
ISSN:
2502-4
752
Belief-Rule-B
ased Intelligent Deci
sion
System
to Select Hospital Location
(Md. M
ahbu
bul Isla
m
)
611
The final
combi
ned
re
sult or output
g
enerated b
y
ER is rep
r
e
s
ente
d
by
{(
C
1
,
1
),
(
C
2
,
1
),(
C
3
,
1
),
……,(
C
N
,
N
)}
whe
r
e
j
is the
final belief degree atta
che
d
to the
j
th
referential val
ue
C
j
of the
con
s
e
que
nt attribute, obtai
ned after
co
mbining
all a
c
tivated rul
e
s in
the BRB by using ER.
2.3. Outpu
t
o
f
the
BRB Sy
stem
The output of
the BRB system is not cri
s
p/nume
r
ic
al value. Hen
c
e,
this output ca
n be conve
r
te
d
into cri
s
p/numeri
cal value by a
ssigni
ng utility score to each
refere
ntial value of the consequent
attribute [17].
∗
(10
)
Whe
r
e,
H
(
A*
)
is th
e
expect
ed
score
exp
r
esse
d a
s
nu
meri
cal val
u
e
and
u
(
C
j
)
is the utility score
of each refe
rential valu
e. For
exampl
e, in this
p
aper the ov
erall a
s
se
ssment re
sult
is
{(Excell
ent, 0.55), (Go
od, 0.25), (Avera
ge, 0.20)
, (B
ad, 0.00)} fo
r sele
cting ho
spital, then the
expecte
d utility sco
re is 0.
675 or 6
8
% which rep
r
e
s
en
ts good
risk for suita
b
le ho
spital lo
cation
.
In this p
ape
r the
RIMER met
h
odolo
g
y to ad
dre
ss va
rio
u
s type of unce
r
tainty su
ch a
s
incom
p
leten
e
ss, ign
o
ra
nce
and impreci
s
ene
ss by u
s
in
g equatio
n (7
) and eq
uatio
n (11
)
.
The i
n
comple
teness
as me
ntioned
o
c
curs d
u
e
to ig
no
ran
c
e, m
eani
ng that
belief
degree
has n
o
t been
assi
gne
d to any spe
c
ific
evaluati
on g
r
ade an
d this
can b
e
re
pre
s
ente
d
usi
ng
the
equatio
n as g
i
ven belo
w
.
N
n
n
H
1
1
(
1
1
)
Whe
r
e
H
is t
he belief deg
ree una
ssi
gne
d to any specific grad
e? If the value of
H
i
s
z
e
r
o
t
h
e
n
i
t
can a
r
g
ued th
at there is a
n
absen
ce of ig
nora
n
ce or in
compl
e
tene
ss. If the value of
H
is greater
than ze
ro th
en it can b
e
inferred th
at there
exi
s
ts ign
o
ran
c
e or in
com
p
letene
ss in t
h
e
as
se
ssm
ent
.
3. Rese
arch
Metho
d
This section
will
contai
n a br
i
e
f introduction to how the m
e
thod
implemented in thi
s
domain, toget
her with a
n
e
x
planation of
some of the i
m
pleme
n
tatio
n
details.
3.1. BRB IDS
Architectur
e
Archite
c
tural
desi
gn
rep
r
e
s
ents th
e st
ru
cture of
data a
nd p
r
og
ram
compon
ents th
at are
requi
re
d to build a
com
puter-ba
s
e
d
system. It
also
con
s
ide
r
s the pattern
of the syst
em
orga
nization, kno
w
n a
s
architectural style. BRB
IDS adopts the thre
e-laye
r archit
ecture [23] [24],
whi
c
h con
s
ist
of pre
s
entati
on layer, ap
p
lication laye
r and data
proce
s
sing lay
e
r a
s
sho
w
n
in
figure 1.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 25
02-4
752
IJEECS
Vol.
