Internati
o
nal
Journal of Ele
c
trical
and Computer
Engineering
(IJE
CE)
Vol.
4, No. 6, Decem
ber
2014, pp. 931~
938
I
S
SN
: 208
8-8
7
0
8
9
31
Jo
urn
a
l
h
o
me
pa
ge
: h
ttp
://iaesjo
u
r
na
l.com/
o
n
lin
e/ind
e
x.ph
p
/
IJECE
Learning Style Classificati
on via EEG Sub-band Spectral
Centroid Frequency Features
M
e
gat
Sya
h
irul A
m
in M
e
gat
Ali, Aisy
ah
Ha
rtini Ja
hidi
n,
Noo
r
ita
w
ati Md T
a
hir,
Mohd
Na
sir T
a
ib
Faculty
of Electr
ical
Eng
i
neering
,
Univ
ersiti Tekn
ologi MARA
, Selangor, Malay
s
ia
Article Info
A
B
STRAC
T
Article histo
r
y:
Received Sep 11, 2014
Rev
i
sed
O
c
t 12
, 20
14
Accepted Oct 30, 2014
Kolb’s Experien
tial Learn
i
ng Th
eor
y
pos
tulates that in learning,
knowledge
is cre
a
ted
b
y
th
e
lea
r
ners’ ab
ili
t
y
to absorb
and tr
ansform
experie
n
ce.
Ma
n
y
studies hav
e
previously
sugge
sted that at rest,
th
e br
ain emits sig
n
atures
that
can b
e
associ
at
e
d
with cogn
itiv
e
and b
e
haviour
a
l
pat
t
erns.
Henc
e,
the stu
d
y
att
e
m
p
ts
to charact
eris
e and c
l
as
s
i
f
y
l
earning
s
t
y
l
es
from
EEG us
ing the
spectral centroid
frequency
featu
r
es. Init
ially
,
learning sty
l
e of 68
univers
ity
stude
nts ha
s be
en a
sse
sse
d using Kolb’s
Learnin
g
Sty
l
e Inventor
y
.
Resting
EEG is
then re
corded from
th
e prefr
ont
al cor
t
ex. Nex
t
, th
e
EEG is
pre-
processed and f
i
lter
e
d into
alph
a and
theta sub-b
a
nds in which the spectr
a
l
centro
i
d frequen
c
ies are computed
from the corresponding power spectral
densities
.
The d
a
tase
t is further
enhanc
ed to 160
sam
p
les via sy
n
t
heti
c EEG
.
The obtained features are th
en
used as input
to the
k
-nearest neighbour
clas
s
i
fi
er that
is
incorporat
ed with
k
-fold cross-validation. Feature
clas
s
i
fi
cat
ion vi
a
k
-n
ear
est n
e
ig
hbour has attain
ed fiv
e
-fold
mean tr
ainin
g
and testing accu
racies of 100% a
nd 97.5%
, respectiv
ely
.
Hence, r
e
sults show
that
the
a
l
pha
a
nd the
t
a s
p
ectr
a
l
cen
tr
oid
frequencies r
e
presen
t
distinct an
d
stable EEG
sign
ature to
distingu
ish
learn
i
ng sty
l
es from the r
e
stin
g brain
.
Keyword:
EEG
k
-f
ol
d c
r
oss-
va
l
i
d
at
i
o
n
k
-
n
ear
e
st n
e
ighb
our
Learni
ng style
Spectral ce
ntroid freque
ncy
Copyright ©
201
4 Institut
e
o
f
Ad
vanced
Engin
eer
ing and S
c
i
e
nce.
All rights re
se
rve
d
.
Co
rresp
ond
i
ng
Autho
r
:
Meg
a
t Syah
irul Am
in
Meg
a
t Ali,
Facu
lty of Electri
cal Engineering,
Un
i
v
ersiti Tekn
o
l
o
g
i
M
A
RA,
4
045
0 Sh
ah A
l
a
m
, Selan
gor
,
Malaysia.
Em
a
il: meg
a
tsyah
i
ru
l@salam.u
itm
.ed
u
.
m
y
1.
INTRODUCTION
Learni
ng
st
y
l
es an
d e
xpe
ri
en
t
i
a
l
l
earni
ng
ha
ve
bot
h be
en
p
a
ram
ount
t
o
t
h
e co
nst
r
uct
i
v
i
s
m
t
h
eory
i
n
educat
i
o
n,
w
h
i
c
h p
r
o
p
o
ses k
n
o
wl
e
dge as
bei
ng c
r
eat
ed t
h
ro
ug
h a p
r
oc
ess
of i
n
t
e
ract
i
on
bet
w
ee
n ex
per
i
ence
and i
d
eas.
Although
receivi
ng c
r
iticis
m
,
t
h
e conce
p
t ha
s been acce
pt
ed due to
its success in
fostering
effective
teaching, whic
h ai
m
s
to provi
de
optim
a
l
e
xpe
rience t
o
lea
r
ners
with
va
rying style prefe
r
ences
.
Seve
ral
l
earni
ng
st
y
l
e
m
ode
l
s
have
bee
n
est
a
bl
i
s
hed
w
h
i
c
h i
n
cl
u
d
e
C
u
r
r
y
’
s
Oni
o
n
M
odel
,
R
i
di
n
g
an
d
Ch
eem
a’
s Fu
nd
am
en
tal D
i
men
s
ion
s
,
D
unn an
d
Dun
n
’
s
Lear
n
i
n
g
Style Mo
d
e
l, and K
o
lb’
s
Exp
e
rien
tia
l
Learni
ng The
o
ry (ELT
) [1].
Com
p
ara
tiv
ely,
Ko
lb’s
ELT h
a
s b
een
wid
e
l
y
ad
op
ted
i
n
th
e f
i
eld of
educatio
n
and
al
so
i
m
pl
em
ent
e
d i
n
m
a
n
a
gem
e
nt
l
earni
ng
[
2
]
.
Ko
l
b
’s ELT has d
e
fin
e
d
th
at k
n
o
wledg
e
i
s
created
th
rou
g
h
th
e creati
v
e ab
ility o
f
in
d
i
v
i
du
als in
gras
pi
n
g
an
d t
r
ans
f
orm
i
ng e
xpe
ri
ence
. T
h
e
g
r
aspi
ng
di
m
e
nsi
o
n i
s
re
pr
esent
e
d
by
a
pai
r
o
f
di
al
ect
i
cal
l
y
-
rel
a
t
e
d l
ear
ni
n
g
m
odes c
o
m
p
ri
si
ng
o
f
C
o
nc
ret
e
Ex
peri
e
n
c
e
an
d A
b
st
ract
C
once
p
t
u
al
i
s
a
t
i
on. M
e
a
n
w
h
i
l
e, t
h
e
t
r
ans
f
o
r
m
a
ti
on
di
m
e
nsi
o
n
i
s
descri
bed
by
anot
her pai
r
of
di
al
ect
i
cal
l
y
-rel
a
t
e
d l
earni
n
g
m
odes whi
c
h c
onsi
s
t
s
of R
e
fl
ect
i
v
e
Obse
r
v
at
i
on a
nd
Act
i
v
e Ex
p
e
ri
m
e
nt
ati
on. I
n
ex
peri
e
n
t
i
a
l
learni
ng
, k
n
o
w
l
e
dge i
s
bei
ng
creat
ed
th
ro
ugh
a
pro
c
ess th
at im
p
lic
ates a creative
ten
s
ion
b
e
tw
ee
n t
h
e l
ear
ni
n
g
di
m
e
nsi
ons
w
h
i
c
h i
s
resp
o
n
si
ve t
o
th
e con
t
ex
tual d
e
m
a
n
d
s
. Th
e
m
o
d
e
l p
o
r
trayed
th
e learn
i
ng
p
r
o
cess as a
recu
rsi
v
e cycle wh
ere ind
i
v
i
d
u
als will
expe
rience
,
reflect, think a
n
d
act [3].
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJECE Vol. 4, No. 6, D
ecem
ber 2014
:
931 – 938
93
2
Vari
at
i
o
ns i
n
i
ndi
vi
d
u
al
l
ear
n
i
ng
style arise
from
unique prefere
n
ces
to reso
lv
e t
h
e co
nflict o
f
b
e
ing
conc
rete or a
b
stract, a
nd
active or
refl
ective [3].
