Indonesi
an
Journa
l
of El
ect
ri
cal Engineer
ing
an
d
Comp
ut
er
Scie
nce
Vo
l.
1
3
,
No.
3
,
Ma
rch
201
9
, p
p.
999~
1006
IS
S
N: 25
02
-
4752, DO
I: 10
.11
591/ijeecs
.v1
3
.i
3
.pp
999
-
10
06
999
Journ
al h
om
e
page
:
http:
//
ia
es
core.c
om/j
ourn
als/i
ndex.
ph
p/ij
eecs
Sn
ake s
p
ec
i
es
id
en
tificatio
n by usi
ng natu
ral l
angu
age
processi
ng
Nu
r Li
yana Iz
z
at
i Rusli
1
, A
mi
z
a
Amir
2
, N
ik Ad
il
ah H
ani
n Z
ah
ri
3
, R
.
Badli
shah
Ahm
ad
4
1,2,3
School
of
Co
m
pute
r
and
Com
m
unic
at
ion
Enginee
ring
,
Univ
ersit
i
M
al
a
y
si
a
Per
l
is,
Mal
a
y
s
ia
4
Facul
t
y
of
Infor
m
at
ic
s a
nd
Com
puti
ng,
Univer
sit
i
Sulta
n
Z
ai
n
al
Abidin
(UniSZA
),
22200
Besut,
Te
ren
gg
anu
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Oct
1,
2018
Re
vised Dec
10, 2
018
Accepte
d Dec
25, 201
8
The
pape
r
pre
se
nts
the
snake
sp
ec
i
es
ide
nti
f
ic
a
tion
b
y
using
nat
ura
l
la
ngua
g
e
proc
essing.
I
t
aim
s
to
hel
p
m
edi
cal
profe
ss
ion
al
s
in
pre
d
ictin
g
the
snak
e
spec
ie
s
for
snak
e
-
bite
tr
ea
tment
s
base
d
on
the
pat
i
ent
’s
de
scr
ip
ti
on
of
th
e
snake
.
The
de
ci
s
ion
in
suita
ble
a
nti
-
venom
cri
tic
al
l
y
dep
ends
on
the
t
y
pe
of
snake
spec
ie
s.
W
rong
ant
i
-
ven
om
m
ay
resul
t
in
seve
re
m
o
rbidi
t
y
and
m
orta
li
t
y
.
Thi
s
rese
arc
h
inv
esti
g
at
es
the
hum
an
per
ce
p
ti
on
and
t
he
sele
c
ti
o
n
of
words
in
desc
ri
bing
a
snake
base
d
on
th
ei
r
v
isual
vi
ew.
Th
e
desc
riptions
were
pre
sente
d
in
unstruct
ure
d
te
xt
,
and
the
NLP
proc
essing
invol
ves
pre
-
proc
essing,
feat
ure
ext
ra
ct
ion
and
cl
assifi
ca
t
i
on.
Four
m
ac
hine
le
arn
ing
al
gorit
hm
s
(na
ïv
e
Ba
y
es
,
k
-
Nea
r
est
Neighbour
,
Suppor
t
Vec
tor
Mac
hin
e,
an
d
Dec
ision
Trees
J48)
were
used
during
tra
ini
ng
an
d
cl
assificat
ion
.
Our
result
s
show
tha
t
J48
al
gorit
hm
obtai
ned
the
high
est
cl
assificat
ion
ac
cur
acy
of
71.
6%
cor
re
ct
pr
edi
c
ti
on
for
the
NLP
-
Snake
dat
a
set
with
high
pre
ci
sion
and
rec
a
ll
.
Ke
yw
or
ds:
N
at
ural
lan
gu
a
ge
Hu
m
an
pe
rce
ption
Sn
a
ke
im
ages
TF
-
IDF
Copyright
©
201
9
Instit
ut
e
o
f Ad
vanc
ed
Engi
n
ee
r
ing
and
S
cienc
e
.
Al
l
rights re
serv
ed.
Corres
pond
in
g
Aut
h
or
:
Am
iz
a A
m
ir,
School
of Com
pu
te
r
a
nd Com
m
un
ic
at
ion
Enginee
rin
g,
Pauh
P
utra
Cam
pu
s,
U
niv
e
rsi
ti
Mal
ay
sia
Per
li
s,
02600 A
ra
u,
M
al
ay
sia
Em
a
il
: a
m
iz
aa
m
ir@u
nim
ap.
e
du.m
y
1.