1, No. 3, March 20
16 : 607 – 618
612
Figure 1. BRBIDS Archite
c
ture
3.2. Sy
stem
Compon
ents
The inp
u
t cla
r
ification
of input ante
c
e
d
ent
W1
1 (Se
c
urity
ward a
r
oun
d),
W12
(Vand
al
Proof), W13
(Ope
n Lo
cati
on), W21 (Ex
pan
sion
Ca
p
a
city), W2
2 (Parki
ng Spa
c
e), W2
3 (Sto
rey
Numb
er
), W
31 (
N
e
u
tral l
o
catio
n
),
W3
2 (T
raffi
c A
c
ce
ss
),
W
33
(
P
ublic
Tra
n
sport Li
nk),
W41
(Co
n
st
ru
ction
Co
st), W4
2 (Land
Cost),
W51
(Land
Risk),
W52
(C
onstructio
n
Risk), W53
(Ti
m
e
Frame
an
d d
e
livery Spe
e
d
)
a
r
e t
r
an
sformed to
refe
re
ntial value
by
equ
ation
(4
), (5
) o
n
b
ehalf
of
expert. The in
put clarifi
c
atio
ns of this BRB system
tran
sform
ed to re
ferential is
sh
own in tabl
e 1.
Table 1. The
Input are T
r
a
n
sformed in
Referential V
a
lue
Sl.No.
Input Antecedent
Expert Belief
Referential Value
Excellent
Good
Average
Bad
0 W11
0.2
0.05
0.1
0.3
0.55
1 W12
1
1
0
0
0
2 W13
0.8
0.5
0.5
0
0
3 W21
0.5
0.1
0.8
0.1
0
4 W22
1
0.8
0.2
0
0
5 W23
0.9
0.86
0.14
0
0
6 W31
0.5
0.1
0.4
0.5
0
7 W32
1
0.8
0.2
0
0
8 W33
1
0.8
0.2
0
0
9 W41
0.4
0.1
0.5
0.4
0
10 W42
0.5
0.5
0.4
0.1
0
11 W51
1
0.8
0.2
0
0
12 W52
0.6
0.5
0.3
0.1
0.1
13 W53
0.7
0.65
0.2
0.1
0.05
3.3. Kno
w
l
e
d
g
e Bas
e
Co
n
s
truc
ted
Usi
ng BR
B
In
p
r
e
s
en
t pap
e
r
,
w
e
wo
rke
d
on
ass
e
s
s
me
n
t
pr
oc
es
s to
se
lec
t
th
e s
u
itab
le
lo
catio
n
fo
r
hospital e
s
ta
blishm
ent. In orde
r to
con
s
truct BR
B kn
owle
dge base
of
this syst
em
we de
sig
ned
a
BRB frame
w
ork to
site asse
ssm
ent a
c
cordin
g to do
main expe
rt. The BRB fra
m
ewo
r
k of su
itable
location assessment as
illustrated in Figure 2, from the fram
ework, it can be observed that input
factors that
determi
ne
su
itable lo
catio
n
for
ho
spital
. The B
R
B knowl
edge
ba
se
ha
s different
traditional
rul
e
to assessm
ent, which ne
ed to conve
r
t belief rule
s.
Evaluation Warning : The document was created with Spire.PDF for Python.
IJEECS
ISSN:
2502-4
752
Belief-Rule-B
ased Intelligent Deci
sion
System
to Select Hospital Location
(Md. M
ahbu
bul Isla
m
)
613
Figure 2. Hierarchical Rel
a
tionship
amo
n
g
locatio
n
evaluation Va
ria
b
le
In su
ch
situa
t
ions, b
e
lief rules may p
r
ovide an
alte
rnative
soluti
on to a
c
com
m
odate
different type
s an
d deg
ree
s
of un
ce
rtai
nty in
representing do
mai
n
kn
owle
dge.
A BRB can
be
establi
s
h
ed i
n
the followin
g
four way
s
[15]- (1
) Extra
c
ting beli
e
f ru
les from exp
e
r
t kno
w
led
ge
(2)
Extracting b
e
lief rule
s b
y
examining
histo
r
ical
d
a
ta; (3)
Usin
g the p
r
evio
us
rule
base
s
if
available, a
n
d
(4
) Rand
o
m
rule
s
witho
u
t any pre-kn
owle
dge. In t
h
is
p
ape
r, we
con
s
tructe
d i
n
itia
l
BRB by th
e
dom
ain
exp
e
rt
kno
w
led
g
e
. Thi
s
B
R
B co
nsi
s
ts of
four sub
-rul
e
-ba
s
e
s
nam
ely
environ
ment
and safety (W1
), size (W2), acce
ssib
ili
ty (W3), cost
effectiveness (W4),
risk (W5
)
and lo
cation
of healthca
re center
(S
). W4
(Co
s
t
Effectiveness)
sub
-
rule-b
ase h
a
s th
ree
antecede
nt attributes. Ea
ch ante
c
ede
nt attribute con
s
ist
s
of four
referential val
ues.