Such c
h
aracteristics are a
ttrib
u
t
ed
to
ind
i
v
i
du
al
specialisations
in education, past expe
riences
, contex
t
an
d
g
e
nde
r [
4
]
.
He
n
ce ove
r a l
o
n
g
peri
od
of t
i
m
e, t
h
e
co
nstru
c
t
represen
ts a stab
l
e
trait o
f
p
e
rso
n
a
lity [5
].
Ind
i
v
i
du
al learn
i
n
g
style is assessab
l
e v
i
a
Ko
l
b
’s
Learni
ng St
y
l
e In
vent
ory
(
L
SI
), i
n
w
h
i
c
h t
h
e d
o
m
i
nant
m
odes f
r
o
m
t
h
e graspi
n
g
an
d t
r
ans
f
o
r
m
a
ti
on
di
m
e
nsi
ons are
i
d
ent
i
f
i
e
d an
d
m
a
pped t
o
ei
t
h
er t
h
e Di
ver
g
i
ng,
Assi
m
i
l
a
t
i
ng
, C
o
n
v
e
r
gi
n
g
or
Accom
m
odat
i
ng
styles [3].
St
udi
es
ha
ve s
h
o
w
n t
h
at
ge
n
d
er
p
o
ses a si
gni
fi
cant
i
m
pact
on t
h
e
g
r
as
pi
n
g
di
m
e
nsi
o
n,
but
i
s
n
o
t
si
gni
fi
ca
nt
i
n
t
h
e t
r
ans
f
o
r
m
a
t
i
on di
m
e
nsi
on [4]
.
S
u
c
h
charact
eri
s
t
i
c
s are i
n
fl
ue
nce
d
by
t
h
e di
ffer
e
nces i
n
t
o
p
o
l
o
gi
cal
o
r
g
a
ni
sat
i
o
n
o
f
t
h
e brai
n’s
f
u
nct
i
onal
net
w
o
r
k
whi
c
h af
fect
s i
ndi
vi
d
u
al
s i
n
t
e
rm
s of be
ha
vi
ou
r
a
n
d
cog
n
i
t
i
on
[
6
]
.
Fi
ndi
ngs
ha
ve al
so i
n
di
c
a
t
e
d t
h
at
rel
i
a
bl
e pat
t
e
r
n
s
of
brai
n de
act
i
v
at
i
on are
oft
e
n
co
m
p
le
m
e
n
t
ed
b
y
in
crease i
n
cog
n
itiv
e d
e
m
a
n
d
s. Low level b
a
selin
e cond
itio
n
s
were activ
e states and
th
at
pat
t
e
rn
of act
i
v
at
i
on a
n
d dea
c
t
i
v
at
i
on o
f
t
h
e brai
n i
n
di
cat
e shift in
balance from
a focus on t
h
e internal state
of t
h
e s
u
b
j
ect
and i
t
s
rum
i
nat
i
ons
, t
o
t
h
e ext
e
rnal
e
nvi
ronment. He
nce, i
t
is possi
ble to characte
r
ise network
dy
nam
i
cs wi
t
hout
a
n
ex
pl
i
c
i
t
st
im
ul
us t
o
dr
i
v
e brai
n act
i
v
i
t
y
[7]
.
It
was al
so di
sc
ove
re
d t
h
at
t
h
e anat
om
i
c
al
stru
cture an
d
fun
c
tion
a
l conn
ectiv
ity o
f
t
h
e b
r
ai
n
m
a
tures duri
ng a
dolescence.
Altho
ugh
n
o
sign
ifican
t
diffe
re
nces ha
ve been observed
bet
w
een
t
h
e
a
dolesce
nts
and adults, s
ubtle s
p
ectral
electroence
pha
l
ogram
(EE
G
)
di
f
f
ere
n
ces exi
s
t
bet
w
e
e
n t
h
e
t
w
o
gr
o
ups
[
8
]
.
EEG is the bi
oelectrical recordin
g
of co
llectiv
e n
e
u
r
o
n
a
l activ
ity in
th
e b
r
ain
.
Th
e brain sig
n
a
l h
a
s
been act
i
v
el
y
st
udi
ed t
o
e
n
ha
nce u
nde
rst
a
n
d
i
ng o
n
t
h
e
un
d
e
rl
y
i
ng ne
ur
o
p
h
y
s
i
o
l
o
gi
cal
pr
ocesses i
n
t
h
e
brai
n.
Th
ese in
clud
e ch
aracterisatio
n of
br
ain
sig
n
a
t
u
res
d
u
ring
sleep
[9
], an
d psycho
log
i
cal con
d
ition
s
su
ch
as
schi
zo
p
h
re
ni
a,
bi
p
o
l
a
r
di
so
r
d
ers
[
10]
a
n
d
aut
i
s
m
[11]
.
R
e
searche
r
s
h
a
ve al
so
at
t
e
m
p
t
e
d t
o
un
ra
vel
ne
w
th
eories in
in
tellig
en
ce b
y
stud
yin
g
v
a
ri
o
u
s asp
ects of hu
man
cogn
itio
n
.
It h
a
s b
e
en
well
-estab
lish
e
d
that th
e
fro
n
t
al cortex
i
s
m
u
ch
related with
cog
n
itiv
e fun
c
tio
n
i
n
g
o
f
th
e b
r
ai
n
[1
2
]
. Hem
i
sp
h
e
ric sp
ecialisatio
n
of the
pre
f
r
o
nt
al
co
rt
ex
has
bee
n
o
b
ser
v
e
d
wi
t
h
t
h
e l
e
ft
hem
i
sphere
bei
n
g
i
n
v
o
l
v
e
d
wi
t
h
l
o
g
i
cal
and
se
que
nt
i
a
l
p
r
o
cesses,
wh
il
e th
e
righ
t h
e
m
i
sp
h
e
re is m
o
re i
m
m
e
rsed
in
em
o
t
io
n
a
l an
d so
cial in
teraction
cap
a
b
ilities [1
3
]
.
In ge
neral
,
t
h
e EEG
ca
n be
se
gre
g
at
ed
i
n
t
o
f
o
u
r
m
a
jor fre
q
u
ency
ba
nds
c
onsi
s
t
i
n
g of de
l
t
a
(0.
5
Hz
–
4 Hz)
,
t
h
et
a (4
Hz – 8
Hz), al
pha
(8 H
z
– 1
3
Hz) an
d be
ta (13
Hz – 30 Hz
) waves [14]. E
ach of the fre
quency
sub
-
ban
d
s
h
o
l
d
s
uni
que
i
n
fo
rm
ati
on
pert
ai
ni
n
g
di
ffe
re
nt
neu
r
op
hy
si
ol
o
g
i
cal
p
r
oces
ses
.
The
del
t
a
wa
ves a
r
e
essent
i
a
l
l
y
do
m
i
nant
i
n
dee
p
sl
eep a
nd a
r
e oft
e
n a p
r
ec
urs
o
r
f
o
r c
o
m
a
t
o
se co
n
d
i
t
i
on [
15]
. M
e
a
n
whi
l
e
, t
h
e
t
h
et
a waves a
r
e gene
ral
l
y
associ
at
ed wi
t
h
l
i
ght
sl
eep a
n
d are l
i
nke
d t
o
cre
a
t
i
v
i
t
y
and em
ot
i
on
[1
6]
. T
h
e
brai
n
in
its restin
g state is
m
a
rk
ed
b
y
in
crease in
alp
h
a
activ
ity.
Und
e
r in
ten
s
e
men
t
al activ
ity
h
o
wev
e
r, th
e
alp
h
a
wav
e
is rep
l
aced
with
th
e faster b
e
ta
rh
ythm
[1
5
]
.
In
relatio
n
with
th
e
cog
n
itiv
e processes,
it h
a
s b
e
en
rev
ealed
t
h
at th
e t
h
eta su
b-ban
d
co
n
t
ri
b
u
t
es to
wo
rk
ing
me
m
o
ry d
e
m
a
n
d
s
[17
]
.
In ad
d
ition
,
th
e t
h
eta and
l
o
we
r al
pha s
u
b
-
ban
d
s are
al
so l
i
nke
d t
o
at
t
e
nt
i
onal
re
qui
rem
e
nt
s t
h
at
dom
i
n
at
e duri
n
g enc
odi
ng
of
ne
w
in
fo
rm
atio
n
.