INTROD
U
CTION
Sn
a
kes
t
hat
ar
e
cold
-
blood
e
d
ve
rtebr
at
es
fa
ll
s
into
tw
o
ca
te
gories;
ve
no
m
ou
s
an
d
non
-
ve
nom
ou
s.
Ma
ny
ve
no
m
ou
s
s
na
kes
hav
e
ap
pear
e
d
in
m
any
co
untrie
s,
and
they
are
a
real
threat
t
o
t
he
public
safet
y
an
d
healt
h.
T
he
re
a
re
m
or
e
than
3000
s
pecies
of
the
s
nak
e
n
owadays
,
60
0
of
them
are
ve
nom
ou
s,
a
nd
ov
er
20
0
are
co
ns
i
der
e
d
i
m
po
rtant
i
n
m
edical
reco
r
d
[1
]
.
Highest
m
edical
i
m
po
rtant
treat
m
ents
are
s
nak
e
bites
from
a
highly
venom
ou
s
sn
a
ke
that
a
re
necessa
ry
to
be
recog
nized
since
they
can
cause
seve
re
pain
a
nd
e
ven
death
(e.
g.
,
Bl
ac
k
M
a
m
ba,
Ki
ng
C
obra,
I
nd
ia
n
Kr
ai
t).
Sec
onda
ry
m
edical
i
m
po
rtant
sn
a
ke
bites
are
du
e
to
the
venom
ou
s
s
na
kes
(e
.g.,
Al
bin
o
B
ur
m
ese
Pyt
hon,
Ba
ll
Pyt
ho
n,
Re
d
Ra
t
Snake)
that
ca
n
r
esult
in
disa
bili
ty
and
sever
e
pai
n
bu
t
in
le
ss
i
m
pacted
due
to
t
heir
act
ivit
y
or
m
ay
be
beca
us
e
of
t
he
ha
bitat
that
near
of
hum
an
popula
ti
on
.
In
Ma
la
ysi
a,
the
five
-
ye
ar
re
vie
w
of
sn
a
keb
it
e
patie
nts
show
s
that
there
we
re
260
cases
of
sn
ake
bites re
ported
, a
nd 52.9%
of t
he
s
na
ke bit
es w
ere
fro
m
u
nk
now
n
[
2].
In
m
any
e
m
er
ge
ncy
cases,
one
has
to
identif
y
the
sn
ake
sp
e
ci
es
m
erely
based
on
the
te
xt
descr
i
ption
giv
e
n
to
them
by
the
victi
m
or
witness
without
any
gr
a
phic
al
ai
ds
.
Be
ing
able
to
rec
ognize
the
ty
pe
of
sn
a
ke
base
d
on
the
descr
i
ption
of
the
pe
op
le
ha
ve
bec
om
e
ver
y
im
p
or
ta
nt
in
s
ocial
an
d
m
edical
progr
ession.
To
pe
rfor
m
optim
a
l
cl
inica
l
t
reatm
ent,
the
diag
nosis
of
th
e
sn
ake
s
pecie
s
respo
ns
ible
f
or
the
s
nake
bi
te
is
cru
ci
al
.
The
sl
igh
te
st
delay
m
igh
t
giv
e
res
ult
in
sever
e
m
or
bid
it
y
and
m
or
ta
li
ty
.
Thu
s,
it
is
i
m
per
at
ive
t
o
pr
eci
sel
y
and
c
on
ci
sel
y
deter
m
ine
the
ty
pe
or
s
pecies
of
th
e
sn
akes
.
The
colle
ct
ed
inf
orm
at
ion
is
i
m
po
rtant
to
identify
if
the s
nak
e
is venom
ou
s
o
r
not,
t
hus
hel
ps
m
edical
prof
e
ssio
nal
t
o
determ
ine
th
e
su
it
able anti
-
venom
and f
ur
t
her tre
atm
ent p
la
n.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
1
3
, N
o.
3
,
Ma
rc
h
201
9
:
999
–
1006
1000
Ty
pi
cal
ly
,
sn
ake
sp
eci
es
a
re
recog
nized
m
a
nu
al
ly
base
d
on
the
vis
ual
fe
at
ur
es
s
uch
as
head
s
ha
pe
,
sk
in
col
or,
ey
e
sh
ape
,
and
body
sh
a
pe.
