TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol. 12, No. 8, August 201
4, pp. 5793 ~ 5800
DOI: 10.115
9
1
/telkomni
ka.
v
12i8.622
5
5793
Re
cei
v
ed Ma
rch 2
1
, 2014;
Re
vised Ma
y 6, 2014; Acce
pted May 2
7
, 2014
Brain Emotional Learn
i
ng for Classification Problem
Re
za Mah
d
i Hadi*
1
, Saee
d Shokri
1
, Omid Sojodishijani
2
1
Departme
n
t of Computer En
g
i
ne
erin
g, Scien
c
e and R
e
se
ar
ch Branch, Isla
mic Azad Un
iv
ersit
y
Qazvi
n,
Iran
2
F
a
cult
y
of Co
mputer Eng
i
n
e
e
rin
g
, Qazvin b
r
anch, Islamic
Azad Un
iversit
y
Qazvin, Iran
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: reza.mh67
@
g
mail.c
o
m
A
b
st
r
a
ct
Emoti
o
n
a
l l
ear
nin
g
is new
tool i
n
the fie
l
d of
machi
ne
lear
nin
g
that the ins
p
ir
ed fro
m
li
mbi
c
system
. The v
a
rious
m
o
dels
of em
otio
nal learning (BEL)
have been succ
e
ssfully utili
z
ed
in m
a
ny
learning
prob
le
ms. F
o
r exa
m
p
l
e, contr
o
l a
ppl
icatio
ns
and
pred
ictio
n
pro
b
le
ms. In
this pa
per a
new
architectu
r
e
base
d
o
n
a br
a
i
n e
m
otion
a
l l
e
arni
ng
mo
del t
hat can
be
use
d
in cl
assific
a
ti
on pr
obl
e
m
(B
ELC). T
h
is
mo
del
is suita
b
l
e
for
hig
h
d
i
mens
io
nal c
l
assific
a
ti
on
app
licati
ons
. T
o
eval
uate
the
prop
ose
d
meth
od
hav
e
bee
n
compar
e it w
i
th the Multi
l
aye
r
Pe
rceptron (
M
LP), K-Neare
s
t Neigh
bor (K
NN), Naiv
e Ba
yes classifi
er a
n
d
Brain E
m
otion
a
l L
earn
i
ng-B
a
sed Pattern
R
e
cog
n
i
z
e
r
(BE
L
PR)
meth
ods
. T
he obtai
ne
d
results sh
ow
the
effectiveness
a
nd efficie
n
cy of
the propos
ed
meth
od for cla
ssificatio
n
prob
le
ms.
Ke
y
w
ords
: bra
i
n e
m
oti
ona
l le
arni
ng, classifi
cation, or
b
i
tofrontal cortex, a
m
yg
da
la, mach
ine l
earn
i
n
g
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
No
wad
a
ys cl
assificatio
n
method
s hav
e been
wid
e
l
y used in t
he area of
sci
en
ce,
engin
eeri
ng, indu
stry, busi
ness and me
dicin
e
[1]. Numero
us meth
ods have b
e
en pro
p
o
s
ed
for
cla
ssifi
cation
pro
b
lem
s
t
hat can
be
refe
r to
st
atistical
an
d
mathem
atical metho
d
s
and
intelligen
ce
-b
ase
d
meth
od
s. They
ca
n
be catego
ri
ze
d as: i
ndu
ctive or tran
sd
uctive, statistica
l-
based or non-stati
s
tical–based,
supervi
s
ed or unsuperv
ised methods [2]. Artifici
al intelligence-
based mod
e
l
s often bio-inspi
r
ed, such as ne
ural
networks a
nd evolution
a
ry com
puti
ng
techni
que
s
h
a
ve a l
ong
tradition
of bei
ng u
s
e
d
a
s
d
a
ta-d
riven
ap
proa
ch
es for
compl
e
x
syst
em
modelin
g. Neural n
e
two
r
ks a
r
e one
of the pow
e
r
ful cla
ssifie
r
s method
which have b
een
extensively u
s
ed fo
r cla
s
sification p
u
rp
o
s
e
s
. Emotion
a
l learni
ng ba
sed
comp
utat
ional mod
e
ls
[3
-
5] is a fairly n
e
w a
r
ea i
n
m
a
chi
ne lea
r
ni
ng that
of bio
-
inspired mo
d
e
ls. Recently, resea
r
chers
in
artificial i
n
tell
igen
ce try
