TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol. 16, No. 3, Dece
mbe
r
2
015, pp. 583
~ 590
DOI: 10.115
9
1
/telkomni
ka.
v
16i3.937
0
583
Re
cei
v
ed Au
gust 24, 20
15
; Revi
sed
No
vem
ber 1
6
, 2015; Accepte
d
No
vem
ber
30, 2015
Combining Fuzzy Logic and Dempster-Shafer Theory
Andino Ma
s
e
leno
*
, Md. Mahmud Ha
san, Norjaidi
Tuah
ST
MIK Pringse
w
u, Pri
ngs
e
w
u,
Lampu
ng, Ind
ones
ia
F
a
cult
y
of Information T
e
chno
log
y
, Kazak
h
B
r
itish T
e
chni
cal Universit
y
, Kazakhstan
Comp
uter Scie
nce Progr
am, Univers
i
ti Brun
ei Dar
u
ssal
a
m, Negar
a Brun
ei
Darussa
lam
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: andi
nomas
el
eno
@mai
l.ru
A
b
st
r
a
ct
T
h
is res
earch
ai
ms to
co
mb
in
e the
math
e
m
a
t
ical th
eory
of
evid
ence
w
i
th t
he r
u
le
b
a
sed
l
ogics
to
refine th
e pr
ed
ictabl
e o
u
tput. Integratin
g F
u
zz
y
Log
ic
a
nd
De
mpster-S
haf
er theory
is ca
l
c
ulate
d
fro
m
th
e
similar
i
ty of Fu
zz
y
me
mb
ership fu
nction.
The nov
el
ty
aspect of this
w
o
rk is that basic pr
ob
ab
ilit
y
assig
n
m
ent is prop
osed b
a
se
d on the si
mil
a
rity me
as
ure
betw
een
me
mbersh
ip functio
n
. T
he simi
lari
t
y
betw
een F
u
zzy me
mbers
h
ip
function
is ca
lculate
d
to
g
e
t a
basic
prob
ab
il
ity assig
n
m
ent.
T
he De
mpste
r
-
Shafer
math
e
m
atic
al the
o
ry
of evide
n
ce h
a
s
attrac
ted co
nsid
erab
le atte
ntion as
a pro
m
is
ing
metho
d
of
dea
lin
g w
i
th s
o
me
of the
b
a
s
ic pr
obl
e
m
s
a
r
ising
i
n
c
o
mbi
natio
n
of evi
d
e
n
ce
an
d d
a
ta f
u
sio
n
. De
mpst
er-
Shafer th
eory
provi
des th
e a
b
ility to
de
al w
i
t
h ig
nora
n
ce
a
nd
miss
in
g i
n
fo
rmati
on. Th
e fo
und
atio
n of Fu
zzy
logic is
natura
l
lan
gua
ge w
h
ic
h can he
lp to
mak
e
full us
e o
f
expert infor
m
ation.
Ke
y
w
ords
:
fu
zz
y
lo
gic, De
mp
ster-Shafer the
o
ry, me
mb
er
sh
ip functio
n
, bas
ic prob
abi
lity a
ssign
ment
Copy
right
©
2015 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
Larg
e
a
m
ou
n
t
of literatu
r
e
is avail
able
o
n
Fu
zzy Lo
gic a
nd it
s a
ppl
ication
s
. Fu
zzy Lo
gic
can
handl
e p
r
oblem
s with i
m
pre
c
i
s
e dat
a and give
m
o
re a
c
cu
rate
results. Profe
s
sor L.A. Zad
eh
introdu
ce
d the con
c
ept
of Fuzzy Lo
gic [1].
Several
re
sea
r
chers have i
n
vestigate
d
th
e
r
e
lations
h
ip betw
een Fuzzy s
e
ts
and
Demps
t
er
-
S
haf
er
mathematic
al theor
y
of
evidenc
e
and
sug
g
e
s
ted dif
f
erent
ways
of integrating
them. In
tegration withi
n
symboli
c
, rul
e
-ba
s
e
d
mo
d
e
ls
have bee
n u
s
ed fo
r control and
cla
ssif
i
cation p
u
rpo
s
e
s
[2, 3]. Yager
and Fil
e
v attempted
to
pre
s
ent
a Fu
zzy infe
re
nce
system
ba
sed on
Fu
zzy
Dem
p
ste
r-S
hafer m
a
the
m
atical th
eory of
eviden
ce whi
c
h integ
r
ated
the proba
bil
i
stic info
rmati
on in the ou
tput [4].
In their works, the
con
s
e
que
nt is sh
ape
d as
a Demp
ste
r
-Shafer beli
e
f stru
cture, wh
ere ea
ch fo
ca
l element ha
s the
same
me
mbe
r
shi
p
fu
nctio
n
.
Binaghi
et al
. [3] pro
p
o
s
e
d
a
structu
r
e
for
cla
ssifi
cati
on ta
sks simil
a
r
to that of [4], wh
ere
the f
o
cal
elem
ent
is a
set re
prese
n
ting the
cla
s
s label.
Dymova et
al
.[5
]
prop
osed a critical an
alysi
s
of conventi
onal ope
ra
tio
n
s on intuitio
nistic Fu
zzy values an
d their
appli
c
ability to the sol
u
tio
n
of multiple
criteri
a
de
ci
sion m
a
ki
ng
probl
em
s in the intuitionistic
Fuzzy setting
. Ghasemi, et al., [6] studie
d
the ma
in
ch
ara
c
teri
stic
of the pro
p
o
s
ed
method
whe
r
e
is that
the
rul
e
s
of Fu
zzy i
n
feren
c
e
syst
em a
r
e
co
nsi
dere
d
as evid
ences in
whi
c
h the
firing
le
vel
of each
rule
and Fu
zzy Naive Bayes
method a
r
e
e
m
ployed for
cal
c
ulatin
g the basi
c
p
r
ob
ability
assignm
ent o
f
focal el
eme
n
t. In the Na
ive Baye
s
cl
assifier, all v
a
riabl
es are
assume
d to
be
nominal
vari
able
s
, which
mean
that
each vari
abl
e ha
s
a finit
e
num
be
r of
value
s
a
n
d
also
assume
s ind
epen
den
ce o
f
features. Howeve
r, in
la
rge d
a
taba
se
s, the variab
les often take
contin
uou
s value
s
or have
a large nu
mb
er of nume
r
ical values.