T
his
process
re
qu
i
res
knowle
dg
e
of
cha
ract
erist
ic
s
of
the
sn
akes
wh
ic
h
is
no
t
quit
e
com
m
on
fo
r
m
os
t
peo
ple
wh
er
e
on
ly
th
e
e
xp
e
rts
hav
e
this
us
ef
ul
knowle
dge.
Co
ns
i
der
i
ng
the
dif
ficult
y
f
aced
by
m
os
t
people
in
id
ent
ify
ing
the
s
na
ke
sp
eci
es,
the
m
ai
n
aim
of
this
work
is
t
o
pe
rfor
m
sp
eci
es
recog
niti
on
base
d on the
descr
i
ptio
n from
the w
it
ne
ss or
victim
in
un
st
ru
ct
ur
e
d
te
xt for
m
.
In
this
w
ork,
a
n
i
ntell
igent
sy
stem
that
will
help
the
m
edical
pro
fessio
nal
to
be
able
t
o
pr
e
dict
the
ty
pe
of
t
he
s
na
ke
ba
sed
on
the
desc
riptio
n
in
the
un
st
ruct
ur
e
d
te
xt
by
us
in
g
natu
ral
la
ngua
ge
pro
cessi
ng
(N
L
P)
.
C
ommon
per
ce
ptio
n
and
wor
ds
us
e
d
by
m
any
diff
ere
nt
pe
op
le
to
desc
ribe
m
a
ny
diff
e
re
nt
typ
es
of
sn
a
kes
will
be
analy
zed.
T
he
te
xt
will
be
prep
ro
ce
ssed
,
re
le
van
t
ke
ywo
r
ds
will
be
e
xtr
act
ed
base
d
on
thei
r
weig
ht
in
the
c
on
te
xt,
a
nd
the
se
key
words
w
il
l
be
use
d
as
f
eat
ur
es
duri
ng
cl
assifi
cat
ion
by
m
ach
ine
le
ar
ning
to lear
n
a
nd pr
edict
the s
peci
es of s
na
kes.
Li
m
it
ed
stud
ie
s
ha
ve
bee
n
c
onduct
ed
f
or
s
pecies
recog
niti
on
by
us
i
ng
m
achine
le
ar
ni
ng.
B
utterfly
sp
eci
es
recog
ni
ti
on
in
[
3]
us
e
s
ne
ur
al
net
work
s
to
rec
ogniz
e
butt
erf
ly
s
pe
ci
es
base
d
on
butt
erf
li
es’
s
ha
pe
.
T
he
br
a
nc
h
le
ngth
si
m
il
arity
(BLS)
e
ntr
opie
s
f
r
om
the
boun
da
ry
pi
xels
of
a
butt
erf
ly
s
hap
e
wer
e
ext
racted
in
t
his
stud
y. Woo
d
sp
eci
es
rec
ogniti
on
wa
s
pro
pos
ed
by
Zha
o
et
al
.
[4
, 5
]
and
Z
a
m
ri
et
a
l.
[6
]
.
In
[
4]
an
d
[
6],
i
m
age
base
d
fe
at
ur
e
s
(co
l
or
,
te
xt
ur
e,
an
d
s
pectral
f
eat
ur
es
)
wer
e
e
xtracted
to
ide
ntify
the
w
ood
sp
eci
es
by
us
i
ng
the
back
prop
a
gati
on
neural
network.
I
n
[5
]
,
k
-
near
e
st
neig
hbor
(k
-
N
N)
was
us
e
d
to
cl
assify
wo
od
sp
eci
es
thr
ough im
ages.
Im
age
-
based plant
sp
eci
es
re
cogniti
on
by
us
in
g k
-
N
N
wa
s also su
ggest
ed by Fa
ria et
al. [7]
.
Christi
anse
n
et
al
.
[8
]
us
e
a
k
-
NN
cl
assifi
e
r
to
discrim
inate
ani
m
al
and
non
-
anim
al
ba
sed
on
heat
char
act
e
risti
cs
of
ob
j
ect
s.
Whi
le
in
the
wo
rk
of
Y
u
et
al
.
[9
]
.,
Su
pp
or
t
Ve
ct
or
Ma
chine
(
SV
M)
has
bee
n
us
e
d
to
extrac
t
feat
ur
es
a
nd
cl
ass
ify
i
m
ages
of
57
a
nim
al
sp
eci
es
captur
e
d
by
cam
era
tr
aps
with
a
n
a
ver
a
ge
cl
assifi
cat
ion
a
ccur
acy
of 82
%.