to
present
co
mputat
ional
model
s of
Li
mbic System
(LS
)
. The
fi
rst a
comp
utationa
l model
of t
he LS
mod
e
l
wa
s
pro
p
o
s
ed by
Lucas
[4]. Furthe
rmore,
nume
r
ous
method
s of e
m
otional le
arning
have
be
en propo
se
d
for variou
s
appli
c
ation
s
. Lucas
et al. [6]
explicitly dete
r
mine
d the re
ward sig
nal a
nd pr
o
p
o
s
ed
the Brain Em
otional Lea
rni
ng (BEL) b
a
se
controlle
r whi
c
h ha
s b
een
su
ccessfully modified an
d
utilized in va
ri
ous
cont
rol a
pplication
s
[7, 8]
and p
r
edi
ctio
n pro
b
lem
s
[9, 10]. In this model, the re
ward si
gnal i
s
vital for upda
ting the learni
ng
weig
hts of
sy
stem. BEL i
s
a sim
p
le
co
mpositio
n of
Amygdala a
n
d
Orbitofront
al co
rtex in t
h
e
brain. An
othe
r mod
e
l in
spi
r
ed
emotion
a
l
learning
i
s
brain
emotio
nal lea
r
nin
g
based intelli
g
ent
controlle
r (BE
L
BIC) [6] th
at ha
s b
een
p
r
oven to
overcome u
n
certai
nty and
co
mp
lexity issu
es
of
other intellig
e
n
t controll
ers.
In the BELBIC algor
ith
m
, the emotion
a
l deci
s
ion m
a
kin
g
is neith
er
compl
e
tely cognitive n
o
r
behavio
ral. T
h
is
emotion
a
l
co
ntrolle
r i
s
widely
used i
n
differe
nt fie
l
ds
su
ch a
s
d
e
cision m
a
ki
ng
and
cont
rol
engin
eeri
n
g
appli
c
ation
s
and
rob
o
tics [8, 11]. Ot
her
model
s,
su
ch
as Brain E
m
otional
Lea
rning
Ba
se
d
Fuzzy Infere
nce
System
(BELFIS) [12]
. All
reviewed BE
L mo
dels a
r
e ba
se
d o
n
monotoni
c
re
inforceme
n
t l
earni
ng
and
need
an
inp
u
t
reward extra
c
ted fro
m
in
put data [13]
. In this
pap
er a ne
w a
r
chite
c
ture
ba
sed o
n
a b
r
ain
emotional le
a
r
ning mo
del that can be u
s
ed in cl
assifi
cation p
r
obl
e
m
(BELC). O
t
her model
s
are
develop
ed ba
sed o
n
brai
n
emotional le
arnin
g
for cl
a
ssifi
cation p
r
oblem
s. Such as [17] that is
prop
osed bra
i
n
em
otional
learni
ng
-ba
s
e
d
patte
rn re
cogni
zer
(BEL
PR)
to
solve multiple
in
put
–
multiple o
u
tp
ut cla
s
sificati
on a
nd
ch
ao
tic time
se
rie
s
p
r
e
d
iction
probl
em
s. Also, in
[18] t
he
prop
osed
a
classifier (ELi
EC)
ba
sed
o
n
b
r
ain
em
oti
onal l
earning
that
can
be
con
s
id
ere
d
a
s
a
n
ensemble
cl
assificatio
n
with a
different in
teg
r
atio
n me
cha
n
ism and
co
mb
ination al
gori
t
hm.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 8, August 2014: 579
3 –
5800
5794
Nume
ro
us
efforts h
a
ve b
een p
u
t into
developi
ng
regul
ari
z
ation
method
s to
increa
se th
e
gene
rali
zatio
n
of supe
rvised cla
ssifi
cat
i
on
alg
o
ri
thm
s
a
nd
red
u
ce the time
complexity of
the
learni
ng procedure. The p
r
ope
rtie
s of emotional lea
r
ning a
s
low
computation
a
l compl
e
xity and
fast training,
and its sim
p
licity has m
ade it a
powerful metho
d
o
logy in sup
e
rvise
d
learn
i
ng.
Whe
r
e the g
r
adient ba
se
d
method
s an
d evolutiona
ry algorith
m
s a
r
e hard to be
use
d
due to t
heir
high computa
t
ional com
p
le
xity.