M
ean
wh
ile, the
up
p
e
r al
p
h
a
sub-b
a
nd
is
p
r
edomin
an
t in
sem
a
n
tic in
fo
rm
atio
n
p
r
o
cessi
n
g
[18
]
.
In orde
r
to quantify
the spectra
l inform
ation in each
band, im
pl
e
m
e
n
tation of
a
dvanced signal
pr
ocessi
ng
ap
p
r
oac
h
w
oul
d
b
e
re
qui
re
d.
A
s
suc
h
, e
v
al
uat
i
o
n
of
spect
ral
f
eat
ure ca
n
be
per
f
o
r
m
e
d usi
n
g
t
h
e
param
e
t
r
i
c
and n
o
n
-
pa
ram
e
tri
c
m
e
t
hods.
The pa
ram
e
t
r
ic t
echni
q
u
e i
s
depe
nde
nt
o
n
m
odel
-
base
d
po
we
r
spect
r
u
m
estim
at
i
on whi
c
h
i
n
cl
udes a
u
t
o
-re
gres
si
ve, m
ovi
ng a
v
era
g
e
or aut
o
-re
g
r
e
ssi
ve m
ovi
ng
avera
g
e
m
e
thods
. Mea
n
while, the
non-param
e
tric approac
h
utilis
es techni
que
s
suc
h
a
s
the
W
e
lc
h’s
m
e
thod t
o
app
r
oxi
m
a
t
e
t
h
e p
o
w
er
spect
r
u
m
of a
t
i
m
e
seque
nce.
Al
bei
t
ha
vi
n
g
i
t
s
dr
awbac
k
s
,
t
h
e
p
o
we
r s
p
ect
r
u
m
has
been
succes
sfully im
ple
m
ente
d in a
variety of E
E
G
rese
a
r
ches [19]. Suc
h
inform
a
tion is usually com
pute
d
i
n
t
o
qua
nt
i
f
i
a
bl
e desc
ri
pt
o
r
s
s
u
ch
as
ba
nd
p
o
w
er
[
18]
.
Spectral ce
ntroid
freque
ncy (
SC
F
)
i
s
a
n
est
a
bl
i
s
he
d
f
r
eq
ue
ncy
-
de
pe
nde
nt
feat
u
r
e
t
h
at
i
s
a
n
approxim
a
tion of t
h
e spectrum
’
s centre of
gra
v
ity within
each s
u
b-ba
nd. Its
advanta
g
es
are attributed
to its
ro
b
u
st
ness a
g
a
i
nst
w
h
i
t
e
Ga
ussi
an
n
o
i
s
e a
nd
re
duce
d
c
o
m
put
at
i
onal
re
qui
rem
e
nt
s. The feat
ure
has
bee
n
successfully im
ple
m
ented for speec
h re
cognition
[20], stress character
isation [21] and intel
ligence
assessm
ent
[2
2]
. He
nce,
bei
ng
rel
a
t
i
v
el
y
n
e
w, i
m
pl
em
entat
i
on o
f
s
p
ect
r
a
l
cent
r
oi
d fea
t
ures ca
n be
f
u
rt
her
extended
to characterise br
ai
n
si
g
n
at
ure
s
i
n
r
e
l
a
t
i
on wi
t
h
l
e
arni
ng
st
y
l
es.
The EE
G s
u
b
-
ban
d
fe
at
ure
s
are o
f
t
e
n
used
fo
r cl
assi
fi
cat
i
on
pu
rp
oses
u
s
i
ng t
e
c
hni
que
s suc
h
as k
-
nearest nei
g
hbour (
k
-N
N) cla
ssifier [
23]
. I
n
k
-NN, feature
s
are classified ba
sed o
n
v
o
t
i
ng c
r
i
t
e
ri
a. Th
ro
u
gh
suc
h
m
e
t
hod
, t
h
e nea
r
est
nei
g
h
b
o
u
r
feat
ure
s
fr
om
t
h
e
training set are conside
r
ed
, and the ne
w feat
ure
s
are
being as
signed to t
h
e class
of the
m
a
j
o
rity [24
]
. Vari
o
u
s
d
i
stan
ce
m
e
trics
can b
e
u
s
ed,
wit
h
th
e
Eu
clid
ean
di
st
ance
bei
n
g
am
ong t
h
e m
o
st
com
m
on.
The cl
assi
fi
cat
i
on t
ech
ni
que
has
been i
m
pl
em
ent
e
d f
o
r
v
a
ri
o
u
s
b
i
o
m
ed
ical ap
plicatio
n
s
such
as reh
a
b
ilitatio
n
[25
]
and
d
i
sease
d
e
tection
s
[23
]
.
C
u
r
r
ent
l
ear
ni
ng
st
y
l
e assessm
ent
i
nvol
ve
s t
h
e
use
of e
s
t
a
bl
i
s
hed
q
u
e
s
t
i
o
n
n
ai
res.
Su
ch t
ech
ni
q
u
e
ho
we
ver i
s
e
x
pos
ed t
o
i
n
c
o
n
s
i
s
t
e
ncy
i
ssues
whe
n
l
a
n
g
u
ag
e pr
ofi
c
i
e
ncy
b
ecom
e
s an obs
t
acl
e. Hence i
n
or
de
r
to
eli
m
in
ate such
limitatio
n
,
EEG is
propo
sed
as a v
i
ab
le
so
lu
tion
t
o
assess learn
i
ng
st
yles fro
m
th
e restin
g
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Lea
r
n
i
ng
S
t
yle Cla
ssifica
tio
n
via
EEG S
ub-ba
nd
Sp
ectra
l
C
e
n
t
ro
i
d
Frequ
e
n
cy
Fea
t
u
r
es
(
Meg
a
t S. A.
M
.
A.
)
93
3
brai
n st
at
e. Fo
l
l
o
wi
n
g
suc
h
p
r
o
p
o
si
t
i
on, t
h
e
st
udy
at
t
e
m
p
ts to characte
r
ise Kol
b
’s learning styles using the
ro
b
u
st
spect
ral
cent
r
oi
d fre
q
u
e
ncy
feat
ure
s
.
The i
n
vest
i
g
at
i
on f
o
c
u
ses o
n
l
y
on t
h
e al
pha
and t
h
et
a s
u
b
-
ban
d
s
as t
h
e i
n
her
e
nt
charact
e
r
i
s
t
i
c
s pert
ai
ni
n
g
t
o
di
ffe
re
nces i
n
at
t
e
nt
i
onal
de
m
a
nds a
n
d
o
r
g
a
ni
sat
i
on
o
f
w
o
r
k
i
n
g
me
m
o
ry exists at these
frequency ra
nges
. T
h
e
features a
r
e
then classifie
d
via
k
-NN t
o
ascertain
its
v
a
lid
ity as
a stable EE
G si
gnat
u
re
.
2.
R
E
SEARC
H M
ETHOD
Thi
s
sect
i
o
n el
abo
r
at
es e
x
t
e
n
s
i
v
el
y
on
t
h
e
m
e
t
hods
bei
n
g
im
pl
em
ent
e
d i
n
t
h
i
s
st
udy
.
It
com
p
ri
ses o
f
EEG ac
qui
si
t
i
on a
n
d i
m
pl
em
ent
a
t
i
on o
f
Kol
b
’s L
S
I
fo
r
dat
a
cl
ust
e
ri
n
g
, si
g
n
al
p
r
e-
p
r
oces
si
n
g
, e
x
t
r
act
i
on o
f
alpha and thet
a sub-ba
nd
SC
F
, rem
o
v
a
l o
f
ex
trem
e o
u
tlie
rs and
p
a
ttern
o
b
s
erv
a
tio
n, gen
e
ration
of syn
t
h
e
tic
EEG, and class
i
fication
of fea
t
ures
via
k
-N
N with
k
-f
ol
d c
r
o
ss-val
i
dat
i
o
n
.
2
.
1
.
EEG Acquisitio
n
a
n
d Ko
lb’s LSI
68
heal
t
h
y
u
n
i
versi
t
y
u
nde
r
g
ra
d
u
at
e an
d
post
g
ra
d
u
ate s
t
ude
nts (m
ale, right-hande
d
, m
ean age /
st
anda
rd
devi
at
i
on = 23
.9 /
3.