To
our
knowle
dg
e
,
the
cl
os
es
t
wo
r
k
to
our
r
esearch
ca
n
be
fo
un
d
in
[10]
and
[
11]
.
In
th
ese
works
,
autom
at
ic
sn
ake
sp
eci
es
ide
ntific
at
ion
te
chn
i
qu
e
s
from
sn
a
ke
i
m
ages
wer
e
pr
opos
e
d
by
us
ing
m
a
chine
le
arn
in
g
al
gori
thm
s.
A
m
ir
et
al
.
[10]
ap
plied
te
xture
based
appr
oach
as
fe
at
ur
es,
w
hile
Ja
m
es
et
al
.
[1
1]
us
e
d
featur
e
s
desc
ribing
top,
side
and
body
view
s
of
sna
ke
im
a
ges.
I
n
co
ntras
t,
NLP
was
util
iz
ed
in
our
w
ork
t
o
enab
le
sn
a
ke
s
pecies rec
ognit
ion
t
hro
ugh
te
xt
-
base
d
in
f
or
m
at
ion
from
a h
um
an.
2.
RESEA
R
CH MET
HO
D
This
w
ork
in
volve
d
the
colle
ct
ion
of
the
te
xt
-
based
desc
riptio
n
of
s
na
ke
s
based
on
th
e
pr
ese
nted
sn
a
ke
i
m
ages
by
us
in
g
surv
e
y
m
e
tho
ds
(qu
est
ionnaire)
.
T
hen,
i
m
po
rta
nt
featur
es
wer
e
extracte
d
by
us
in
g
te
rm
fr
eq
uen
c
y
–
inv
erse
docum
ent
fr
eq
ue
ncy
(TF
-
I
DF),
an
d
these
f
eat
ur
es
we
re
pro
vid
e
d
to
m
achin
e
le
arn
in
g
al
gorithm
s to
le
arn
an
d pr
e
dict t
he sn
ake s
pecies
us
i
ng
W
e
ka
t
oo
l
[12
]
.
2.1.
Ra
w
D
ata Co
ll
ecti
on
60
respo
nd
e
nt
s
fr
om
m
ulti
pl
e
ranges
of
ag
e
par
ti
ci
pated
in
the
qu
e
sti
onnai
re
survey
durin
g
data
colle
ct
ion
proc
ess.
T
he
res
po
nd
e
nts
wer
e
s
how
n
with
se
ries
of
sn
a
ke
’s
pictures.
Im
ages
of
t
hr
ee
sp
eci
es
of
sn
a
kes
(
Na
j
a
Trip
ud
ia
ns
,
B
oa
Con
stric
tor
,
and
D
og
-
T
oothed
Ca
t)
w
ere
us
e
d
in
t
his
s
urvey.
Af
te
r
that,
th
e
respo
nd
e
nts
w
ere
aske
d
a
fe
w
quest
io
ns
in
a
qu
e
sti
onnair
e
to
desc
ribe
t
he
s
nak
e
im
age
that
they
ha
d
see
n
base
d
on
their
per
ce
ptio
n
a
nd
opinio
n
of
the
sn
a
ke.
T
hey
wer
e
al
lowe
d
t
o
us
e
t
heir
ow
n
wor
ds
to
e
xpla
in
the
sn
a
kes’ cha
racteri
sti
cs. Two
e
xam
ples o
f sna
ke
im
ages ar
e s
how
n
in
Fig
ure
1
.
Durin
g
the
sur
vey,
the
res
pond
e
nts
are
gui
ded
to
wa
rd
s
e
xp
la
ini
ng
th
e
ei
gh
t
physi
cal
char
act
erist
ic
s
of
the
sn
a
ke
–
that
play
s
a
m
a
j
or
ro
le
in
deci
di
ng
wh
at
ki
nd
of
the
s
nak
e
t
hat
is
venom
ou
s
or
non
-
ve
nom
ou
s.
A
s
nake
ob
se
r
ver is al
ways
usi
ng these c
ha
r
act
erist
ic
s to
re
cognize t
he
s
na
ke
s
pecies:
a)
Len
gth
of it
s bod
y
b)
The
s
ha
pe of
it
s bod
y
c)
Its h
ea
d
a
nd
ne
ck
s
ha
pe
d)
The
c
olor a
nd
patte
rn on its
body
e)
Scal
e tex
tu
re
f)
Ey
e pup
il
s
hape
g)
Tai
l scal
es
h)
An
al
plat di
vis
ion
180
sam
ples
of
unstr
uctu
re
d
te
xt
re
pres
ent
the
s
na
ke
s’
de
scripti
on
are
obta
ine
d
from
the
qu
e
sti
onnaire
.