The pap
er i
s
orga
nized a
s
follows: the brain
em
otiona
l learnin
g
is p
r
esented in S
e
ction
s
2 and
propo
sed meth
od i
s
pre
s
e
n
ted in
Section
3 .Section
4 p
r
e
s
ents the
Experime
n
tal re
sults
whe
r
e th
e
propo
sed
meth
od i
s
com
pared
with m
u
ltilayer perce
ptron (MLP),
K
-
Nea
r
e
s
t
Neig
hbor
(KN
N
), Naiv
e
Bay
e
s cla
s
s
i
fier
and
BELPR. In mult
ilayer pe
rcep
tron a
r
e va
ri
ous ve
rsion
s
of
backp
rop
agat
ion al
gorith
m
; the g
r
adie
n
t
desce
nt
ba
ckpro
pag
atio
n (GDBP) [1
6]. And finally
c
o
nc
lus
i
on
s
ar
e
ma
de
in
Se
c
t
io
n
5
.
2. Brain Emotional Learni
ng
The emotion
a
l
learnin
g
method is an inte
lligent algorit
hm in the field of machine
learni
ng
that develop
ed to red
u
ce
the compl
e
xity of co
mputations a
nd re
duce the time com
p
lexity in
learni
ng p
r
ob
lems [3, 4]. Emotional le
arnin
g
is in
spired f
r
om li
mbic
system.
The emotio
nal
learni
ng m
o
d
e
ls that first i
n
trodu
ce
d by
More
n an
d B
a
lke
n
iu
s in hi
s Ph.D. the
s
i
s
an
d devel
o
ped
a netwo
rk
re
pre
s
entatio
n as a computa
t
ional model
t
hat mimics the amygdala, the orbitof
r
ont
al
cortex, the th
alamu
s
, the sensory input
cortex, and,
g
enerally, those parts
of the brain tho
ught
to
be
re
spon
sibl
e for p
r
o
c
e
s
sing e
m
otion
s
[4, 5]. Lim
b
i
c
system
is t
he
cent
ral
ba
se
of em
otio
nal
intelligen
ce t
hat ha
s several
sub m
o
dule
s
that e
a
ch
ha
s its
spe
c
ial fu
ncti
onality and t
h
e
emotional
lea
r
ning
o
c
curs
mainly in th
e
amy
gdal
a. T
w
o im
po
rtant
pa
rts
of the
limbic
syste
m
,
amygdala a
n
d
orbito
-front
al cort
ex
(OF
C
), are re
sp
o
n
sibl
e for pr
o
c
e
ssi
ng the e
m
otional si
gn
als.
The main fea
t
ure of the limbic sy
stem is that
the we
ights of amyg
dala ca
nnot d
e
crea
se (call
e
d
monotoni
c le
arnin
g
).
Figure 1
sh
ows the
gra
phical mo
del
of t
he
se
n
s
ory
sig
nal
and l
earning
network
con
n
e
c
tion m
odel in
side th
e brain [3]. In Thalamu
s
, some poo
r pre
-
processin
g
o
n
sen
s
o
r
y inp
u
t
sign
als
such as noi
se red
u
c
tion or
filteri
ng can
be
do
ne in
this pa
rt. The a
m
ygd
a
la, is very
well
placed to re
ceive stimuli from all sen
s
o
r
y cortice
s
an
d the orbitofrontal co
rtex
is thoug
ht to inhibit
inapp
rop
r
iate
respon
se
s from the amyg
dala, ba
sed
on the conte
x
t giv
en by the hipp
ocam
pus
[3].
Table
s
and Fi
gure
s
a
r
e pre
s
ente
d
ce
nter, as
sho
w
n b
e
low a
nd cite
d in the manu
script.
Figure 1. Gra
phical Model
of the Brain Emotional Le
arning Pro
c
e
ss
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Brain Em
otional Lea
rnin
g for Cla
s
sificati
on Probl
em
(Re
z
a Mah
d
i Hadi
)
5795
The follo
win
g
limbic sy
ste
m
structu
r
e
s
are
currentl
y
thought to
be
mo
st inv
o
lved in
emotion:
a) Amygdala
b) Hipp
ocampu
s
c) Forni
c
ate
gyrus
d) Hypothal
amu
s
e) Orbitofrontal
cortex
a
) Amy
gdala
The
amygdal
a is pa
rt of
the limbi
c sy
stem
[3]
whi
c
h
ba
sed
on
the
brai
n e
m
otiona
l
learni
ng (BEL) creates
e
m
otional
i
n
te
lligen
ce.