1, ra
nge = 1
8
– 3
7
y
ears) f
r
o
m
vari
ous
di
sci
p
l
i
n
es ha
ve vo
l
unt
eere
d
i
n
t
h
e EE
G
reco
rdi
n
g
.
Ap
pr
o
v
al
o
n
t
h
e
exp
e
ri
m
e
nt
al pr
ot
oc
ol
was
obt
ai
ned
fr
o
m
t
h
e uni
vers
i
t
y
’s researc
h
et
hi
cs
co
mmittee (6
00
-RMI
(5
/1
/6)). Prior to
t
h
e
record
i
n
g
sessi
o
n
, su
bj
ects
were in
itially b
r
iefed
o
n
t
h
e
ov
erall
p
r
o
c
ed
ure. All th
e
su
bj
ects
h
a
v
e
g
i
v
e
n
written
co
nsen
t.
Su
bject
s
were
req
u
i
r
e
d
t
o
sea
t
i
n
rel
a
xed
po
si
t
i
on wi
t
h
ey
e
s
cl
osed.
Ne
xt
,
EEG we
re rec
o
r
d
e
d
fr
om
t
h
e pre
f
r
o
nt
al
cort
e
x
(scal
p l
o
cat
i
o
n
s
AF
3 a
nd
AF
4)
usi
n
g
t
h
e Em
ot
i
vneu
r
o
h
ea
dset
wi
t
h
sam
p
l
i
ng rat
e
of
12
8
Hz.
A fee
d
bac
k
loop was
formed via the
P3 and P4 s
calp l
o
cations. T
h
e e
l
ectrode
place
ments conform to the
10
-
20 El
ect
r
o
d
e
Pl
acem
e
nt
Sy
st
em
of t
h
e In
t
e
rnat
i
o
nal
Fed
e
rat
i
o
n
.
Each
s
e
ssi
on
was r
e
c
o
r
d
e
d
f
o
r
du
rat
i
on
of
three m
i
nutes.
The subjects
were als
o
required to c
o
m
p
le
te the on
line
Kolb’s LSI. The
scores
obtaine
d we
re then
u
s
ed
to clu
s
ter
th
e su
bj
ects i
n
to
Div
e
rg
er,
Assi
m
ila
to
r, C
o
nv
erg
e
r and
Acco
mm
o
d
a
to
r [3
]
.
2.
2.
Si
gn
al
pre
-
Proce
ssi
ng
a
nd Fe
at
ure E
x
trac
ti
on
The reco
r
d
ed EEG
si
gnal
w
e
re
p
r
e
-
p
r
oce
s
sed of
f
line
usi
n
g MATL
AB
R2012a. Basel
i
ne correction
was accom
p
lished
using a 0.5
Hz hi
ghpass filter. Any am
pl
itudes exce
eding ±100
μ
V is assum
e
d as EOG
artefact and
he
nce re
jected [26]. So as to standa
rdis
e t
h
e si
gnal
d
u
r
at
i
on f
o
r f
u
rt
her a
n
al
y
s
i
s
, onl
y
2.
5 secon
d
s
seg
m
en
t was co
n
s
i
d
ered
[8
]. Nex
t
, th
e
p
r
e-p
r
o
cessed
EEG were filtered
in
to
alph
a an
d
th
eta wav
e
s u
s
ing
eq
u
i
ripp
le b
a
nd
p
a
ss filters [27
]
. In
o
r
d
e
r t
o
o
b
s
erv
e
th
e h
e
misp
h
e
ric correlatio
n
,
th
e stud
y also
con
s
iders bot
h
th
e left an
d ri
gh
t sid
e
of th
e prefron
t
al cortex
.
Prior to
SC
F
com
putation,
po
we
r spectral
density
(PS
D
) fo
r the res
p
ective sub
-
b
a
n
d
s was fi
rst
obt
ai
ne
d
vi
a
Wel
c
h t
e
c
hni
q
u
e
usi
n
g
Ham
m
i
ng wi
nd
o
w
wi
t
h
50%
o
v
e
r
l
a
ppi
n
g
e
p
och
s
. As
sh
o
w
n
i
n
(1
), t
h
e
su
b-b
a
nd
SC
F
is then c
o
m
puted as
the a
v
erage o
f
am
pli
t
ude wei
g
ht
ed
fre
q
u
enci
es,
di
vi
de
d by
t
h
e
t
o
t
a
l
am
pl
i
t
ude, wh
ere
N
i
s
t
h
e
n
u
m
ber of
fre
que
ncy
bi
ns,
i
i
s
t
h
e EE
G s
u
b-
ba
nd
, a
nd
S
[
f
]
w
i
[
f
] is th
e p
o
wer o
f
t
h
e
spectral
distrib
u
tion
co
rre
sp
o
ndi
ng
to
fre
q
u
e
n
cy
,
f
at
bin
i
[
20]
.
N
i
i
N
i
i
i
f
w
f
S
f
w
f
S
f
SCF
1
1
(1
)
The
SC
F
features were
t
h
en clustered
into Dive
rg
er
,
A
s
s
i
mila
to
r
,
Co
nve
r
g
er
an
d A
c
co
mmo
d
a
to
r
gr
o
ups
,
whe
r
e
si
gni
fi
ca
nt
pat
t
e
rn
i
s
o
b
se
rve
d
vi
a S
PSS
1
9
.
2.
3. Synthe
tic EEG
It has been
noted that performance of
k
-NN
classifier de
teriorates w
ith s
m
all c
l
ass sep
a
ration
an
d
une
ve
n sam
p
l
e
di
st
ri
b
u
t
i
o
n
am
ong t
h
e c
ont
rol
gr
ou
ps
[2
8]
. He
nce,
i
n
o
r
de
r t
o
m
i
nim
i
se such
effect
,
gene
rat
i
o
n of
s
y
nt
het
i
c
EE
G has bee
n
reco
m
m
e
nded.
It
i
s
im
p
o
r
tan
t
to no
te th
at
EEG i
s
sto
c
h
a
stic in
n
a
ture.
Hence
,
i
t
s
sy
nt
het
i
c
ve
rsi
o
n ca
n
be ge
n
e
rat
e
d
by
i
m
p
l
em
ent
i
ng w
h
i
t
e Gaus
si
an
n
o
i
s
e wi
t
h
s
u
f
f
i
ci
ent
l
y
con
d
i
t
i
one
d si
gnal
-
t
o
-
n
oi
se,
SN
R
ratio
to
m
a
in
tain
similar
ch
aracteristic
s. Alteratio
n
o
f
EEG ch
aracteristics is
im
m
i
nent
wi
t
h
very
l
o
w
SN
R
and t
hus
, m
a
y l
ead t
o
m
i
scl
a
ssi
fi
cat
i
on
of s
a
m
p
l
e
s. Fo
r t
h
ese reas
ons
, an
SN
R
of
3
0
dB
i
s
i
m
pl
em
ent
e
d.
Noise array,
V
noise
, is o
b
t
ai
n
e
d
b
y
m
u
ltip
ly
i
n
g th
e
no
ise
vo
ltag
e
,
V
attn
, a
n
d
w
h
i
t
e
Ga
us
si
an
noi
se,
W
noise
, where
V
attn
, is
th
e atten
u
a
ted
vo
ltag
e
deriv
e
d
fro
m
th
e
SN
R
dB
rel
a
t
i
onshi
p.
W
i
t
h
t
h
e 30 dB
SNR
, t
h
e
noi
se
po
we
r,
P
noise
was com
puted vi
a (2)
where
P
si
gnal
is the ave
r
a
g
ed power
fo
r th
e
o
r
ig
in
al EEG
,
V
EEG
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJECE Vol. 4, No. 6, D
ecem
ber 2014
:
931 – 938
93
4
SNR
P
V
signal
attn
(2
)
The synt
hetic EEG,
V
synt
,
wa
s t
h
en c
o
m
put
ed by
a
ddi
ng t
h
e ge
ner
a
t
e
d n
o
i
s
e,
V
noise
, to
th
e o
r
i
g
in
al
EEG,
V
EEG
. S
u
ch
pr
oce
d
u
r
e c
a
n
be e
x
p
r
esse
d
by
(
3
)
an
d
(
4
).
attn
noise
noise
V
W
V
(3
)
noise
EEG
synt
V
V
V
(4
)
The m
o
re det
a
i
l
e
d el
aborat
i
o
n o
n
t
h
e sy
nt
h
e
t
i
c
EEG has
been
pre
v
i
o
usl
y
repo
rt
ed [
2
9
]
. In o
r
de
r t
o
ach
iev
e
statistically sig
n
i
fican
t n
u
m
b
e
r of sa
m
p
les, syn
t
h
e
tic EEG were
g
e
n
e
rated, amo
u
n
ting
to
40
sa
m
p
les
p
e
r group
an
d
h
e
n
c
e, to
talling
to 160
sam
p
les prior to
k
-NN classification
[30].