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Snake
sp
eci
es
i
den
ti
fi
catio
n b
y u
si
ng natur
al
lang
uage pr
oc
essing
(
Nur L
i
yana Izz
ati Ru
s
li
)
1001
(a)
(b)
Figure
1. Tw
o exam
ples o
f sn
ake im
ages f
r
om
the sp
eci
es
of Do
g
-
To
oth
e
d C
at
(
a)
and B
oa
Co
ns
tric
tor
(b)
2.2.
Te
xt
Pre
-
pr
ocessin
g
The
m
et
ho
d
of
pr
e
-
proces
sin
g
te
xt
is
the
first
ste
p
and
an
i
m
po
rtant
ste
p
in
te
xt
m
ining
te
chn
iq
ues
.
Pr
e
-
proces
sin
g i
s p
e
rfor
m
ed
to m
ini
m
ise
the d
im
ension
al
it
y of t
he rep
res
entat
ion
sp
ace
wh
ic
h
incl
uded
[13
]
:
a)
Da
ta
to
ken
iz
at
i
on
To
ken
iz
at
io
n
was
pe
rfor
m
ed
us
ing
Wek
a
t
o
br
ea
k
do
wn
a
te
xt
into
pieces
of
w
ords.
In
this
w
ork,
tok
e
nizat
ion i
s
bro
ken into
wo
rd
s
. E
xam
ple of to
ke
nizat
ion
i
s sho
wn as
fo
ll
ow
s:
Inp
ut: I
sa
w
a
gr
ee
n
s
na
ke,
a
nd it
h
as
tw
o
f
angs
Ou
t
pu
t:
I,
sa
w,
a,
gr
ee
n, sn
a
ke
, and, i
t, h
a
s,
t
wo, fan
gs
b)
Ste
m
m
ing
Ste
m
m
ing
is
t
he
process
of
fin
ding
the
r
oot
of
the
w
ord
f
r
om
diff
ere
nt
w
ord
form
s
,
w
her
e
the
su
f
fixes
an
d
prefixes
will
be
rem
ov
ed.
A
w
ord
s
uc
h
as
“
pl
ay
ing
”
an
d
“
pl
ay
ed”
can
be
stemm
ed
as
“play
”.
Ste
m
m
ing
was
nee
ded
as
it
preve
nts
ov
e
rf
l
ow
of
t
he
diff
e
r
ent
w
ord
with
the
sam
e
m
eaning
i
n
the
li
br
a
ries.
Exam
ple o
f st
em
m
ing
p
r
ocess
shown a
s foll
ows:
Inp
ut: I,
saw
, a
, gree
n,
s
na
ke,
and, it,
has, t
w
o,
fa
ng
s
Ou
t
pu
t:
I,
see
, a
, gree
n,
s
na
ke
, and, i
t, h
a
s,
t
wo,
f
an
g
c)
Sy
m
bo
ls a
nd S
top
-
w
ord
el
im
i
nation
The
ste
m
m
ed
te
xt
ob
ta
ine
d
pr
e
viously
unde
rw
e
nt
the
pro
cess
in
rem
ov
ing
al
l
the
sp
ec
ia
l
sy
m
bo
ls
su
c
h
as
‘(‘,
‘
)’
,
‘#
’
,
‘!’,
‘
?
’
,
‘
_’,
‘+
’,
‘
-
‘,‘
*’
,
a
nd
‘/’.
Sto
p
w
ord
or
sto
p
w
ord
li
st
are
the
set
of
com
m
on
wor
d
that
h
um
an
us
e
ever
y
day
in
any
la
nguag
e
.
It
do
es
not
ha
ve
le
ss
sign
ific
ant
m
eaning
in
th
e
te
xt
or
par
a
grap
h.
Com
m
on
w
ord
s (
e.
g.
“a”, “a
n”
, and
“t
he”
)
ar
e elim
inate
d
by
u
sing
sto
p
-
w
ord
rem
ov
al
fu
nction
i
n W
e
ka
. Th
is
process
can
m
i
nim
ise
the d
im
ensio
nalit
y about
15%
to 2
0%
r
e
du
ct
io
n
i
n
th
e colle
ct
ed dat
a [
13
]
.