Am
ygdala
re
cei
v
es pla
s
tic conne
ction
s
from
sen
s
o
r
y co
rt
ex and thal
amus
and t
he internal reinforce
cau
s
ed
by external
reward
and
puni
shme
nt [3]. The amy
gdala
are rel
a
ted in
det
e
c
ting and
lea
r
ning of
our surroun
ding
s
are
importa
nt an
d have
emoti
onal
signifi
ca
nce. T
her
e a
r
e
several in
puts to th
e a
m
ygdala th
at are
comin
g
st
raig
ht from the
sensory a
r
ea
s.
Out
puts from
the amygdal
a are th
e con
d
itioned
sign
als
to the OFC, a
nd the emotio
nal co
nditioni
ng.
b) Orbito
fro
n
tal Cor
t
e
x
Biological Fo
undatio
ns
Worki
ng
clo
s
el
y with the a
m
ygdala i
s
t
he o
r
bitofro
n
tal co
rtex
(OF
C
), which
will evaluate
the acti
vity of the amygdal
a
in context.
T
he orbitofront
al co
rtex (OF
C
)
is
a regi
on o
f
asso
ciation cortex of
the
huma
n
b
r
ain
involved in
cog
n
itive pro
c
e
s
ses su
ch
as
deci
s
io
n ma
king. It is critically en
gage
d in em
otio
n
a
l pro
c
e
s
sin
g
and i
nhibit
o
ry co
ntrol f
o
r
behavio
r mo
nitoring
by a
ssi
gnin
g
valu
e in de
cisi
on
makin
g
me
chani
sm whereas
orbitof
r
o
n
tal
cortex
le
sion
s a
r
e
kno
w
n
to h
a
ve a
b
n
o
rmal
be
havi
o
r
and
emoti
onal i
r
regul
arities. Th
e m
a
in
function
of orbitofrontal
co
rtex is th
oug
ht to be
in
hi
bitory when
e
v
er the e
m
oti
onal
rea
c
tion
is
assume
d to
be i
n
conv
enient
due
to reinforcement, in
fact o
r
bitofro
n
tal cortex
ha
s
interconn
ecti
ons
with hipp
ocam
pu
s, an
d also it
ha
s bidire
ction
a
l conne
ction
s
wi
th amygdala.
In Figure
1, there i
s
on
e A node for e
v
ery st
imulu
s
X, including
one for the
thalamic
stimulu
s
. The
r
e is al
so on
e
O node for e
a
ch of the
sti
m
uli, except for the
thalami
c
nod
e. There is
one
output n
ode E th
at is
comm
on fo
r
all the o
u
tput
s of th
e mo
de
l. The E n
ode
simply
sum
s
the
outputs from
the A no
des
and the
n
subt
ract
s the
inhi
bitory outp
u
ts from the
O
n
ode
s. The
re
sult
is the output from the mod
e
l. In other word
s, E can b
e
obtaine
d fro
m
:
∑
∑
(
1
)
One of amygdala inp
u
ts is called th
alamic
con
n
e
c
tion and
cal
c
ulate
d
as the
ma
x
min
overall Sen
s
ory Input
as equatio
n. Like
wise, the
node
sum
s
the o
u
t
puts from
, except
and then subtract
s from in
hibit
o
ry output
s from the
node
s. The le
arnin
g
rul
e
of the amygdala
is given as:
(2)
∆
m
a
x
0
,
∑
(
3
)
Whe
r
e
the weig
ht in the amygdal
a con
n
e
c
tion is,
is the lea
r
ning
step i
n
the
amygdala, an
d
is
the valu
e of reward si
gnal. The term
in (3) is fo
r making the learni
ng
cha
nge
s mo
n
o
tonic, implyi
ng that the a
m
ygdala
wei
ght ca
n neve
r
be de
crea
se
d. This
rule i
s
for
modelin
g the
incapability
of unlea
rni
n
g
the em
otion
sign
al p
r
evio
usly lea
r
n
ed i
n
the a
m
ygd
a
la.