Hen
c
e, this
sub
-
rule-ba
s
e
con
s
ist
s
of
16 rul
e
s. Th
e entire B
R
B (whi
ch
co
nsi
s
ts of six
sub
-rul
e
ba
se
s)
con
s
i
s
ts
of (6
4+6
4
+64
+
16
+64
+
1
024
)
=
1296
beli
e
f ru
les. It i
s
a
s
su
med th
at all
b
e
lief rules h
a
v
e
equal rule weight; all antece
dent equ
al weigh
t, an
d the initial belief degree
assign
ed to each
possibl
e co
n
s
eq
uent by two expe
rt fro
m
accumul
a
ted the data.
To better h
a
n
d
le un
ce
rtainties,
each belief
ru
le co
nsid
ered
the three
ref
e
rential
val
u
e
s
are Excelle
nt (E),
Goo
d
(G), Average
(A)
and Bad
(
B).
Locat
i
on
of
He
alt
h
care
Ce
n
t
e
r
Environment
&
Safety
Security
wa
r
d
Around
Vandal
Proof
Op
en
Locat
i
on
Size
Ex
pans
ion
Cap
a
cit
y
Park
ing
Space
Storey
Numbers
Accessibility
Neutral
Locat
i
on
Traffic
Acce
ss
Public
Transport
Links
Co
st
Effec
t
iveness
C
o
ns
truc
tion
Co
st
Land
Co
st
Risk
Land
Risk
C
o
ns
truc
tion
Risk
Timeframe
&
Delivery
Speed
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 25
02-4
752
IJEECS
Vol.
1, No. 3, March 20
16 : 607 – 618
614
Table 2. Initial Belief Rules of
Sub-Rule
-Base (Cost
Effectiveness)
Rule
No.
Rule
Weight
IF
THEN
W41
W42
Cost Effectiveness(
W4)
Excellent
Good
Average
Bad
0
1
E
E
1
0
0
0
1
1
E
G
0.4
0.5
0.1
0
2
1
E
A
0.5
0
0.5
0
3
1
E
B
0.6
0.1
0.1
0.2
4
1
G
E
0
0.8
0.2
0
5
1
G
G
0
0.6
0.3
0
6
1
G
A
0.33
0.66
0
0
7
1
G
B
0
0.93
0.1
0
8
1
A
E
0
0.8
0.2
0
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
14
1
B
A
0.2
0
0.8
0
15
1
B
B
0
0.06
0.93
0
An example of a belief rule taken from T
able 2 is illust
rated bel
ow.
R1:
IF
W41 i
s
‘E’
AND
W42 is ‘E’
THEN
Cos
t
Effec
t
ivenes
s
(W2) is
{E
(1.0
0), G (0.00), A (0.00),
B (0.00)}
3.4. Inferenc
e Engine using ER
This B
R
B IDS desi
gne
d
usin
g the E
R
a
pproa
ch
[15] [18] [20
]
[21] [24] which
is
descri
bed i
n
se
ction 2.2. I
t
is simila
r to
tradi
tional fo
rwa
r
d
ch
ainin
g
. The infe
re
nce
with a B
R
B
usin
g the E
R
app
roa
c
h
al
so involve
s
a
ssi
gnin
g
valu
es to
attribut
es, evalu
a
tin
g
co
ndition
s
and
che
c
king to
see if all of the con
d
ition
s
i
n
a ru
le
are
satisfied. The
BRB infere
nce pro
c
e
s
s usi
ng
the ER a
p
p
r
o
a
ch
de
scribe
d by the follo
wing
step
s
are input tran
sformatio
n
, cal
c
ulatio
n of t
h
e
activation weight, calculati
ng com
b
ine
d
belief deg
ree
s
to all con
s
e
quent
s, belief
degre
e
upd
a
t
e
and ag
gre
gat
e multiple acti
vated belief rules.