2.
4.
k
-ne
a
rest Neighbour and
k
-fo
l
d Cro
s
s-Va
lidatio
n
k
-NN is a
s
upe
rvise
d
learning algorithm
,
where
new f
eat
ures are
classified
base
d
on elective criteri
a
.
In
itially, th
e alg
o
rith
m
sto
r
es th
e
SC
F
feat
ures
from
the
training datase
t with its associated learning style
lab
e
ls. Du
ri
n
g
testin
g
,
th
e
u
n
l
ab
elled
feat
u
r
es will b
e
classified
b
y
assign
i
n
g
the m
o
st freq
u
e
n
t
learn
i
ng
style
l
a
bel
am
ong t
hos
e o
f
k
trai
nin
g
sam
p
les nearest to
it.
In th
is st
ud
y, Euclid
ean
d
i
stan
ce is u
tilised
an
d th
e
larg
est
k
v
a
lu
e
is set at
5
.
8
0
%
o
f
th
e d
a
ta are u
s
ed
fo
r trai
n
i
n
g
, wh
ile th
e re
m
a
in
in
g
20
%
was u
s
ed
fo
r testin
g
[3
1]
.
Fo
r
bo
th
th
e train
i
n
g
and
testin
g
p
h
a
ses, accu
r
acy,
p
o
sitiv
e p
r
ed
ictiv
ity an
d
sensitiv
ity were selected
as per
f
o
r
m
a
nce i
ndi
cat
o
r
s. S
u
ch m
e
t
hod i
s
com
m
on i
n
gau
g
i
n
g t
h
e
per
f
o
r
m
a
nce of cl
ass
i
fi
er f
o
r a sel
e
c
t
ed set
of features
.
Ac
curacy,
Acc
, po
sitiv
e
p
r
ed
ictiv
ity,
Pp
, an
d sen
s
itiv
ity,
Se
,
c
a
n eac
h
be e
x
p
r
esse
d
by
(
5
)
,
(
6
)
an
d
(7
), whe
r
e
TP
is th
e tru
e
po
si
tiv
es,
TN
th
e t
r
u
e
neg
a
tiv
es,
FP
th
e
false po
sitiv
es an
d
FN
is th
e false
neg
a
tiv
e
classifications.
%
100
FN
FP
TN
TP
TN
TP
Acc
(5
)
%
100
FP
TP
TP
Pp
(6
)
%
100
FN
TP
TP
Se
(7
)
In
ord
e
r to
d
e
t
e
rm
in
e th
e tru
e
perform
ance of the classi
fier,
k
-f
ol
d c
r
oss-
v
a
l
i
d
at
i
on
was i
n
co
r
p
o
r
at
ed
with
th
e
k
-NN. The accuracy
estim
a
tes
are made by const
r
uctin
g dis
j
oi
nt training a
nd te
st sets using random
sam
p
ling
m
e
thod. T
h
e cross
-
validation
estimate of accura
cy is the overa
ll
num
ber of c
o
rrect classific
a
tions,
di
vi
de
d
by
t
h
e num
ber o
f
i
n
st
ances i
n
t
h
e
da
t
a
set
.
Hence
,
a feat
ure i
s
ass
u
m
e
d st
abl
e
fo
r
a gi
ve
n dat
a
set
and a
set
of pert
ur
ba
t
i
ons, i
f
i
t
i
n
d
u
ces t
h
e cl
assi
fi
er t
o
m
a
ke t
h
e sam
e
predi
c
t
i
ons w
h
e
n
i
t
i
s
gi
ven t
h
e
pert
ur
be
d
d
a
tasets [32
]
.
Fo
r th
e pur
po
se of
the stud
y,
th
e fo
ld
v
a
lu
e,
k
was set at
5
.
Hen
c
e at each
in
stan
ce, th
e data will b
e
ran
d
o
m
l
y
di
vi
d
e
d i
n
t
o
fi
ve se
gm
ent
s
, whe
r
e
fo
ur se
gm
ent
s
are use
d
f
o
r t
r
a
i
ni
ng
, w
h
i
l
e
t
h
e rem
a
i
n
i
ng segm
ent
is u
s
ed
for testin
g
.
Throug
h
su
ch
im
p
l
e
m
en
tatio
n
,
th
e cl
assifier will b
e
train
e
d
an
d
tested
fo
r
fiv
e
instan
ces
with
rando
m
l
y selected
train
i
ng
an
d testin
g datasets.
3.
R
E
SU
LTS AN
D ANA
LY
SIS
3.
1. Ch
ar
acter
i
sati
o
n
of
Al
p
h
a an
d
T
h
e
t
a
SCF
Sam
p
l
e
s have
been
cl
ust
e
re
d
i
n
t
o
fo
ur
l
ear
ni
n
g
style gr
ou
p
s
in
accor
d
a
n
ce
with the a
ssessm
ent via
Kol
b
’s L
S
I. The Acc
o
mmodat
o
r a
nd C
o
nve
r
ge
r group
s each c
o
nsists of
14 sam
p
les. Meanwhile, the
Dive
rge
r
a
n
d
Assim
i
lator groups
com
p
rise
s of
20
sam
p
les each. One
extrem
e outl
i
er each from
the
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Learning Style Classificati
on via
EEG Sub-band Spectral
C
e
ntroi
d
Frequency Features
(
Meg
a
t S. A.
M
.
A.
)
93
5
Assi
m
i
l
a
t
o
r
an
d
Acc
o
m
m
odat
o
r gr
ou
p was i
d
ent
i
f
i
e
d
a
n
d rem
oved. Fi
g
u
r
e
1
(
a)
a
n
d
Fi
gu
re 1(
b) sh
o
w
s
t
h
e
mean alpha a
n
d theta
SC
F
(with 95% c
o
nfidence inte
rval
)
for each learning style group.
(a)
(b
)
Figure
1. Mean (a) alpha a
n
d (b) theta
SC
F
in Ko
lb’s learn
i
ng
style group
,
N =
66
sam
p
les (o
ri
g
i
n
a
l
d
a
taset)
As o
b
se
rv
ed
fr
om
Fi
gure
1(a
)
, t
h
e C
o
n
v
e
r
gers yield the
highest m
ean alpha
SC
F
, fo
llowed
b
y
th
e
Assim
i
lato
rs an
d th
en
, th
e
Div
e
rg
ers.
T
h
e
Accom
m
odat
o
rs
on
t
h
e
ot
her
ha
nd
, at
t
a
i
n
e
d
t
h
e l
o
west
m
e
an
fo
r
the alpha
SC
F
. Variation
s
in
th
e alp
h
a
SCF
i
s
at
t
r
i
but
ed t
o
t
h
e di
f
f
e
rent
ap
pr
oac
h
es bei
ng a
d
o
p
t
e
d i
n
i
n
f
o
rm
at
i
on pr
ocessi
n
g
,
where a high alpha
SC
F
wou
l
d
ind
i
cate state o
f
sem
a
n
tic
in
formatio
n
wh
ile
a lo
w
al
pha
SC
F
si
gnifies a state e
n
coding
of
ne
w
inform
ation. Balanced m
ean alpha
SC
F
between t
h
e left
and
ri
g
h
t
hem
i
sphe
re
has
been
o
b
s
e
rve
d
fo
r al
l
l
e
arni
ng
st
y
l
e gr
ou
ps.
Meanwhile, Figure 1(b) re
veal
ed
th
at
Acco
mm
o
d
a
to
rs attain
ed the
highest m
ean theta
SC
F
,
fol
l
o
we
d by
t
h
e
Di
ver
g
e
r
s
and
C
o
nve
r
g
er
s. M
ean
w
h
ile, the
Assim
i
lators
yielded t
h
e lowest t
h
eta
SC
F
.