Similarly, the learni
ng rule in the orbitofrontal co
rtex is given a
s
:
(
4
)
∆
′
(5)
Whe
r
e
is the weight in the
orbitofro
n
tal con
n
e
c
tion, a
nd
is the lea
r
ning ste
p
in the
orbitofrontal cortex. Also
′
is
:
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 8, August 2014: 579
3 –
5800
5796
′
∑
∑
(
6
)
In Figure 1, the top is the
orbitofro
n
tal,
the bottom right is the a
m
ygdala, and
the left
contai
ns the
thalamic a
n
d
sen
s
ory cort
ical mod
u
le
s. The sen
s
o
r
y inputs
ente
r
the thalami
c
part, whe
r
e a
thalamic in
pu
t to the amyg
dala is
calcul
ated as the m
a
ximum or mi
nimum or me
an
over all in
put
s. A prima
r
y reward sig
nal
enters both
the amygdal
o
i
d
and o
r
bitof
r
ontal p
a
rt
s.
In fact, by re
ceiving th
e sensory inp
u
t
, the model
calcul
ates th
e internal
sig
n
a
l
s of amygd
a
l
a
(
) and orbitofrontal co
rtex (
). Since the a
m
ygdala d
o
e
s
not have th
e cap
ability to unlea
rn a
n
y
emotional
respon
se th
at it
ever l
earned,
inhibiti
o
n
of
any ina
p
p
r
op
riate
re
spo
n
se is the
duty
of
the orbitofron
tal cortex. T
he amygd
a
la
part re
ceive
s
input
s from
the thalamu
s
and f
r
om t
he
corti
c
al
are
a
s
, while the
orbitofrontal
part
re
ce
ive
s
inp
u
ts f
r
o
m
the
corti
c
al area
s an
d
the
amygdala o
n
l
y
.
Table
s
and Fi
gure
s
a
r
e pre
s
ente
d
ce
nter, as
sho
w
n b
e
low a
nd cite
d in the manu
script.
3. Proposed
Model
In this
sectio
n, the struct
ure
of pro
p
o
s
ed
method
i
s
introdu
ced.
In this p
ape
r a n
e
w
architectu
re
based on a brain em
otio
nal learni
ng
model that can be used in classification
probl
em (BEL
C). Prop
osed
method can
be use
d
in
cl
assificatio
n
and pre
d
ictio
n
probl
em
s. In the
cla
ssifi
cation
probl
ems
re
quire
s a trai
ning data
s
et.
Every insta
n
ce in
any
dataset use
d
b
y
machi
ne lea
r
ning alg
o
rith
ms is
rep
r
e
s
ented u
s
ing t
he sa
me set of feature
s
[1
4]. In the BELC
model, a
n
y feature
of the
data
s
et i
s
in
put patterns
model. T
he i
nput p
a
ttern
is illu
strate
d
by
vec
t
or
:
,
,…,
,…,
1
,2,
…
,
(
7
)
,
,…,
,…,
1
,2,
…
,
(8)
Whe
r
e
is th
e instan
ce si
ze and m is the feature si
ze and
is a
n
instan
ce a
nd
determi
ne
s the lab
e
l cla
ss of
. Als
o
is cal
c
ulate
by
ma
x
min
. Where the t
r
ainin
g
data
s
et are n
o
rm
al
ized b
e
fore e
n
tering to
model bet
we
en [-1.0, +1.
0
]. Figure 2
illustrated t
he pro
p
o
s
ed
method ba
sed on BEL for
cla
ssifi
cation
probl
em
s. Ou
tputs of the model can be o
b
tained fro
m
:
∑
∑
(
9
)
Whe
r
e the a
c
tivation function ca
n be:
∗
1
(
1
0
)
∗
1
(11
)
Also, the val
ues
of amyg
dala (
)
an
d Orbitofrontal cortex (
) are
cal
c
ulated
b
y
followin
g
equ
ations:
(
1
2
)
(
1
3
)
Output of am
ygdala
(
) that is u
s
e
d
for
adju
s
ting the
plasti
c
conn
ection
wei
ght
s an
d
output of Orb
i
tofrontal co
rtex (
) that is use
d
for inha
biti
ng the amygdala output
. Update the
weig
hts are calcul
ated by followin
g
equ
a
t
ions:
m
a
x
0
,
∑
(
1
4
)
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Brain Em
otional Lea
rnin
g for Cla
s
sificati
on Probl
em
(Re
z
a Mah
d
i Hadi
)
5797
′
(15
)
The value
a
nd
are the
best weig
ht
and
during
training and
is a
rand
om num
ber. Find the
best value
s
and
simply calcul
ated by followin
g
equ
a
t
ions:
(
1
6
)
∑
∑
(
1
7
)
(
1
8
)
(
1
9
)
Figure 2. Gra
phical model
of the BELC
The
and
values
ca
use
a fast conv
erge
nce toward
s the
be
st wei
ght
coeffici
ents.