Walije
wski, et
al., [7] concentrated
on t
he ro
l
e
of Fu
zzy o
perator
s, and on the
probl
em
of discretizati
on of
co
ntin
uou
s attrib
utes.
Dutta
et
al. [9] studi
e
d
Demp
ster-Shafer th
eory of
eviden
ce by
con
s
id
erin
g f
o
cal
elem
ent
s a
s
tria
ngul
a
r
Fu
zzy num
ber. T
he a
u
th
ors have
devi
s
ed
a method fo
r obtaini
ng b
e
lief and pl
a
u
sibility me
a
s
ure from
b
a
si
c proba
bili
ty assig
n
me
nts
assign
ed to
Fuzzy focal
e
l
ements. Bo
u
d
raa, et
al., [8] estimated
basi
c
p
r
ob
abi
lity assig
n
me
nts
usin
g Fu
zzy
membe
r
ship
function
s. Binaghi, et
al.,
[3] presente
d
a supe
rvised cl
assification
model
integ
r
ating F
u
zzy reasonin
g
a
n
d
Demp
st
er-Shafer propa
gation of
evi
den
ce ha
s
b
een
built on top
o
f
conn
ectio
n
i
s
t tech
niqu
es to addr
ess cl
assificatio
n
ta
sk
s
in whi
c
h vaguen
ess
a
n
d
ambiguity co
exist. The ap
proa
ch i
s
the
integrat
ion
within a Ne
uro-Fu
zzy syst
em of kno
w
le
dge
stru
ctures a
n
d
inferen
c
e
s
for evidential re
a
s
oni
ng
base
d
on
Demp
ste
r-Sh
a
fer theo
ry. The
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 16, No. 3, Dece
mb
er 201
5 : 583 – 590
584
comm
on wea
k
ne
ss of neu
ral network, however, is
a probl
em of
determin
a
tio
n
of the optimal
size of a network configu
r
ation, as thi
s
ha
s
a sign
ificant impa
ct on
the effectivene
ss of
its
perfo
rman
ce.
The
Demp
ster-Shafe
r
th
eory o
r
igi
nat
ed from the
con
c
e
p
t of l
o
we
r a
nd
up
per
prob
ability
in
duced by
a multivalued mappin
g
by De
mp
ste
r
[1
0, 11]. Follo
wing
this
wo
rk
hi
s
stude
nt Glen
n Shafer [12] furt
her extended the the
o
ry in his bo
ok ”A Mathe
m
atical The
o
r
y of
Evidence
”
, a more tho
r
ou
g
h
explanation
of belief
function
s. Lyu et al.
[13] studied Dem
p
ste
r
-
Shafer theo
ry
for the rea
s
o
n
ing with im
p
r
eci
s
e
co
ntext. Yao, et al., [
14] used
Dempste
r-S
haf
er
theory for the
multi-attribut
e deci
s
ion m
a
kin
g
pro
b
le
ms with in
co
mplete inform
ation by identify all
possibl
e foca
l elements from the inco
mplete
de
cisi
on matrix, and then calculate the ba
si
c
prob
ability assignm
ent of e
a
ch fo
cal
ele
m
ent and th
e
belief fun
c
tio
n
of ea
ch d
e
cision
altern
ative.
Yu, et al., [15] use
d
Dem
p
ster-Shafe
r
Theo
ry
as
an
applie
d app
roach to sce
n
a
rio fo
re
ca
sting
based o
n
im
pre
c
ise p
r
ob
ability. Uphof
f, et al., [
16] studie
d
a
p
p
lication
of Dempste
r-S
haf
er
theory to
task m
appin
g
u
nder epi
stemi
c
u
n
certai
nty.
De
mpste
r-S
hafer mathe
m
atical th
eo
ry of
eviden
ce im
p
lies
a type
of
uncertainty
a
s
soci
at
ed
wit
h
conditio
n
s
of ambi
guity throu
gh th
e
d
a
ta
by deali
ng
with ig
no
ran
c
e
and
mi
ssing info
rmati
on. Thi
s
ch
a
r
acte
ri
stic i
s
due
to
usi
ng
a
combi
nation
of evidence weight from different
sou
r
ces to obtain a new evide
n
ce weig
ht.
2. Dempster-Shafer Ma
th
ematical The
or
y
of Ev
ide
n
ce
It is difficult to avoid u
n
ce
rtainty whe
n
attempting to
make mod
e
l
s
of the
real
worl
d.
Un
certai
nty is inhe
rent to natural p
h
enome
na,
a
nd it is impossible to create a perf
e
ct
rep
r
e
s
entatio
n of reality.
Cla
ssi
c math
ematics deal
s with ideal
worl
ds
whe
r
e
perfect ge
o
m
etric
figures exist
and can verif
y
extr
aordinary conditions. The formali
s
ation of Fuzzy sets started in
the 19
60
s
wit
h
the
works o
f
Zade
h [1] i
n
Fu
zzy
set
s
a
nd
Dem
p
ste
r
[11] in
belief f
unctio
n
s. B
e
li
e
f
function
s
offer
a n
on B
a
yesia
n
met
hod fo
r
qua
ntifying su
bj
ective eval
u
a
tions by u
s
ing
probability. In the 1970s, it
wa
s further
developed by
Shafer,
wh
ose book Mathematical
Theory
of Evidence [
12] re
main
s
a cl
assic in
b
e
lief functi
o
n
s
, or the
so
-called T
heo
ry of
Evidence. This
theory h
a
s b
een
also
call
ed the
Dem
p
ster-Shafe
r
Mathemati
c
al
The
o
ry of E
v
idence. In t
he
1980
s, the scientific
com
m
unity worki
ng with Artifi
cial Intellige
n
c
e got involv
ed in u
s
ing t
h
e
theory of evidence in appli
c
ation
s
. The
Demp
ste
r-S
h
a
fer theory o
r
the theory of belief functio
n
s
is a mathem
a
t
ical theory of
evidence whi
c
h can
be int
e
rp
reted a
s
a
generalizatio
n of prob
ability
theory in wh
ich the elem
ents of the sampl
e
sp
ace to which non
zero pro
bability mass is
attributed a
r
e not singl
e
points but
sets. Th
e
se
ts that get nonzero m
a
ss are call
ed focal
element
s. The sum of these probability masses i
s
one, however, the ba
si
c difference bet
ween
Demp
ste
r-Sh
a
fer math
em
atical the
o
ry
of eviden
ce
and tra
d
ition
a
l pro
bability
theory is th
at the
focal el
eme
n
ts of a
Demp
ster-Shafe
r
structur
e
may o
v
erlap one a
nother. The Demp
ste
r-Sh
a
fer
mathemati
c
al
theory of eviden
ce also p
r
ovide
s
meth
ods to re
pre
s
ent and co
m
b
ine wei
ghts
o
f
eviden
ce.