Inp
ut: i
, s
ee,
a,
gr
ee
n, sn
a
ke, a
nd, it,
has, t
w
o, fa
ng
Ou
t
pu
t:
i,
see,
gr
ee
n, sn
a
ke, i
t, h
a
s,
t
wo, f
a
ng
2.3.
Fe
ature
Extr
act
i
on
using
TF
-
I
DF
A
high
num
ber
of
w
ords
in
th
e
te
xt
descr
ipti
on
will
cause
high
dim
ension
al
it
y
of
the
re
pr
ese
ntati
on
sp
ace d
ur
i
ng
tr
ai
nin
g
a
nd
te
sti
ng.
T
heref
or
e
, in
this w
ork,
T
F
-
I
DF
wei
gh
ti
ng
is use
d
to
id
entify
im
po
rtant
an
d
releva
nt
key
w
ords
from
each
descr
i
ptio
n.
These
hi
gh
w
ei
gh
ti
ng
keyw
ords
will
be
e
xtracted
as
im
portan
t
featu
re
s
an
d
w
il
l
be
us
ed
to
op
ti
m
ise
the
trai
nin
g
process.
Term
fr
eq
uency
(TP)
re
pr
es
ents
how
m
any
tim
es
the
num
ber
of
t
he
w
ord
that
occ
ur
s
in
a
sin
gle
te
xt
or
doc
um
ent.
We
use
d
a
fle
xib
l
e
filt
er
nam
ed
Strin
gtoWo
r
dV
ect
or
i
n
W
e
ka
t
o
c
onve
rt
strin
g
at
tri
bu
te
s
in
to
a
se
t
of
w
ord
vect
or
w
hich
re
pr
e
sents
the
w
ords
occ
urren
ce
.
Be
lo
w
are
t
hr
ee
e
xam
ples
of
a
s
na
ke
de
scripti
on
by
a
hu
m
an
in
te
xt
f
or
m
a
nd
how
featur
e
ex
t
racti
on is
done.
Exam
ple:
Text
1: “I
saw
a long a
nd a
gr
een s
na
ke.
”
Text
2: “T
he g
reen sna
ke
is
a
dang
e
rous sna
ke.
”
Text
3: “T
he
lo
ng snake
is sca
rier.”
Af
te
r
pre
-
proc
essing, the
text
w
il
l be a
s foll
ow
s:
Text
1: I
,
see,
long, g
ree
n,
a
nd, sna
ke
Text
2: gree
n,
sn
a
ke,
be, da
nger
ous,
s
na
ke
Text
3: lo
ng, snake
, be,
scary
TF
-
ID
F
w
ei
ght o
f
a term
is cal
culat
ed
as
f
ollow
s:
(a)
Ca
lc
ulate
term
fr
e
qu
e
ncy
(TF)
(b)
Ca
lc
ulate
d
oc
um
ent f
re
qu
e
nc
y (D
F
)
a
nd the
inv
e
rse of
the
DF
(IDF
).
(c)
Com
pu
te
TF
-
I
DF
The n
or
m
al
iz
e
d
T
F is m
easure
d
acc
ordin
g
t
o
E
quat
ion (
1).
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
1
3
, N
o.
3
,
Ma
rc
h
201
9
:
999
–
1006
1002
=
.
ℎ
ℎ
ℎ
(1)
In
reali
ty
,
each
te
xt
will
con
t
ai
n
dif
fer
e
nt
si
ze,
an
d
us
ually
,
the
va
lue
of
TF
will
be
higher
t
han
a
n
exam
ple
of
TF
in
Table
1.
N
ext,
th
e
te
xt
w
il
l
be
norm
aliz
ed
ba
sed
on
it
s
siz
e
by
div
i
di
ng
T
F
by
the
total
nu
m
ber
of ter
m
s.
Table
1.
T
F Ba
sic
Cal
culat
ion f
or
Text
1, Te
xt 2 an
d Text
3
Text 1
i
see
lo
n
g
an
d
g
reen
sn
ak
e
TF
1
1
1
1
1
1
No
r
m
TF
0
.16
7
0
.16
7
0
.16
7
0
.16
7
0
.16
7
0
.16
7
Text 2
g
reen
sn
ak
e
be
d
an
g
erou
s
TF
1
2
1
1
No
r
m
TF
0
.20
0
0
.40
0
0
.20
0
0
.20
0
Text 3
lo
n
g
sn
ak
e
be
scary
TF
1
1
1
1
No
r
m
TF
0
.25
0
0
.25
0
0
.25
0
0
.25
0
In
TF,
al
l
te
r
m
s
being
treat
ed
as
e
qual
.