The m
odel
al
so
nee
ds a
ta
rget
asso
ci
at
ed to
input
pa
ttern to
adju
s
t
the
wei
ghts.
Let
Rw b
e
target
value (
) to pattern (
).
1
,2,
…
,
(
2
0
)
The value
′
are cal
c
ulate
d
by this equati
on:
′
∑
∑
(
2
1
)
The value pa
rameters in propo
sed meth
od sh
own in the followi
ng table.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 8, August 2014: 579
3 –
5800
5798
Table 1. Meth
odParamete
r
s
Parameters
Value
Number of
epoch
100
21
0
11
0
Rw Function
4. Experimental Re
sult
In this
s
e
c
t
ion, we tes
t
the
BELC method to
clas
sif
y
sev
e
ral dat
a
s
e
t
s that have obtained
from the Univ
ersity of Calif
ornia, Irvine
(UCI)
m
a
chin
e learni
ng re
posito
r
y [15]. Every instance
in any data
s
et use
d
by
machi
ne le
arning alg
o
rith
ms is
re
pre
s
ented u
s
in
g the sam
e
se
t of
feature
s
. Th
e features m
a
y be
contin
uou
s, cate
go
rical
or bin
a
ry. Table 2
shows the
m
a
in
cha
r
a
c
teri
stics of the data
s
ets. Th
e de
vel
oped m
e
th
od wa
s
comp
ared
with G
D
BP MLP, KNN,
Naive Baye
s cla
ssifie
r
an
d
BELPR. Also, to evaluate the BELC i
n
the cla
s
sification p
r
o
b
le
ms,
mean
squ
a
re error
(MS
E
) and a
ccura
cy are
p
e
rform
a
n
c
e
measures th
at are g
ene
rally
expre
s
sed a
s
follows:
∑
(
2
2
)
(
2
3
)
For all of the learni
ng ste
p
, the training set
contain
s
7
0
% of the data and the testing set
contai
ns 3
0
%.
Table 2. Data
sets
Datase
t
Attrib
ute
c
harac
t
eristics
Insta
n
ces
Attrib
ute
Classes
Iris Real
150
4
3
Glass Real
214
9
7
Sonar
Real
208
60
2
Pim
a
Integer,
R
eal
768
8
2
Table 3. Accu
racy of Cla
s
si
fication Resul
t
for Iris Data
set duri
ng 10
Run
s
M
e
t
h
od
Accuracy (
%
)
Ma
x
Me
a
n
BELC 0.97
0.85
BELPR 0.78
0.70
GDBP M
L
P
0.97
0.91
KNN 1.00
0.95
Bayesian
0.97
0.02
Table 4. Accu
racy of Cla
s
si
fication Resul
t
for Glass Da
taset du
ring 1
0
Run
s
M
e
t
h
od
Accuracy (
%
)
Ma
x
Me
a
n
BELC 0.84
0.75
BELPR 0.71
0.68
GDBP M
L
P
0.76
0.76
KNN 0.84
0.82
Bayesian
0.96
0.86
Table 5. Accu
racy of Cla
s
si
fication Resul
t
for Sonar Dataset du
ring
10 Ru
ns
M
e
t
h
od
Accuracy (
%
)
Ma
x
Me
a
n
BELC 0.86
0.73
BELPR 0.74
0.70
GDBP M
L
P
0.85
0.79
KNN 0.86
0.75
Bayesian
0.79
0.74
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Brain Em
otional Lea
rnin
g for Cla
s
sificati
on Probl
em
(Re
z
a Mah
d
i Hadi
)
5799
Table 6. Accu
racy of Cla
s
si
fication Resul
t
for Pima Dataset du
ring 1
0
Run
s
M
e
t
h
od
Accuracy (
%
)
Ma
x
Me
a
n
BELC 0.79
0.74
BELPR 0.82
0.75
GDBP M
L
P
0.81
0.79
KNN 0.78
0.72
Bayesian
0.78
0.72
Figure 3. Maximum Accura
cy Comp
ari
s
o
n
betwe
en th
e BELC, BELPR, GDBP MLP, KNN and
Bayes
i
an
Figure 4. Minimum MSE Compa
r
ison be
tween the BE
LC, BELPR, GDBP MLP, KNN an
d
Bayes
i
an
4. Conclusio
n
This pap
er p
r
esents
a ne
w cla
ssifie
r
t
hat
is
inspire
d
by the
brai
n emotio
nal l
earni
ng
(BELC). Ho
wever, the BE
LC differs
fro
m
other
meth
ods i
n
the
wa
y that the cla
ssifie
r
s are fe
d
.