2.1. Repre
s
e
n
ta
tion of Ev
idence
The
Dem
p
ste
r-Shafe
r
th
eo
ry as
su
me
s t
hat there i
s
a
fixed
set
of m
u
tually excl
usive and
exhau
stive elements
call
e
d
hypot
he
se
s or propo
sitio
n
s a
nd sym
b
olize
d
by the
Gree
k letter
ϴ
,
r
e
pr
es
e
n
t
ed
a
s
ϴ
=
{h1
,
h2
, …,
hn
}
, wh
ere hi is
calle
d a hypothe
si
s or p
r
op
ositi
on. A hypothesi
s
can
be any
subset of the
frame, in
exa
m
ple, to sin
g
l
e
tons i
n
the f
r
ame
or to
combinatio
ns
o
f
element
s in t
he fram
e.
ϴ
i
s
al
so
calle
d
frame
of discernme
n
t. A basi
c
p
r
o
babi
lity assig
n
me
nt
(bpa
) is rep
r
e
s
ente
d
by a mass fun
c
tion
m
: 2
ϴ
→
[0,
1]. Where 2
ϴ
is the power
set of
ϴ
.
The b
a
si
c
probability a
s
signment i
s
a
primitiv
e of
e
v
idence the
o
ry. Gene
rally
spe
a
ki
ng,
the term
basi
c probability
assignm
ent
does not
refer to probabilit
y in the cl
assical
sense. T
h
e
bpa,
rep
r
e
s
e
n
ted by
m, d
e
fines a
map
p
ing
of the
p
o
we
r
set to
t
he inte
rval
b
e
twee
n 0
an
d 1,
whe
r
e
the
bp
a of th
e n
u
ll
set is 0
and
th
e bp
as of
all t
he
sub
s
et
s of
the
po
wer se
t is
1. In F
u
zzy
Logi
c, two va
lued lo
gic often
con
s
ide
r
s 0 to b
e
false
and
1 to b
e
true. Fu
zzy L
ogic
deal
s
wi
th
truth values
betwe
en 0 a
nd 1, and these value
s
are con
s
ide
r
e
d
as the intensity or degre
e
s of
truth.
The value of
the bpa for
a given set A (rep
r
e
s
ent
ed as
m(
A)
;A
Є
2
ϴ
), expresses th
e
prop
ortio
n
of all relevant a
nd available
eviden
ce
that
supp
orts the
claim that a p
a
rticul
ar el
em
ent
of
ϴ
(the
univ
e
rsal
set) bel
ong
s to th
e
set A but to
n
o
pa
rticul
ar subset of A. T
he valu
e of
m(A)
pertain
s o
n
ly to the set A
and ma
ke
s
n
o
additio
nal
claims a
bout a
n
y sub
s
et
s of
A. Any further
eviden
ce on the sub
s
ets
of A would b
e
represente
d
by another
bpa, in exa
m
ple
B
⊂
A
, m(B)
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Com
b
ining F
u
zzy L
ogi
c an
d Dem
p
ste
r
-Shafer The
o
ry (Andi
no Ma
sele
no)
585
woul
d the b
p
a
for the
su
b
s
et B. Form
a
lly, th
is description of m
can be
rep
r
e
s
ented
with the
followin
g
two
equatio
ns 1 a
nd 2:
(
1
)
(
2
)
From th
e p
r
i
m
itive of evid
ence theo
ry
or m
a
ss fun
c
t
i
on, the u
ppe
r an
d lo
we
r b
ound
s of
an interval
can be defined.
This in
terval contai
ns the
precise proba
bility of a set of interest and
is
b
oun
ded
by
two non
additive conti
nuou
s mea
s
ure
s
calle
d Belief
fun
c
tio
n
an
d
Pla
u
si
bility
function. Eviden
ce theo
ry
use
s
two
m
easure
s
of u
n
ce
rtainty, belief function
and pla
u
si
bi
lity
function, expressed
a
s
Bel
() an
d
Pls
() respectively. Gi
ven a
basic probability assi
gnment
m
, the
corre
s
p
ondin
g
belief functi
on mea
s
u
r
e
and pla
u
sibili
ty function measure a
r
e d
e
termin
ed for all
set
s
A
Є
2
ϴ
an
d
B
Є
2
ϴ
by equation
s
3 an
d 4:
(
3
)
(
4
)
The suppo
rt function o
r
b
e
lief,
Bel
, is the total belie
f of a set and all its sub
s
ets. The
lower bound
Belief for
a
set A is defined as the
sum
of all the basic probability assignment
s of
the pro
per
su
bset
s (B) of t
he set of inte
rest (A)
(
A
⊂
B
).