Variation
s
of
SCF
bet
w
ee
n t
h
e l
e
ft
and
ri
g
h
t
hem
i
sphere w
a
s not
a
b
l
e
f
o
r t
h
e t
h
et
a ba
n
d
.
C
o
m
p
arat
i
v
el
y
,
t
h
e
Accom
m
odat
o
rs an
d C
o
nve
r
g
ers
di
s
p
l
a
y
e
d
hi
g
h
er
SC
F
f
o
r t
h
e ri
ght
he
m
i
sphere,
w
h
i
l
e
b
o
t
h
Di
ve
rg
ers a
n
d
Assim
i
lato
rs attain
ed
h
i
g
h
e
r
SCF
for th
e left
sid
e
of th
e prefron
t
al co
rtex
.
Su
ch
find
ing
s
can
b
e
related
to
th
e
di
ffe
re
nt
at
t
e
nt
i
onal
re
qui
rem
e
nt
s an
d st
rat
e
gi
es i
n
w
o
r
k
i
n
g m
e
m
o
ry
org
a
ni
sat
i
on
bet
w
een t
h
e l
e
ft
an
d ri
g
h
t
hem
i
sphere.
3.
2. Synthe
tic EEG
Each
o
f
t
h
e
l
e
a
r
ni
ng
st
y
l
e g
r
o
ups
were
en
ha
nced
wi
t
h
sy
nt
het
i
c
EE
G,
am
ou
nt
i
n
g t
o
40
s
a
m
p
l
e
s pe
r
gr
o
u
p
.
The sy
nt
het
i
c
EEG u
nde
r
w
ent
si
m
i
l
a
r si
gnal
p
r
e
p
r
o
cessi
ng a
n
d feat
u
r
e ext
r
act
i
on m
e
t
h
o
d
as t
h
e
ori
g
i
n
al
dat
a
se
t
.
As sh
ow
n i
n
Fi
gure
2(a
)
an
d Fi
g
u
re 2
(
b),
sim
i
l
a
r pat
t
e
rn of m
ean al
pha and t
h
et
a
SC
F
(with
9
5
% co
nfid
en
ce in
terv
al)
h
a
s
b
een ob
serv
ed
with
th
e enh
a
nced
d
a
taset.
(a)
(b
)
Figure
2. Mean (a) alpha a
n
d (b) theta
SC
F
in Ko
lb’
s
learn
i
ng
style gr
oup
,
N
=
16
0 sam
p
les (o
r
i
g
i
n
a
l
d
a
taset)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJECE Vol. 4, No. 6, D
ecem
ber 2014
:
931 – 938
93
6
3.
3.
k
-ne
a
rest Neighbour and
k
-fo
l
d Cro
s
s-Va
lidatio
n
The fi
ve-
f
ol
d m
ean t
r
ai
ni
n
g
and testing acc
uracies for
k
= 1 to
k
= 5 i
s
s
h
o
w
n i
n
Fi
g
u
r
e 3. T
h
e best
accuracies
we
re obtaine
d
at
k
=
2
,
w
ith
t
h
e tr
ai
n
i
ng
and
testing
yieldin
g
10
0% an
d 97
.5
% accuracies,
respectively. As
k
increas
es,
bot
h the t
r
aini
ng a
n
d testing
accuracy
decre
a
ses. T
h
is is mainly contri
but
ed by
the fact
that t
h
e classification technique
wo
r
k
s base
d o
n
vo
t
i
ng
c
r
i
t
e
ri
a. W
i
t
h
i
n
creasi
n
g
k
, inte
rfe
renc
e f
r
om
o
t
h
e
r
n
e
i
g
hbou
r
i
n
g
bu
t d
i
ff
er
en
tly lab
e
lled
f
eat
u
r
es
w
o
u
l
d
b
e
i
n
tr
odu
ced
and
h
e
n
c
e, aff
ectin
g
t
h
e
classification a
ccuracies.
Fi
gu
re
3.
Fi
ve
-
f
ol
d m
ean t
r
ai
n
i
ng a
n
d t
e
st
i
n
g
classification a
ccuracies
for alpha
and
t
h
eta
SCF
features
Positive pre
d
ictivity and sensitivity
for each learning
style group for
k
= 2 are as s
h
own in Table 1.
Ove
r
al
l
,
fi
ndi
n
g
s i
n
di
cat
e t
h
at
t
h
e fi
ve-
f
ol
d
m
ean i
ndi
ces
y
i
el
ded
rel
i
a
bl
e
resul
t
s
.
D
u
ri
ng
t
r
ai
ni
n
g
, al
l
l
e
arni
ng
style groups at
tained
perfect
positive
pre
d
ictivity and se
nsitivity. In the
testing sta
g
e
howe
ve
r, acce
pt
able
p
e
rform
a
n
ce hav
e
b
e
en
o
b
s
erv
e
d
.
Th
e
Acco
mm
o
d
a
to
rs
yield
e
d
100
% po
sitiv
e pred
ictiv
ity an
d
sen
s
itiv
ity.
Mean
wh
ile, the Div
e
rg
er
grou
p attain
ed th
e lo
west po
sitive pred
ictiv
ity an
d sen
s
itiv
ity at 9
5
.6% an
d
9
5
.0%,
respectively.
Tab
l
e
1
.
Fiv
e
-Fo
l
d Mean Po
sitiv
e Pred
ictiv
ity an
d
Sensitiv
ity Measu
r
es
fo
r Classificatio
n
at
k
= 2
L
ear
ning Sty
l
es
T
r
aining T
e
sting
Pp (%)
Se (%
)
Pp (%)
Se (%
)
Diver
g
er 100
100
95.
6
95.
0
Assi
m
ilator 100
100
97.
1
95.
0
Conver
g
er
100
100
97.
8
100
Acco
m
m
odator 100
100
100
100
The res
u
lts are deem
ed reliab
l
e since the
k
-NN classifier
has bee
n
incorporated wit
h
k
-f
ol
d cr
oss
-
v
a
lid
ation
.
Th
e con
s
isten
c
y
of th
e
features i
n
classifyin
g t
h
e l
ear
ni
n
g
st
y
l
es were
t
e
st
ed
wi
t
h
fi
ve
ra
nd
om
ly
assigne
d traini
ng a
nd testing datasets
that on ave
r
age re
sul
t
ed in excellent
perform
a
nce in term
s
of accuracy
,
p
o
s
itiv
e pred
ictiv
ity an
d
sen
s
i
tiv
ity.
4.
CO
NCL
USI
O
N
Fi
ndi
ng
s have
pr
o
v
en t
h
at
res
t
i
ng EEG f
r
om
t
h
e pref
ro
nt
al
cort
e
x
cont
ai
ns
brai
n si
g
n
at
u
r
es t
h
at
can
be rel
a
t
e
d wi
t
h
l
earni
n
g
st
y
l
es. The st
u
d
y
h
a
s pr
ovi
ded
a first-h
a
nd
in
sigh
t in
to
th
e ch
aracterisatio
n
o
f
alp
h
a
and t
h
eta s
u
b-band
SC
F
,
wher
e d
i
stin
ct
d
i
ff
er
en
ces can
be o
b
serv
ed
b
e
t
w
een
t
h
e learnin
g
style gr
oups. Su
ch
find
ing
s
are at
trib
u
t
ed
to
th
e v
a
riatio
ns in
atten
ti
onal
re
q
u
i
r
em
ent
s
, i
n
f
o
rm
at
i
on pr
oc
essi
ng st
rat
e
gi
es and
or
ga
ni
sat
i
on
o
f
w
o
r
k
i
n
g m
e
m
o
ry
d
u
ri
n
g
t
h
e
rest
i
n
g
st
at
e.
Classification of the alpha and theta sub-band
SC
F
vi
a
k
-NN techni
que
has attained excellent
accuracy,
positive predictivity and sens
itivity for all learning style groups
. Hence, t
h
e results support the
in
itial o
b
s
erv
a
t
i
o
n
s
wh
ich indicate alp
h
a
and th
eta sub
-
b
a
nd
SC
F
a
s
distinctive and
stabl
e
EE
G si
gnat
u
res
for
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Learning Style Classificati
on via
EEG Sub-band Spectral
C
e
ntroi
d
Frequency Features
(
Meg
a
t S. A.