The p
e
rfo
r
m
ance of BEL
C i
s
evalu
a
ted by
cl
a
s
sifying seve
ral
ben
c
hm
ark data
set
s
and
comp
ared
wi
th differe
nt cl
assifier meth
od. Th
e
re
su
lts indi
cate
a
fairly g
o
od
perfo
rman
ce
of
BELC for c
l
ass
i
fic
a
tion.
Referen
ces
[1]
Sheik
hol
esl
a
mi
N, et al. A
ppl
ying
brai
n em
oti
ona
l
le
arn
i
ng
a
l
gorit
hm for mu
ltivaria
bl
e co
ntrol of H
V
A
C
s
y
stems.
Jour
n
a
l of Intelli
ge
nt and F
u
zz
y
Sys
t
ems
. 20
06; 17
(1): 35-46.
[2]
G
holi
pour
A, Caro Luc
as, Dani
al
Sha
h
m
i
rzadi.
Pr
ed
icti
ng
geo
mag
net
ic activity
in
d
e
x by
bra
i
n
emotio
nal l
ear
nin
g
. WSEAS
T
r
ansactions on S
y
stem
s. 2004; 3(1): 296-299.
[3]
Sutikno T
,
F
a
cta M, Arab Mar
k
ade
h G
.
R. Progress i
n
Artifi
cial Intel
l
i
genc
e T
e
chniqu
es: from Brain t
o
Emotion.
TEL
K
OMNIKA
. 2011
; 9(2): 201-20
2
.
[4]
Caro
Luc
as, D
ani
al S
h
a
h
mirz
adi,
Nima
Sh
ei
khol
eslam
i
.
Introduc
ing
BELBI
C: bra
i
n
e
m
oti
ona
l l
ear
nin
g
base
d
inte
lli
ge
nt controll
er
. Intellig
ent Autom
a
tion & Soft Co
mputin
g. 200
4; 10(1): 11-2
1
.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 8, August 2014: 579
3 –
5800
5800
[5]
Lucas, C
a
ro, R
a
sou
l
M Mil
a
si,
Babak
N Ara
a
b
i. Intell
ig
ent M
ode
lin
g a
nd
Co
ntrol Of W
a
shi
ng Mac
h
in
e
Using
L
o
cal
l
y
Lin
ear Neur
o
‐
F
u
zz
y
(
L
l
n
f) Mode
lin
g A
nd
Modifi
ed Br
ain
Emotio
nal
L
e
a
rni
ng B
a
se
d
Intelli
gent C
ont
roller (Be
l
b
i
c).
Asian Jo
urn
a
l of Control.
20
0
6
; 8(4): 393-4
0
0
.
[6]
Gholi
pour
A,
C L
u
cas, D
Shahm
irzad
i
.
Purpos
eful pr
edicti
on of
sp
ace
w
e
ath
e
r phe
nome
n
a
b
y
simulat
ed emot
ion
a
l le
arni
ng.
Internati
o
n
a
l Jo
urna
l of Model
li
ng an
d Si
mu
lat
i
on.
20
04; 24(
2
)
: 65-72.
[7]
LeD
ou
x, Jose
ph E.
E
m
otio
n circuits i
n
the br
ain
. F
O
CUS: T
he Journa
l of Life
lo
ng L
earn
i
n
g
i
n
Ps
y
c
h
i
atr
y
.
20
09; 7(2): 27
4-2
74.
[8]
Kotsiantis, S
o
tiris B, Ioan
nis
D Z
ahar
akis,
Pana
yi
otis E
Pintel
as.
Ma
chin
e le
arn
i
ng:
a revi
ew
of
classificati
on a
nd co
mb
ini
ng t
e
chn
i
qu
es
. Artificial Intel
lig
enc
e Revi
e
w
. 200
6; 26(3): 15
9-1
90.