The plau
sibili
ty function of a pro
p
o
s
ition,
Pls, is the su
m of the masse
s
of all pro
positio
ns
in
which it is
wh
ol
ly or partially
contai
ned. Th
e
plau
sibility function is d
e
fined as the d
egre
e
to
which the evidence fails to refute A. These two
function
s, wh
ich h
a
ve be
e
n
som
e
time
s referred to
as lo
we
r an
d
uppe
r p
r
ob
a
b
ility functions
,
have the follo
wing p
r
op
erti
es are given
by equation
s
5 and 6:
(
5
)
(
6
)
Whe
r
e
Ā
i
s
th
e compl
e
men
t
ary hypothesis of
A, A
∪
Ā
=
ϴ
and
A
∩
Ā
=
∅
. The plausibility Pls (A)
is defin
ed
as the de
gree
to whi
c
h th
e
eviden
ce fa
ils
to
refute
A. This
term is
given by
the
equatio
n 7:
(
7
)
Due to
a la
ck of info
rmati
on, it is m
o
re
rea
s
o
nable
to present bo
und
s for th
e
result of
uncertainty q
uantificatio
n, as op
po
sed t
o
a singl
e value of pro
babi
lity. The total
degree of bel
ief
in a
given p
r
opo
sition A i
s
exp
r
e
s
sed
within
an inte
rval [
Bel
(A);
Pls
(A)], which lies in the
unit
interval [0,1].
Deali
ng
with
uncertainty i
s
a fu
ndam
ent
al issu
e in
the
study
of m
a
n
-
mad
e
com
p
u
t
ational
device
s
and
system
s whi
c
h can be ma
de to act in a manner
whi
c
h human wou
l
d be incline
d
to
call intellige
n
t. Dempste
r
-Shafer math
ematical th
e
o
ry of evidence is a
n
importa
nt tool of
uncertainty modellin
g wh
en both un
certainty orig
i
n
s from h
u
m
an’s la
ck of kno
w
le
dge of
the
physi
cal
worl
d and un
cert
ainty derives from the natural va
riabilit
y of the physical
worl
d are
pre
s
ent in the
proble
m
und
er co
nsi
d
e
r
ati
on.
In de
cisi
on m
a
kin
g
p
r
o
c
e
s
se
s
with h
u
m
an’s la
ck of
knowl
edge
of t
he p
h
ysi
c
al
world
and
lack of the
ability of me
asu
r
ing
and
modellin
g t
he phy
sical
worl
d, the F
u
zzy Logi
c
and
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 16, No. 3, Dece
mb
er 201
5 : 583 – 590
586
Demp
ste
r-Sh
a
fer math
em
atical theo
ry of evidenc
e h
a
ve gaine
d p
r
omin
en
ce a
s
the method
s of
choi
ce over traditional probabili
stic
m
e
thod
s.
The
fundame
n
tal
and imp
o
rta
n
t
object of t
he
mathemati
c
al
theory of evi
den
ce i
s
the
primit
ive fun
c
tion call
ed a
basi
c
p
r
ob
abi
lity assig
n
me
nt.
In
the ab
sen
c
e of
em
piri
cal data, experts
in
relat
ed field
s
pro
v
ide ne
ce
ssa
r
y informatio
n.
Ho
wever ho
w to o
b
tain
basi
c
p
r
o
babi
lity assig
n
me
nt is
still an
open i
s
sue.
The me
mbe
r
ship
function of a
Fuzzy set is
a gene
rali
zati
on of the indi
cator fu
nctio
n
in cla
ssi
cal
sets. In Fu
zzy
Logi
c, it represe
n
ts the d
egre
e
of truth as an
exte
nsio
n of valuation. Fuzzy Logi
c is a lo
gic
operation met
hod ba
se
d on
many-value
d
logic rath
er t
han bin
a
ry lo
gic or t
w
o-val
ued logi
c.
Dempster-Shafer m
a
them
atical
theory
of evidence,
a proba
bilisti
c reasoni
ng t
e
chnique,
is d
e
si
gne
d t
o
de
al
with
uncertainty a
nd in
com
p
let
ene
ss of ava
ilable info
rma
t
ion. Dem
p
st
er-
Shafer mathe
m
atical theo
ry of evidence
allows
one t
o
combi
ne ev
iden
ce from
different source
s
and arrive at a degree of b
e
lief whi
c
h is
rep
r
e
s
ente
d
by a belief function that ta
ke
s into acco
unt
all the availa
ble evide
n
ce. The d
egree
of belief is
expectin
g
a truth value
whi
c
h is the
rel
a
tion
betwe
en Fu
zzy Logi
c and
Demp
ste
r-Sh
a
fer ma
them
atical theo
ry of evidence.
2.2. Ev
idence Combina
t
ion
Demp
ste
r-Sh
a
fer th
eory
provide
s
a
method
to
combine
the
previou
s
me
asu
r
e
s
of
eviden
ce of
different
sou
r
ce
s. Thi
s
rul
e
a
s
sume
s t
hat the
s
e
so
urces a
r
e in
depe
ndent. T
h
e
combi
nation:
m =
m
1
⊕
m2
, also
called o
r
thogon
al sum
,
is defined a
c
cordi
ng to the Demp
ste
r
’s
rule
of com
b
i
nation [12]. It can
be a
pplie
d re
petit
ively whe
n
the
sou
r
ce
s
are
more than two. After
the combi
nat
ion, a deci
s
i
on can b
e
made amo
ng the different hypothe
se
s according to the
deci
s
io
n rule
cho
s
e
n
.
To use Dem
p
ster-Shafe
r
mathemati
c
al
theory
of evi
den
ce, there
must be th
e feasi
b
le
measures to
determin
e
b
a
si
c proba
bility assi
gn
men
t. The Fuzzy theory al
so
requi
re
s ba
si
c
probability assignm
ent. Basic probability assi
gnment
whi
c
h is
called the primitive function i
s
the
fundame
n
tal
and imp
o
rtan
t object of the mathemat
i
c
al theo
ry of evidence. The memb
ership
function of a
Fuzzy set is
a gene
rali
zati
on of the indi
cator fu
nctio
n
in cla
ssi
cal
sets. In Fu
zzy
Logi
c, it represe
n
ts the d
egre
e
of truth as an
exte
nsio
n of valuation. Fuzzy Logi
c is a lo
gic
operation met
hod ba
sed o
n
many-valued
logic rathe
r
than bina
ry logic or two
-
val
ued logi
c. Two-
valued l
ogi
c
often con
s
ide
r
s 0 to
be
fal
s
e
and
1 to
b
e
tru
e
. Fu
zzy
Logi
c
deal
s
with truth val
ues
betwe
en
0 a
nd 1,
and t
hese value
s
are
con
s
ide
r
ed
as the i
n
tensity o
r
d
egre
e
s of truth.