I
n
c
on
tra
st,
the
inv
e
rse
do
c
um
ent
fr
e
qu
e
nc
y
(I
D
F)
is
a
m
easur
e
of
ho
w
m
uch
in
for
m
at
ion
the
wor
d
prov
i
des
ac
r
os
s
al
l
te
xt
or
do
c
um
ents.
Th
us
,
IDF
is
c
ompu
te
d
as
fo
ll
owin
g
E
qu
at
ion
2:
=
log
(
.
ℎ
ℎ
)
(2)
Fo
r
e
xam
ple, the
te
rm
o
f
t
he “
gr
ee
n” was
use
d
to
f
i
nd IDF:
Total
no. o
f
te
xts: 3
Nu
m
ber
of text
s w
it
h t
erm
g
re
en on i
t:
2
(
)
=
(
3
2
)
Table
2
s
hows
the ex
am
ple of
m
easur
ed ID
F
v
al
ue
for t
erm
s that a
pp
ea
re
d i
n
al
l t
he
text.
Finall
y, TF
-
I
D
F w
ei
gh
t i
s
m
e
asur
e
d usi
ng E
qu
at
io
n (
3)
.
Table
2.
Inve
rs
e Docum
ent Frequ
e
ncy
Ter
m
s
IDF
I
0
.47
7
see
0
.47
7
lo
n
g
0
.47
7
an
d
0
.47
7
g
reen
0
.17
6
sn
ak
e
0
.00
0
be
0
.17
6
d
an
g
erou
s
0
.47
7
scary
0
.47
7
_
=
∗
(3)
Table
3.
E
xam
ple on
w
ord oc
currence
s in
T
F
-
I
DF
W
o
rds
Text 1
Text 2
Text 3
i
0
.08
0
-
-
see
0
.08
0
-
-
lo
n
g
0
.08
0
-
0
.12
0
an
d
0
.08
0
-
-
g
reen
0
.02
9
0
.03
5
-
sn
ak
e
0
.00
0
0
.00
0
0
.00
0
be
-
0
.35
2
0
.04
4
d
an
g
erou
s
-
0
.09
5
-
scary
-
-
0
.12
0
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Snake
sp
eci
es
i
den
ti
fi
catio
n b
y u
si
ng natur
al
lang
uage pr
oc
essing
(
Nur L
i
yana Izz
ati Ru
s
li
)
1003
Fr
om
the
exam
ple
in
Ta
ble
3,
the
w
ord
“sna
ke”
is
c
onsider
ed
c
omm
on
due
to
it
is
an
oc
currence
i
n
al
l
descr
i
ption
with
wei
gh
t
val
ue
of
0.0
00.
In
a
no
t
he
r
word,
t
he
word
“s
na
ke”
is
not
si
gnif
ic
ant
in
determ
ining
th
e
cha
racteri
sti
c
of
a
sn
a
ke
de
scribe
d
i
n
eac
h
te
xt
.
T
her
e
fore,
a
w
ord
wit
h
ze
ro
or
l
ow
weig
ht
will
b
e c
onside
red irrele
van
t
f
eat
ur
e a
nd
om
i
tt
ed
duri
ng trai
ning a
nd classi
ficat
ion
.
2.4.
Tr
aining
a
n
d
Clas
sific
at
i
on
In
s
up
e
rv
ise
d
cl
assifi
cat
ion
,
trai
ning
m
us
t
be
first
co
nduc
te
d,
an
d
cl
assifi
cat
ion
ta
sk
f
ollows
this.
The
trai
ning
involve
s
bu
il
din
g
a
m
od
el
based
on
one
or
m
or
e
nu
m
eric
al
and
cat
egori
cal
var
ia
bles
su
ch
a
s
at
tribu
te
s
or
fe
at
ur
es.
Cl
assifi
cat
ion
is
a
te
xt
m
ining
ta
sk
of
pr
e
dicti
ng
the v
al
ue
of
a
cat
egorical
var
ia
bl
e
su
ch
as target
or cla
ss.
Four
m
achine
le
arn
in
g
al
gorithm
s
wer
e
ch
ose
n
to
pe
rfo
rm
these
ta
s
ks
.