The input
s of data are of two types, obje
c
tive
and su
bj
ective. Input tran
sform
a
tion
of this
system an
d input cla
r
ificati
on are
d
edu
ced in previo
u
s
se
ction an
d
table 1 by using (4
) (5). After
the value
assignment fo
r a
n
tece
dent,
ca
lculatin
g the
combi
ned
ma
tching
deg
re
e
s
b
e
twe
en th
e
inputs a
nd th
e rule’
s
ante
c
edent
s, the next step is
to cal
c
ulate a
c
ti
vation weig
ht for each pa
cket
antecede
nt i
n
the
rule
ba
se
usi
ng
(6
). The
be
lief
d
egre
e
s in th
e
po
ssi
ble
co
nse
que
nt of t
h
e
activated
rul
e
s in
the
rule
b
a
se
a
r
e
upd
ated u
s
in
g
(7).
Then
ag
gre
g
a
ting all
a
c
tivated
rule
s
usi
n
g
the ER ap
pro
a
ch
to
gen
erate a
combin
ed b
e
lief
deg
ree
in
po
ssi
bl
e con
s
eq
uent
s u
s
in
g
(8
)
(9
).
Then
expect
ed re
sult
of suitabl
e lo
cation a
s
sessment was
calcul
ated fro
m
its differe
nt
con
s
e
que
nts of factors.
Finally, pre
s
enting the
sy
stem inferen
c
e results of
suitable lo
cation
con
s
e
que
nt whi
c
h is n
o
t crisp/n
u
me
rica
l value, t
hen it is conve
r
ted
into cri
s
p/nu
meri
cal value
for
recomme
ndat
ion usi
ng (1
0).
3.5. BRB IDS
Interface
System interf
ace i
s
an int
e
rmediate posit
ion that
re
pre
s
ent
s the
intera
ction
b
e
twee
n
use
r
and
syst
em. Figure 3
rep
r
e
s
ent
s the
BRB syste
m
interface of this pape
r.
Evaluation Warning : The document was created with Spire.PDF for Python.
IJEECS
ISSN:
2502-4
752
Belief-Rule-B
ased Intelligent Deci
sion
System
to Select Hospital Location
(Md. M
ahbu
bul Isla
m
)
615
Figure 3. GUI
of the IDS
4. Results a
nd Discu
ssi
on
In the previo
us
se
ction, we have
disc
u
s
sed ab
ou
t the RIMER method a
n
d
how to
impleme
n
t it.
Therefore,
in
this
se
ction
we will
loo
k
at
the results f
r
om u
s
in
g thi
s
metho
d
o
n
t
h
e
different type
s of alternatives. Figu
re
4 sho
w
s the a
s
se
ssm
ent di
stribution
whi
c
h must b
e
do
ne
first by e
m
plo
y
ing the tran
sform
a
tion
eq
uation.
Any measurement
s
of quality can
b
e
tra
n
sl
a
t
ed
to the same
set of grad
e
s
as the top
attri
bute whi
c
h ma
ke it easy for furth
e
r analy
s
is.
The
asse
ssm
ents
given by the
De
cisi
on Ma
ker (DM
)
in
Fi
gure
2 are fe
d into IDS an
d the agg
re
g
a
ted
results a
r
e yielded at the
main criteria l
e
vel (Figu
r
e 2
)
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 25
02-4
752
IJEECS
Vol.
1, No. 3, March 20
16 : 607 – 618
616
A
t
t
r
ib
ute
s
Poduar
Bazar
Kandir
p
ar
Raceco
urse
Securit
y
w
a
rd
aroun
d
B(0.2)
A
(
0.8
)
G(0.4
)
E(0.
6)
G(0.4
)
E(0.
6)
Vandal Pro
o
f
G(0.4
)
E(0.
6)
B(0.2)
A
(
0.8
)
B(0.2)
A
(
0.8
)
Ope
n
L
o
cati
o
n
B(0.2)E
(
0.8)
A
(
1
.
0)
G(1.0
)
Expansi
on Cap
acit
y
E(1.0)
G(1.0
)
G(0.4
)
E(0.