Demp
ste
r-Sh
a
fer math
em
atical the
o
ry
of eviden
ce
, a pro
babili
stic
rea
s
oni
n
g
tech
niqu
e, is
desi
gne
d to deal with un
ce
rtainty and in
compl
e
tene
ss of available
informatio
n. Demp
ste
r-Sh
a
fer
mathemati
c
al
theory of ev
iden
ce allo
ws on
e to
co
mbine evid
en
ce fro
m
different so
urce
s
and
arrive
at a d
e
g
ree
of beli
e
f whi
c
h i
s
rep
r
ese
n
ted
by
a
belief fun
c
tio
n
that takes i
n
to acco
unt
all
the available
evidence. The De
gre
e
of belief is
exp
e
cting a truth
value whi
c
h
is the relati
on
betwe
en Fu
zzy Logi
c and
Demp
ste
r-Sh
a
fer ma
them
atical theo
ry of evidence.
3. Fuzzy
Logic
The
origi
nal
motivation fo
r Fu
zzy L
ogi
c i
s
to
provide the
ba
si
s for
re
aso
n
in
g un
der
nonbi
nary i
n
formatio
n. Th
e en
sui
ng
re
aso
n
ing
sy
stem
often
this is
refe
rre
d t
o
a
s
a
pproximate
rea
s
oni
ng o
r
Fuzzy rea
s
o
n
ing. Ho
wev
e
r, this
sh
o
u
l
d
not be ta
ken to imply that the re
sult
ing
system i
s
a
n
y
less
exact
than that affo
rded
by
crisp
logic. Ind
e
e
d
, Fuzzy re
a
s
oni
ng mig
h
t be
con
s
id
ere
d
m
o
re
exact p
r
e
c
isely be
cau
s
e it doe
s n
o
t assume
a bi
n
a
ry unive
rse. The b
a
si
s fo
r
formal
re
aso
n
ing i
s
an i
n
feren
c
e
p
r
o
c
e
dure,
it
self ba
sed
up
on an approp
riate model
for
’if
-then
’
rule
s, or mo
dus po
nen
s. The gen
eral
goal is to
infer the deg
re
e of truth asso
ciated with
a
prop
ositio
n,
B, from the i
m
plicatio
n, A, or
A
→
B
. Consi
der, ’A’
d
enote
s
”sh
a
rp co
rn
er” an
d
’B’
”app
ro
ach slo
w
ly” than the
implic
atio
n ca
n naturally ex
pre
ss by:
premi
s
e 1
(f
act):
x
is A;
premi
s
e
2
(fact): IF
x
is
A T
H
EN
y
is B; co
nse
quen
ce
(
c
on
c
l
u
s
io
n)
: y is
B.
Or
premi
s
e 1
(f
act):
x
is A’;
premi
s
e
2 (fact): IF
x
is A THEN
y
is B; co
nse
quen
ce
(
c
on
c
l
u
s
io
n)
:
y
is
B’.
Fuzzy rea
s
o
n
i
ng Let A and
A’ be Fuzzy
sets o
n
the u
n
iverse X, and B a Fuzzy set on
Y. Implication
,
A ! B, is defined in term
s
of
a Fuzzy rel
a
tion R on th
e Carte
s
ia
n p
r
odu
ct, X
Y.
The Fu
zzy p
r
opo
sition B induced by the premi
s
e “x
is
A’”and the Fu
zzy rule “if x is A then y is
B”is defin
ed i
n
the form of the Fuzzy co
mpositio
n.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Com
b
ining F
u
zzy L
ogi
c an
d Dem
p
ste
r
-Shafer The
o
ry (Andi
no Ma
sele
no)
587
Assu
me that
the spe
c
ific
ca
se of com
positio
n ba
se
d on the Ma
x-Min ope
rat
o
r, then
thesp
e
ci
al ca
se of th
e ab
o
v
e gene
ral m
odel of F
u
zzy
rea
s
o
n
ing
ca
n be d
e
fined
usin
g eq
uatio
n 8
as:
(
8
)
By now, i
n
a
po
sition to
build
som
e
reasonin
g
e
n
g
ine
s
. Con
s
i
der the foll
o
w
ing
three
spe
c
ial ca
se
s
:
1)
Single Rul
e
with Single Pre
m
ise
The premise simplifie
s to the sp
eci
a
l ca
se of a scala
r
thresh
old
s
, or
2)
Single Rul
e
with Multiple Premise
s
The p
r
emi
s
e
simplifie
s to the spe
c
ial
ca
se of the
mini
mum of two
scala
r
threshol
ds.
Thus, the
co
mpositio
nal rule implie
s
.
3)
Multiple Rul
e
s with Multipl
e
Premises
The ’max’ op
erato
r
of equ
ation (8
) no
w applie
s, thus the are
a
of the impli
c
atio
n is
the maximum
of each mini
mally thresho
l
d premi
s
e.
At this poi
nt, the ba
sis for
Fuzzy rea
s
on
ing
with
the
remainin
g p
r
o
b
lem of
esta
blishi
ng
what the Fu
zzy con
s
eq
ue
nt actually mean
s in pra
c
tice. Tsu
k
a
m
ot
o Fuzzy rea
s
oning a
r
e mo
dels
based
on F
u
zzy Lo
gic. T
h
e
s
e
rule
s
are e
a
sy to l
earn a
nd u
s
e
an
d
can b
e
mo
difie
d
a
c
cordi
ng t
o
the situatio
n. It helps to m
a
ke d
e
ci
sio
n
s
and
c
an
be u
s
ed i
n
de
ci
sio
n
analy
s
is. T
s
ukam
oto Fu
zzy
rea
s
oni
ng
do
es
map
p
ing
from
given i
n
put to
an
out
put u
s
ing
Fu
zzy
Logi
c. T
s
ukam
oto F
u
zzy
rea
s
oni
ng ha
s a numb
e
r of
rules b
a
sed
on
if - th
en
co
ndition
s.