[9]
Lotfi, Ehsan, MR Akbarza
de
h T
.
Brain Emotion
a
l Le
arn
i
n
g
-Base
d
Pattern Reco
gni
z
e
r
. Cy
be
r
n
e
t
i
cs
and S
y
stems. 201
3; 44(5): 40
2-42
1.
[10]
Hua
ng, Guo
y
o
ng, Z
i
yan
g
Z
h
e
n
, Dao
bo W
a
n
g
.
Brain
emotio
nal l
ear
nin
g
b
a
s
ed i
n
tell
ige
n
t control
l
er fo
r
nonlinear syst
em
. Inte
lli
gent
Information
T
e
chnolog
y A
pplic
atio
n. IIT
A
'
08. Secon
d
Internatio
na
l
S
y
mp
osi
u
m on
. IEEE. 2008; 2.
[11]
Mahd
i Ha
di, R
e
za, Sae
i
d S
h
okri, Pe
yma
n
A
y
u
b
i.
Ur
mi
a L
a
ke l
e
vel for
e
c
a
sting
usin
g Br
ain E
m
otion
a
l
Lear
nin
g
(BEL
)
. Computer a
nd Kno
w
l
e
dg
e
Engin
eeri
ng (I
CCKE), 3th Internati
ona
l eCo
n
ferenc
e on.
IEEE. 2013.
[12]
Mohamm
d
i Mil
a
si R, C Luc
as, B Nadj
ar-
A
rrabi.
Spe
e
d
control of a
n
interior
per
mane
nt magn
et
synchro
nous
motor
us
ing
B
E
LBIC (br
a
in
e
m
oti
o
n
a
l
le
arni
ng
base
d
intel
l
i
ge
nt contr
o
ll
er
)
. Automation
Con
g
ress. Pro
c
eed
ings. Worl
d. IEEE. 2004; 16.
[13] Parsap
oor,
Ma
hbo
ob
eh, Urba
n Bilstrup.
Bra
i
n Emotio
nal
L
earn
i
ng B
a
sed
F
u
zz
y
Inferen
c
e Syst
e
m
(BELFIS) for Solar
Ac
tivity
Forecasting
.
T
ools
w
i
th
Artificial Int
e
ll
i
genc
e (ICT
AI). IEEE 24th
Internatio
na
l C
onfere
n
ce o
n
. IEEE. 2012; 16.
[14]
Pasrap
oor, M
a
hbo
ob
eh, Ur
ba
n Bi
lstrup.
A
n
Emoti
o
n
a
l Le
ar
nin
g
-ins
pire
d
E
n
se
mb
le Class
i
f
ier
(EL
i
EC).
Co
mp
uter Scie
nce an
d Info
r
m
ation Syste
m
s
(F
edCSIS).
Federate
d
Co
nfer
ence o
n
. IEEE. 2013.
[15]
J Moren, C Bal
k
eni
us. Emotio
n and
Le
arni
ng
-A Computati
o
nal Mo
del
of the Am
ygda
la, P
h
.D.
T
hesis,
Lun
d Univ
ersit
y
, S
w
e
den. 2
0
02.
[16]
CC Blak
e, C
Merz. UCI Re
posit
or
y
of Ma
chin
e Le
arn
i
ng
Datab
a
ses. D
ept. Informatio
n
& Com
pute
r
Sciences, Univ. California,
Irvine. 1998. Avai
lable: http://arc
hi
ve.ics.uci.edu/ml/.
[17] C Balk
eni
us, J Moren.
A co
mputatio
na
l
mod
e
l of e
m
otio
nal
cond
ition
i
n
g
in
the bra
i
n
. Proc. Workshop
Ground
ing Em
otion Ad
aptiv
e S
y
st, Z
u
rich, Sw
itz
e
rla
nd. 19
9
8
.
[18]
NNg, S
i
n
Ch
u
n
, Shu
H
ung
Leu
ng, A
ndre
w
L
u
k. F
a
st
conver
gent
ge
ner
a
lize
d
back
-
prop
agati
o
n
alg
o
rithm
w
i
th
constant le
arn
i
ng rate.
Neur
al
Processin
g
Le
tters.
1999; 9(1
)
: 13-23.
Evaluation Warning : The document was created with Spire.PDF for Python.