M
.
A.
)
93
7
id
en
tifying
indiv
i
d
u
a
l
learn
i
ng
styles. Reliab
ility o
f
t
h
e
featu
r
es was al
so
con
f
irm
e
d
v
i
a th
e
k
-
f
ol
d cross
-
v
a
lid
ation
.
Fu
ture work
will fo
cus on
m
o
d
e
llin
g
o
f
t
h
e restin
g
EEG in
relation
with
Ko
lb’s learn
i
n
g
styles.
Co
m
p
arativ
e stu
d
y
b
e
tween
differen
t
m
o
d
e
llin
g
techn
i
qu
es
will also
b
e
perfo
r
m
e
d
so
as to
p
r
ov
id
e th
e
m
o
st
feasib
le so
lu
tion
for a
real-time EEG-b
as
ed learni
ng style assessm
ent syste
m
.
ACKNOWLE
DGE
M
ENTS
Au
t
h
ors
wou
l
d lik
e to
th
ank
Un
i
v
ersitiTekno
log
i
MARA
an
d
t
h
e Min
i
st
ry o
f
Edu
cation
,
Malaysia
fo
r t
h
e
s
u
p
p
o
rt
t
h
i
s
st
u
d
y
t
h
r
o
ug
h t
h
e F
u
nda
m
e
nt
al
R
e
search
Gra
n
t
Sc
he
m
e
(6
00
-R
M
I
/
F
R
G
S
5/
3
(
7
2/
20
1
2
)
)
,
as well as
SL
AB and MyPhD
scholars
hips.
REFERE
NC
ES
[1]
S. Cassid
y
, "Learning sty
l
es: An overvie
w of theories, mode
ls
,
and m
eas
ures
",
Educational Ps
ycholog
y,
vol. 24
,
pp. 419-444
, 20
04.
[2]
A.
Y.
Kolb and D.
A. Kolb, "Experiential
Learning
Theor
y
:
A D
y
namic, Ho
listic Approach
to Man
a
gemen
t
Learn
i
ng, Education and
Development",
in
T
h
e SAGE Hand
book of Manag
ement Learning
, Educa
tion an
d
Developmen
t
, S. J. Armstrong an
d C. V
.
Fukami,
Eds., L
ondon: SAGE Publicatio
ns
Ltd, 2009, pp. 42-69.
[3]
A.
Y.
Kolb and D.
A.
Kolb,
The Kolb Learn
i
ng Style Inven
tory–V
ersion 3.1
2005 Technica
l Specifica
tions
.
Massachusetts: Hay
Re
source D
i
rect, 2005.
[4]
S
.
J
o
y
and D. A. Kolb, "Are the
r
e cultur
a
l diff
er
ences
in le
arnin
g
s
t
y
l
e
?
",
International Journal
of Intercultural
Relations,
vo
l. 3
3
, pp
. 69-85
, 20
09.
[5]
R. Dunn, "Understanding th
e Dunn and D
unn learning sty
l
es
model and the
n
eed for ind
i
vid
u
al diagnosis an
d
prescription",
In
t
e
rnational
Journ
a
l of
Read
ing,
W
r
iting, and Lear
ning Disabili
ti
es,
vol. 6
,
pp
. 223-
247, 1990
.
[6]
N. J
a
uš
ovec and
K. J
a
uš
ovec, "
R
es
ting brain
ac
tivit
y:
Differ
e
nc
es
between g
e
n
d
ers
"
,
Neuropsychologia
,
vol. 4
8
,
pp. 3918-3925
,
2010.
[7]
G.
Deco,
V.
K.
Jirsa,
and A.
R.
McIntosh,
"Resting br
ains never rest: Co
mputati
onal insights into potential
cognitiv
e arch
ite
ctures",
T
r
ends
i
n
Neur
os
cien
ces
,
vol. 36
, pp
. 268
-274, 2013
.
[8]
R. Lüch
inger,
L
.
M
i
che
l
s
,
E. M
a
r
tin
, and D. Br
andeis, "EEG–BOLD correla
tion
s
during (post-)adolescent brain
m
a
turation",
NeuroImage,
vol. 5
6
, pp
. 1493-150
5, 2011
.
[9]
V. Brodbeck, A
.
Kuhn, F. von
Wegner, A. Mo
r
zelewski, E.
Tag
lia
zucch
i, S. Bor
i
sov
, et al.
, "EEG microstates
of
wakefulnes
s
and
NREM
s
l
eep"
,
NeuroImage,
vo
l. 62
, pp
. 2129-2
139, 2012
.
[10]
J. W. Y. Kam, A. R.
Bolbecker
,
B. F. O'
Donnell, W.
P. Hetrick,
and C. A. Br
enn
e
r, "Resting state EEG power
an
d
coheren
ce
abnor
m
a
lities in b
i
pol
ar disorder and s
c
hizophr
enia"
,
Journal of Psychiatric Res
e
arch,
vol. 47, pp
. 1893
-
1901, 2013
.
[11]
K
.
J
.
M
a
t
h
e
w
s
o
n
,
M
.
K
.
J
e
t
h
a
,
I
.
E
.
D
r
m
i
c
,
S
.
E
.
B
r
y
s
on, J. O
.
G
o
ldberg, and
L.
A.
Schm
idt, "R
e
g
ional
EEG
alph
a
power, coher
e
nce, and behav
i
or
al s
y
mpto
m
a
tolo
g
y
in autism
spectrum
disorder
",
Clinica
l
Neurophysiology,
vol.
123, pp
. 1798-1
809, 2012
.
[12]
J. R. Gr
ay
and P
.
M. Thom
pson,
"Neurobiolog
y
o
f
intell
igenc
e
:
S
c
ienc
e and
e
t
hics
"
,
Natur
e
R
evi
ew
s
Neur
os
cienc
e
,
vol. 5
,
pp
. 471-4
82, 2004
.
[13]
L. J. van der Knaap and I. J. M. va
n der Ham
,
"How does
the corpus
callos
u
m
m
e
diat
e interh
em
is
pheric tr
ans
f
er
?
A
review",
Beha
vioural Brain
Res
e
arch,
vo
l. 223,
pp. 211-221
, 20
11.
[14]
B. P. Harn
e, "H
iguchi fr
ac
tal dimension analy
s
is of EEG signal before
and after OM chanting
to observe over
a
ll
effect on
brain",
International Jo
urnal of
Electr
ical and Computer Engin
eering
,
vo
l. 4
,
pp
. 585-592
, 2014
.
[15]
J. G. Webster,
M
e
dica
l Instrumen
t
ation:
Application and Design
,
4th ed
. New
Jersey
: Wiley
,
2009.
[16]
T. L. Huang
an
d C. Char
y
t
on, "
A
comprehensive review
of the
ps
y
c
holog
ical ef
fects of brainwave entrainment",
Alternative Ther
apies in
Health
&
M
e
dicine,
vol. 14, pp. 38-49, 2
008.
[17]
W. Klimesch, "EEG alpha and
th
eta oscillations reflect cognitive
and memo
ry
per
f
ormance: a review and analy
s
is",
Brain Research
Reviews,
vol. 29, pp. 169-195, 19
99.
[18]
M. Doppelm
ay
r
,
W. Klim
esch, W. St
adler, D. Pöllhuber, and
C. Heine,
"E
E
G
alpha power and intellig
ence",
Intell
igen
ce
,
vol. 30, pp. 289-302
, 2002
.
[19]
S. Motam
e
di-Fa
khr, M. Moshrefi-Torba
ti, M
.
Hill, C
.
M. Hil
l
,
and P. R.
W
h
it
e, "Signal pro
c
e
ssing techniques
applied to human sleep EEG sig
n
als—A review"
,
Biomed
ical S
i
g
nal Processing
and Control,
vol. 10, pp. 21-33
,
2014.
[20]
B. Gajic and K.
K. Paliwal, "Ro
bust speech reco
gnition
in nois
y
environm
ents ba
sed on subband
spectra
l cent
r
oid
histograms",
IEEE Transactions
on Aud
i
o, Spe
ech, and
Language Processing,
vol. 14, pp. 600-608
, 2006
.