In this metho
d
, the con
s
eq
uen
ce of ea
ch Fuzz
y rule i
s
re
pre
s
e
n
te
d by a Fuzzy set with
a mon
o
toni
c
membe
r
ship f
unctio
n
. The
rule
ba
se
ha
s the form a
s
:
R
i
: if u is
A
i
a
nd v i
s
B
i
, then
w is
C
i
, i
= 1,
2, , n. Whe
r
e
μ
C
i
(w) i
s
a
monoto
n
ic f
unctio
n
. As a
re
sult, the in
ferre
d outp
u
t of
each rule i
s
d
e
fined a
s
a crisp value ind
u
ce
d by
the rules mat
c
hin
g
degree (firi
ng stre
ngth
)
. The
overall outp
u
t
is take
n as t
he wei
ghted
averag
e of e
a
ch
rule
s out
put. Suppo
se
, that the set
C
i
has a mo
noto
n
ic mem
bership functio
n
μ
C
i
(w) an
d tha
t
α
i
is the matchin
g
deg
ree
of its rule.
For the Fu
zzy set input (A’, B’) is given b
y
the equatio
n 9:
(
9
)
Then the result of its rule is obtain
ed by
the equation
10:
(
1
0
)
The final resu
lt is derive
d
from the weigh
t
ed
avera
ge li
ke in the foll
o
w
ing
whe
n
th
ere a
r
e
two rule
s. Thi
s
term is give
n by the equa
tion 11:
(
1
1
)
Since ea
ch
ru
le infers a crisp result, the
Tsu
k
am
oto F
u
zzy model a
ggre
gate
s
ea
ch rule
s
output by the weighte
d
averag
e metho
d
. Therefo
r
e,
it avoids the time-co
n
su
ming process of
defuzz
i
fication.
4. Fuzzy
Logic and Demp
ster
-Sha
fer
Mathem
atica
l
Theor
y
of Ev
idence
Fuzzy set the
o
ry propo
sed
by Zadeh in
1965 i
s
a
kin
d
of theoretical rea
s
o
n
ing
schem
e
for dealin
g with imperfe
ct data. A Fuzzy set, as th
e name impli
e
s, is a
set without a cri
s
p
boun
dary. T
hat is, th
e transitio
n i
s
g
r
adu
al a
nd
t
h
is smo
o
th
t
r
an
sition
i
s
chara
c
te
rized by
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 16, No. 3, Dece
mb
er 201
5 : 583 – 590
588
membe
r
ship
function
s. Th
e Fuzzy inference sy
ste
m
or Fuzzy m
odel is a p
o
pular
com
put
in
g
frame
w
ork b
a
s
ed
on the
co
nce
p
ts of Fu
zzy set th
e
o
ry, Fuzzy if-then
rule
s an
d Fu
zzy
rea
s
oni
n
g
.
The ba
si
c structure of a F
u
zzy in
fere
nce system
co
n
s
ist
s
of th
re
e con
c
e
p
tual
compon
ents which
are
rul
e
base
,
databa
se
a
n
d
rea
s
oni
ng
mech
ani
sm.
A rule
b
a
se o
r
d
e
ci
sio
n
m
a
trix of the
Fu
zzy
kno
w
le
dge
-b
ase,
whi
c
h
contain
s
a
sel
e
ction
of
F
u
zzy rule
s a
nd
a data
b
a
s
e,
whi
c
h
define
s
th
e
membe
r
ship f
unctio
n
s u
s
e
d
in the Fu
zzy rules
com
p
ose
d
of expe
rt IF
< ant
ece
dent
s >
TH
EN
<
concl
u
sions >
rule
s. A reasoning me
ch
a
n
ism, whi
c
h p
e
rform
s
the F
u
zzy rea
s
oni
ng ba
sed on
the
rule
s and giv
en facts to d
e
rive a rea
s
onabl
e outpu
t or con
c
lu
si
on. Mathema
t
ically spea
ki
ng,
c
o
ns
ider
F to r
e
pr
esent a
Fuzz
y s
e
t in
the
domain of dis
c
ours
e
U
.
The Fuzzy s
e
t F c
an
be
defined by th
e membe
r
shi
p
function fro
m
equation 1
2
as follo
ws:
(
1
2
)
The m
e
mbe
r
ship
fun
c
tion
of a Fu
zzy set is
a ge
ne
ralizatio
n of th
e indi
cato
r fu
nction
in
cla
ssi
cal sets. In Fuzzy Logic, it repre
s
ent
s the
de
gree of truth
as an exten
s
ion of valuation.
Prope
rties of
membe
r
ship functio
n
are:
1)
The memb
ership fun
c
tion
should b
e
stri
ctly monotonically incr
easi
ng, or st
rictly
monotoni
cally
de
cre
a
si
ng,
or
stri
ctly m
onoto
n
ica
lly increa
si
n
g
then
stri
ctly
monotoni
cally
de
crea
sing
with the
in
cre
a
sin
g
valu
e o
f
eleme
n
ts i
n
the
universe
of
discou
rse X.
2)
The memb
ership fun
c
tion
sho
u
ld be
co
ntinuou
s or pi
ece
w
i
s
e conti
nuou
s.
3)
The memb
ership fun
c
tion
sho
u
ld be diff
erentia
ble to provide
smo
o
t
h result
s.
4)
The me
mbe
r
ship fu
nctio
n
sho
u
ld b
e
of simpl
e
straight segme
n
ts to ma
ke
the
pro
c
e
ss of fu
zzy mo
dels e
a
sy and to hi
gh accu
ra
cy.
A new
method to obtain basi
c
probabilit
y assi
gnment is
proposed based on the
simila
rity measu
r
e b
e
twe
en mem
bership fun
c
tion.
Method to
integrat
e F
u
zzy Logi
c
and
Demp
ste
r-Sh
a
fer mathem
atical t
heo
ry of evidence a
s
follows:
1)
Define ling
u
istic variable a
nd Fuzzy ra
n
ge. Define a
variable
who
s
e value
s
ca
n be
expre
s
sed by
mean
s of na
tural lang
uag
e te
rms.