6)
Parking Sp
ace
G(1.0
)
B(0.2)E
(
0.8)
E(1.0)
Store
y
N
u
m
b
er
s
B(0.2)
A
(
0.8
)
G(0.4
)
E(0.
6)
G(0.4
)
E(0.
6)
Neutral Loc
atio
n
G(0.4
)
E(0.
6)
B(0.2)
A
(
0.8
)
B(0.2)
A
(
0.8
)
Traffic
A
c
ces
s
B(0.2)E
(
0.8)
A
(
1
.
0)
G(1.0
)
Public Trans
por
t
Lin
k
G(1.0
)
B(0.2)E
(
0.8)
E(1.0)
Cons
truc
tio
n
C
o
st
B(0.2)
A
(
0.8
)
G(0.4
)
E(0.
6)
G(0.4
)
E(0.
6)
Land
Cos
t
G(0.4
)
E(0.
6)
B(0.2)
A
(
0.8
)
B(0.2)
A
(
0.8
)
Land
Risk
B(0.2)E
(
0.8)
A
(
1
.
0)
G(1.0
)
Time Frame and
deli
v
e
r
y
Sp
eed
B(0.2)E
(
0.8)
A
(
1
.
0)
G(1.0
)
Figure 4
.
Assessment Scores Of suita
b
le
location Ba
sed On Sub Criteria
(E-Excell
ent, G-G
ood, A-A
v
erage, B-Ba
d)
A
l
tern
a
t
iv
e
E
xcel
len
t
G
o
o
d
A
ver
ag
e
B
a
d
T
o
t
al
D
o
B
Podua
r Baza
r
0.80 0.10
0.10
0.00
1.00
Kandirpar
0.15 0.45
0.20
0.20
1.00
Ra
ce
cou
r
s
e
0.18 0.52
0.10
0.20
1.00
F
i
gure 5. T
he Overall Assess
ment (A
lternati
v
es) (DoB-De
g
r
ee of Beli
ef)
A
l
tern
a
t
iv
es
E
x
p
e
ct
e
d Utility
Sc
or
e
/
BRB
System
Result
Manuel
Result
Benchm
ark
Result
Stage
Podua
r Baza
r
92%
85%
90%
Excellent
Kandirpar
87%
77%
85%
Good
Ra
ce
cou
r
s
e
75%
80%
78%
Bad
Figure 6. Overall Asse
ssm
ent for suitabl
e locatio
n
s
The th
ree
al
ternatives (l
o
c
ation
)
simul
a
t
ed d
a
ta set with a
s
se
ssment o
u
tcome i
s
pre
s
ente
d
in
figure 6. This figure re
pre
s
ent
s
ove
r
all asse
ssm
ent outcom
e
from locati
on
informatio
n. The re
sult of this system
is
measure
d
in percent
age for re
co
mmend
ation.
The
output of this system was gene
rated b
a
se
d on
outp
u
t utility equation (10
)
. In this pape
r, the
utility sco
re
of (10
0
-9
0)
% assign
ed t
o
‘Exce
lle
nt’, (85
-
89
) % a
ssi
gne
d to ‘
G
ood’,
(80
-
8
4
) %
assign
ed to ‘Average’ a
n
d
(0-7
9) % assigned to ‘Bad’
.
In the case
study
, the lo
catio
n
a
s
sessm
e
n
t
of thre
e
alternative
s
usi
ng thi
s
syste
m
,
manual
syst
em and
ben
chma
rk re
sul
t
is sho
w
n i
n
figure
6. The hi
stori
c
a
l
results
were
con
s
id
ere
d
a
s
be
nchma
r
k.
From fig
u
re
4 it
can
be o
b
se
rved that
IDS gene
rate
d re
sult ha
s l
e
ss
deviation tha
n
from
be
nch
m
ark result.
Hen
c
e, it
ca
n
be
arg
ued
th
at IDS outp
u
t is m
o
re relia
ble
than
m
anual
system. The
r
efore,
it ca
n be con
c
lu
ded
that if the
a
s
se
ssm
ent
of
suitabl
e lo
cati
on
evaluation i
s
carried out by usi
ng the IDS, eventuall
y
this will pl
ay an important role in taki
ng
deci
s
io
n to avoid un
certai
nty issue.
Evaluation Warning : The document was created with Spire.PDF for Python.