[21]
N. Sulaiman, M. N. Taib, S.
Lias, Z. Murat, S.
A. Mat
Aris, an
d N. H. Abdul
Hamid,
"EEG-b
ased stress featu
r
es
using spectral
centro
i
ds technique and
k-near
est neighbor classifier", in
Pr
oceed
ings of 2011 UKSim 13th
International Co
nference on
Co
mputer Modellin
g and Simula
tio
n
, 2011
, pp
. 69-
74.
[22]
A
.
H
.
J
a
h
i
d
i
n
,
M
.
N
.
T
a
i
b
,
M
.
S
.
A
.
M
e
g
a
t
A
l
i
,
N
.
M
d
T
a
h
i
r
,
S
.
L
i
a
s
,
M
.
H
.
H
a
r
o
n
,
R
.
M
.
I
s
a
,
W
.
R
.
W
.
O
m
a
r
,
a
n
d
N. Fuad, "Evalu
ation of br
ainwave sub-band sp
e
c
tra
l
cen
troid in
hum
an intell
igen
ce", in
Proceedings of 2013 IEEE
9th International Colloquium
on
Signal
Processin
g
and its App
lica
tions
, 2013
, pp
.
295-298.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJECE Vol. 4, No. 6, D
ecem
ber 2014
:
931 – 938
93
8
[23]
W. A. Chaovalitwongse,
F. Ya-
J
u, and R. C
.
Sachdeo
,
"On the time
series k-n
earest n
e
ighbor
classification of
abnormal brain
activity
"
,
I
E
EE Transactions on
Systems, Man and Cy
bernetics,
Part A: Systems
and Humans,
vol.
37, pp
. 1005-10
16, 2007
.
[24]
S. Zokaee and
K. Faez, "Human iden
tification
based on ECG and palmprint",
I
n
ternational
Jou
r
nal of Electrica
l
and Computer Engineering
,
vol.
2, pp
. 261-266
,
2012.
[25]
A. R. Gur
a
liu
c,
P
.
Bars
occh
i,
F
.
P
o
torti
,
and
P
.
Nepa, "
L
im
b m
ovem
e
nts
cl
as
s
i
fica
tion us
ing
wearab
le wi
rel
e
s
s
transce
i
vers",
IEEE Transactions
on Informatio
n
Technology in
Biomedicine,
vol.
15, pp
. 474-480
, 2011.
[26]
A. Schlögl, C. Keinrath, D
.
Zimmermann,
R. Sc
herer,
R.
Le
eb,
a
nd G. Pfurtsche
l
ler,
"A full
y
aut
o
m
a
ted corr
ec
tio
n
method
of EOG
artif
acts in EEG recordings",
Clinical Neurophys
i
ology,
vol. 118
,
pp. 98-104
, 200
7.
[27]
M.
S.
A.
Megat
Ali,
M.
N.
Taib, N.
Md Tahir,
A.
H.
Jahidin
,
and
I. M. Yassin, "
E
EG sub-band spectr
a
l centro
id
frequencies extr
action based on
Hamming and
equiripple filters
: A comparativ
e stud
y
"
, in
Proceedings of 2014
IEEE 10th
Inter
national Co
lloqu
ium on Signa
l
Processing and
its
Applica
tions
, 20
14, pp
. 199-203
.
[28]
J. E. Goin
, "Classification
bias
of
the k-nearest
neighbor algorithm",
IEEE Transactions on Pattern Analysis an
d
Machine
Int
e
ll
ig
ence
,
vo
l. PAMI-6, pp
. 379-381
,
1984.
[29]
A
.
H
.
J
a
h
i
d
i
n
,
M
.
S
.
A
.
M
e
g
a
t
A
l
i
,
M
.
N
.
T
a
i
b
,
N
.
M
d
T
a
h
i
r
,
I
.
M
.
Y
a
s
s
i
n
,
a
n
d
S
.
L
i
a
s
,
"
C
l
a
s
s
i
f
i
c
a
t
i
o
n
o
f
intelligen
ce quotient
via brainw
ave sub-band po
wer
ratio f
eatur
es and ar
tificial n
e
ural network",
Computer Metho
d
s
and Programs in Biomed
icin
e,
vo
l. 114
, pp
. 50-59
, 2014
.
[30]
A.
Field,
Disco
v
ering Sta
tistics
Using SPSS
, 3rd
ed. London
: Sage
Publications,
2009.
[31]
K. Polat, B. Ak
de
mir, and S. Güne
ş
, "Computer aided diagnosis of ECG da
ta on the least squar
e
support vecto
r
m
achine",
Digital Signal Processing,
vol. 18
, pp
. 25-32, 2008.
[32]
R. Kohavi, "A st
ud
y
of cross-vali
dation and boots
t
ra
p for accur
a
c
y
estim
ation and m
odel select
ion", in
Pr
oceed
ings
of 1995 In
ternational Jo
int Con
f
er
ence on
Artifici
al Intelligence
, Montreal,
1995
, pp.
1137-1145
.
BIOGRAP
HI
ES OF
AUTH
ORS
Megat S
y
ahirul
Am
in Megat Ali rec
e
ived
th
e B.Eng (Biom
e
dic
a
l) from
Universiti Ma
la
ya
,
Malay
s
ia, and
M.Sc. (Biomedical
Engin
eering
)
from University of Surrey
,
United Kingdom. He
is currentl
y
a senior lecturer at
the Facult
y
of
Elect
rical Engi
neering
,
Univer
siti Teknolog
i
MARA, Mala
y
s
ia. His resear
c
h
interests inc
l
ude EEG and
intel
ligen
t m
odelling of bra
i
n
behavior
with
ap
plic
ation
to
expe
rienti
al
le
arning
theor
y
.
Ais
y
ah Har
tini Jahidin ob
tain
ed the B.Eng (Tel
ecommunication) and M.Eng.
S
c
(El
ectr
i
ca
l) from
Universiti Mal
a
ya
, Mal
a
y
s
ia
. She is current
l
y
a postgraduat
e
research
er at
t
h
e Facult
y of
Ele
c
tri
cal
Engin
eering
,
Universi
tiTekno
logi M
A
RA, Mala
y
s
i
a
. Her m
a
in res
earch
inter
e
sts
includ
e hum
an intel
ligen
ce, EE
G and non-linear m
odelling of
brain behav
i
or
via intellig
ent
signal pro
cessin
g
technique.
Nooritawati Md
Tahir received
the B
.
Eng (
E
l
ectronics) from th
e Universiti
Tek
nologi MARA,
Malay
s
ia, M.Sc. (Micro
electr
o
n
ics & Teleco
mmunications) fro
m University
of Liv
e
rpool,
United Kingdo
m, and Ph.D. in Electrical E
ngineer
ing (Pattern Recognition
& Artificial
Intell
igen
ce) fro
m
Universiti Kebangsaan Mal
a
ysia
, Mala
ysi
a
.
She is currentl
y
an Associate
Professor at the Faculty
of Electrical
Engine
er
ing and the Dir
ector of R
e
sear
ch Innovation
Business Unit, Universiti T
e
kn
ologi MARA, Mala
y
s
ia
. Her r
e
search
inter
e
sts include
im
age
processing,
pat
t
e
r
n recogn
ition
,
c
o
m
puter vision
a
nd art
i
fic
i
a
l
in
tel
ligen
ce.
Mohd Nasir Taibobtain
e
d th
e B
.
Eng
(
E
lectr
i
cal)
from the Univ
ersity
of
Tasmania, Australia,
M.Sc. (Control
S
y
stems) from
University
of She
ffield,
and Ph.D. (Control
& I
n
strumentation
)
from
Universit
y
of Manch
e
ster
Institute of
Science
and T
echno
log
y
, Uni
t
ed Ki
ngdom
. He is
current
l
y
a Profe
ssor and the Dea
n
of the Facu
lt
y
of Ele
c
tri
c
a
l
Eng
i
neer
ing, Univ
er
siti Tekno
logi
MARA, Malay
s
ia. He is leading
an active r
e
search group and supe
rvising a pool
of research
ers
in advan
ced sig
n
al processing with application
s
in control s
y
s
t
ems and process, biomedical
engineering, and
nonlin
ear
s
y
stems.
Evaluation Warning : The document was created with Spire.PDF for Python.