Whe
n
defining a li
ngui
stic varia
b
le,
it is also to sp
ecify minimu
m and maxim
u
m values.
2)
Define the
F
u
zzy rule
s. T
he Fu
zzy rul
e
s
a
r
e n
early
a se
rie
s
of if-then
statem
ents.
These statem
ents are de
ri
ved by an exp
e
rt to achieve
optimum re
sults.
3)
Define the f
o
rmul
a. Vari
ous oth
e
r
membe
r
ship
function
s such a
s
trian
gular,
trape
zoid
al,
Gau
ssi
an, a
n
d
si
gmoid
a
l
can b
e
u
s
ed
in
the formulati
on of m
e
mbe
r
shi
p
function
s.
4)
Define the in
put. The infe
ren
c
e
schem
es a
r
e ba
se
d
on the com
positio
nal rul
e
of
inference, and the result
is derived from
a set of Fuzzy rules and gi
ven inputs.
5)
Cal
c
ulate m
e
mbershi
p
val
ue. Cal
c
ul
ate
informatio
n
contai
ned i
n
a fuzzy set which
is de
scribe
d by its membe
r
shi
p
value.
6)
Cal
c
ulate th
e
rule.
Cal
c
ula
t
e the si
milari
ty between
F
u
zzy mem
bership fu
nctio
n
to
get a basi
c
probability assi
gnment a
s
sh
own in the e
q
uation 13.
(
1
3
)
7)
For a
de
ci
si
on p
r
oble
m
, all the p
o
ssible
re
sult
s from a
set
m
(
ϴ
). Th
en
, any
prop
ositio
n is a sub
s
et of
m
(
ϴ
), whic
h is
c
a
lled identific
a
tion frame.
8)
For a
n
id
entification f
r
ame
m
(
ϴ
) i
s
calle
d a fun
c
tion
m : 2
ϴ
→
[0,
1] (2
ϴ
is the
power
set
of
ϴ
) basic probabilit
y assignment
if m sa
tisfies the followi
ng conditions, as
sho
w
n i
n
the
equatio
n 1.
m(A)
i
s
called basic possibility assi
gnm
ent value,
which
pre
s
ent
s the level of trust to prop
ositio
n
A
.
9)
For
an i
dentif
ication
fram
e
m
(
ϴ
),
m : 2
ϴ
→
[0, 1] i
s
the basi
c probability assignm
ent
of
ϴ
, d
e
fine
function
Bel
as
Bel
:
2
ϴ
→
[0; 1].
Bel
(
A
)
is called
belie
f func
tion, as
sho
w
n in the equatio
n 3 prese
n
ts the su
m of
all the p
o
ssibilitie
s of A, which is al
so
the total evaluation of A.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Com
b
ining F
u
zzy L
ogi
c an
d Dem
p
ste
r
-Shafer The
o
ry (Andi
no Ma
sele
no)
589
10)
Given two
bel
ief function
s
Bel
1
,
Bel
2
, which a
r
e th
e sa
me identification fram
e,
m1
,
m2
are corresponding ba
sic
probability assignm
ents and their focal
element
s are
A1,
A2, …,
Am
an
d
B
1,
B
2
,
…, B
n
. The combin
ation:
m = m
1
⊕
m2
is defin
e
d
accordi
ng to
the Demp
ste
r
’s rule of
com
b
inat
ion a
s
shown in the e
quation 1
4
.
(
1
4
)
5. Conclusio
n
Fuzzy re
aso
n
ing d
o
e
s
th
e mappi
ng from given in
p
u
t to an outp
u
t usin
g Fu
zzy Logi
c.
Fuzzy re
ason
ing mod
e
ls
h
a
ve a num
be
r of rul
e
s
ba
sed on if th
en conditio
n
s. In fact, these
rule
s a
r
e e
a
sy to learn
an
d use an
d ca
n be m
odified
acco
rding
to
the situatio
n. It helps to m
a
ke
deci
s
io
ns a
n
d
can
be u
s
ed in de
cisi
on analy
s
is.
Demp
ster-S
hafer math
e
m
atical the
o
ry o
f
eviden
ce i
s
o
ne of th
e imp
o
rtant to
ol for de
ci
si
on m
a
king
un
der u
n
ce
rtainty. Dempste
r-S
haf
er
theory ma
ke
s inferen
c
e
s
from in
com
p
lete and
u
n
ce
rtain
kno
w
led
ge, provided by diffe
rent
indep
ende
nt kn
owl
edg
e
so
urce
s.
Dempste
r-S
haf
er
th
eory
provides expli
c
it e
s
timatio
n
of
impre
c
i
s
ion a
nd co
nflict b
e
twee
n information
from
different source
s and
can
deal with a
n
y
union
s of hyp
o
theses. De
mpste
r-Sh
a
fe
r mathem
atic
al theory of e
v
idence is a formal fra
m
e
w
ork
for plau
sibl
e
rea
s
oni
ng
whi
c
h p
r
ovid
es te
chni
qu
es for
ch
aracteri
zin
g
th
e eviden
ce
s b
y
con
s
id
erin
g
all the
availa
ble evid
en
ce
s. Demp
st
er-Shafer th
eo
ry has b
een
use
d
in
de
ci
sion
makin
g
. The kno
w
le
dge is unce
r
tain in
the collect
io
n of basic ev
ents can be
dire
ctly used
to
dra
w
co
ncl
u
si
ons in sim
p
le
case
s,
ho
we
v
e
r,
in
many
ca
se
s the various event
s a
s
soci
ated wit
h
each oth
e
r.
Rea
s
o
n
ing
u
nder un
ce
rtai
nty that u
s
ed
so
me
of ma
thematical
ex
pre
ssi
on
s, ga
ve
them a
different inte
rpreta
tion which i
s
each pi
ec
e of
eviden
ce
ma
y sup
port
a
subset contain
i
ng
several hypot
heses. Thi
s
is a gene
rali
zation of
the pure proba
bilistic frame
w
ork in which eve
r
y
finding
corre
s
pond
s to a
value of a
vari
able. In thi
s
rese
arch it i
s
Fuzzy Lo
gic
and
Demp
ste
r
-
Shafer th
eory, which resulted in
a 0
% reje
ct
ion.
Finally, Fu
zzy Logi
c an
d
Demp
ste
r-Sh
a
fer
mathemati
c
al
theory of eviden
ce have
shown goo
d re
sults.
Referen
ces
[1]
Z
adeh LA. F
u
z
z
y
Sets.
Information a
nd C
ont
rol
. 196
5; 8: 33
8-35
3.
[2]
Yen Z
.
General
izing T
he Dem
p
ster-Shafer
T
heor
y to F
u
zz
y Sets. 1990; 20
: 559-57
0.
[3]
Bina
ghi E, Gal
l
o I, Madel
la P
.
A Neural M
o
del for F
u
zz
y
Dempster-S
haf
er Class
ifiers.
Internatio
na
l
Journ
a
l of Appr
oxi
m
ate R
easo
n
in
g
. 200
0; 25: 89-12
1.
[4]
Yager
RR, F
i
l
e
v P. Inclu
d
in
g Prob
abi
listic
Unce
rta
i
nt
y i
n
F
u
zz
y Lo
gic
Contro
ller M
o
deli
ng
Usin
g
Dempster-S
haf
er T
heor
y
.
IEEE Transactions on System
s
,
Man, and Cy
bern
e
tics
. 199
0; 25: 122
1-
123
0.
[5]
D
y
m
o
va
L, Sev
a
stjan
o
v P. T
he op
erati
ons
o
n
intu
it
ionistic f
u
zz
y
values
in
the frame
w
o
r
k
of Dempster-
Shafer the
o
r
y
.
Know
led
ge-B
a
sed Syste
m
s
. 201
2; 35: 132-
143.
[6]
Ghasemi J, G
had
eri R, M
o
ll
aei
M
R
K, Ho
jj
atoles
lami S
A
. A Nov
e
l Fuzz
y
Dempster-S
hafer Inference
S
y
stem for Bra
i
n MRI Segme
n
tation.
Infor
m
ation Sci
enc
es
. 2013; 2
23: 20
5-22
0.
[7]
W
a
lije
w
s
ki JS,
Sosno
w
s
k
i ZA. Genetic T
unin
g
Fuzz
y
Demp
ster-Shafer D
e
cision
Rul
e
.
Task Quarterly
.
200
2; 6: 631-6
40.
[8]
Boudr
aa AO,
Bentab
et A, S
a
lze
n
stein
F, Guillo
n
L. Dem
p
ster-Shafer
’s Basic
Pro
b
a
b
il
i
t
y
Assi
gnme
n
t
Based
on F
u
z
z
y
Mem
bersh
i
p
F
unctio
n
s.
El
ectronic
Letter
s
on Co
mputer
Vision
an
d Image A
nalys
is
.
200
4; 4: 1-9.
[9]
Dutta P, Ali
T
.
F
u
zz
y
F
o
cal
Eleme
n
ts in Demp
ster-Sh
a
fe
r
T
heor
y
of Ev
ide
n
ce: Case
stud
y
i
n
Risk
Anal
ys
is.
Internatio
nal J
ourn
a
l of
Co
mp
uter
Applic
ations
. 2
011; 34: 4
6
-53.
[10]
Dempster AP. Upper an
d lo
w
e
r prob
ab
iliti
e
s ind
u
ced b
y
a
multiv
alu
e
d
map
p
in
g.
Ann. Math. Stat
.
196
7; 38: 325-
339.
[11]
Dempster AP. A
Gener
aliz
ati
on
of Ba
yes
i
a
n
infer
ence.
J
o
u
r
nal
of the
Roy
a
l Statistic
a
l S
o
ciety
. 19
68
;
30: 205-
24
7.
[12]
Shafer G. A Mathematic
al T
heor
y
of Evide
n
c
e. Ne
w
J
e
rse
y
: Princet
on U
n
iversit
y
Press.
1976.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 16, No. 3, Dece
mb
er 201
5 : 583 – 590
590
[13]
L
y
u
CH, C
hoi
MS, Li Z
Y
, Yo
un
HY.
Re
aso
n
in
g w
i
th I
m
pr
ecise
Co
ntext
Using
I
m
prov
e
d
D
e
mpster-
Shafer T
h
e
o
ry
. In: 2010 IEEE
/W
IC/ACM Inte
rnatio
nal C
onf
erenc
e on W
e
b Intelli
ge
nce
and Inte
lli
gen
t
Agent T
e
chnol
og
y. T
o
ronto,
Can
ada. 2
010:
475-4
78..
[14]
Yao S, Hua
n
g
W
.
A New
Appro
a
ch for Multi-attrib
ute D
e
ci
sion Making Problem
with Incom
p
lete
Information B
a
sed o
n
D
e
mpst
er-Shafer T
h
eo
ry
. In: Internationa
l Co
nfere
n
c
e
on M
a
n
agem
ent Scie
nc
e
and Ind
u
stria
l
Engi
neer
in
g (MSIE). Harbin. 2
011: 62
0-6
23.
[15] Yu X, Li M, Ping Z
.
Dempster-
S
hafer T
heory
as an Appl
ie
d A
ppro
a
ch to Scenar
io F
o
reca
sting Base
d
on Imprec
ise
Proba
bil
i
ty
. In: IEEE 12th International Confer
ence on Computer
and Information
T
e
chnolog
y. C
hen
gd
u, Chin
a. 2012: 9
75-9
8
0
.
[16]
Uph
o
ff C, Gritschne
der
DM, Schlichtm
ann
U.
Appl
icati
on
of De
mpster-S
haf
er T
heory
to T
a
sk Map
p
i
n
g
und
er Episte
mi
c Uncertai
nty
. In: IEEE International S
y
stems
Conf
er
ence (S
y
s
Con). Orlando, FL, USA.
201
3: 536-
541.
Evaluation Warning : The document was created with Spire.PDF for Python.