T
hey
are
naïve
Ba
ye
s
[13],
Suppor
t
Ve
ct
or
Ma
chine
(SV
M)
[14],
k
-
Ne
arest
Neig
hbou
rs
(k
-
NN)
[
15
]
,
and
decisi
on
tree
J4
8
[
16]
.
I
n
this
work,
18
0
sam
ples
of
te
xt
-
base
d
desc
ription
c
ollec
te
d
from
60
respo
ndents
will
be
us
e
d
for
trai
ni
ng
a
nd
cl
assifi
cat
ion
.
Du
e
to
li
m
i
ted
sam
ple,
in
order
to
ob
ta
i
ned
m
or
e
accurate
resu
lt
,
10
-
fo
l
d
strat
ifie
d
cr
oss
validat
io
n
was
app
li
e
d
to
e
nsure
t
he
vali
dity
of
our
re
su
lt
be
co
nduc
te
d
f
or
eac
h
al
go
rithm
.
The
trai
ni
ng
a
nd
cl
assifi
cat
ion
proces
ses
wer
e
perform
ed
on
diff
e
re
nt sets
of the
d
at
a as
to
gen
e
rali
ze the
new in
form
at
io
n.
3.
RESU
LT
S
A
ND AN
ALYSIS
The
quest
io
nn
ai
re
was
fill
ed
by
the
so
ci
et
y
throu
gh
s
ocial
m
edia
as
the
m
edium
,
and
we
obta
ine
d
180
sam
ples
from
60
pa
rtic
ipants.
Eac
h
par
t
ic
ipant
was
a
s
ked
t
o
de
scri
be
three
s
na
ke
im
ages
(r
e
prese
nting
a
sp
eci
es
each
).
The
ra
w
datase
t
in
te
xt
fo
rm
t
hen
was
im
po
rted
into
an
Attr
ibu
te
-
Re
la
ti
on
Fil
e
Fo
rm
at
(A
RFF)
file
.
The
n,
pre
processi
ng
an
d
featu
re
e
xtract
ion
we
re
perform
ed.
ARFF
fi
le
is
le
ss
m
e
mo
ry
i
ntensiv
e
,
faster
and b
et
te
r
for a
naly
sis beca
use
it
inclu
des
m
et
a d
at
a a
bout
colum
n
hea
der an
d data col
um
n.
Convertin
g
a
word
t
o
a
vect
or
is
sim
ply
a
m
echan
ism
to
input
an
d
pro
c
ess
w
ords
f
or
any
natu
ral
la
nguag
e
proc
essing
ta
sk.
A
s
m
entioned
in
Se
ct
ion
V
,
durin
g
prep
ro
c
essing,
We
ka
pack
a
ge
was
us
e
d
t
o
conve
rt
word
into
the
vecto
r.
This
res
ults
in
483
at
tribu
te
s
wer
e
f
ou
nd
in
the
da
ta
set
.
Then,
fe
at
ur
es
extracti
on
m
eth
ods
s
uc
h
as stop
w
ord
el
im
in
at
ion
,
ste
m
m
er
,
an
d
to
ke
nizer
wer
e
pe
rfor
m
ed.
Fi
nally
,
TF
-
IDF
T
trans
form
w
as
execu
te
d
t
o
cal
culat
e the
weig
ht of eac
h w
ord
in
each
doc
um
ent.
These
feat
ur
e
extracti
on
ta
sks
resu
lt
in
a
re
du
ct
io
n
of
the
dim
ension
al
it
y
of
at
trib
utes
to
30%.
Af
te
r
featur
e
s
extrac
ti
on
,
the
num
ber
of
at
tribu
te
s
decr
eases
t
o
346
at
trib
utes.
Hen
ce
,
the
re
s
ulti
ng
data
set
wh
ic
h
cal
le
d
the
NLP
-
Snake
d
at
a set
consist
s
of
18
0
sam
ples w
it
h 3
46 att
r
ib
utes.
3.1.
Clas
si
ficat
i
on
Accur
ac
y
Cl
assifi
cat
ion
accuracy
is
pr
esented
as
a
per
ce
ntage
w
he
re
100%
is
the
best
a
n
al
gorithm
can
achieve.
F
our
m
achine
le
arn
i
ng
al
go
rithm
s
decisi
on
tree
J4
8,
SV
M
(Li
ne
ar
Kernel)
,
na
ïve
Ba
ye
s,
k
-
N
N
are
sel
ect
ed
as
cl
assif
ie
rs
in
this
pro
j
ect
.
The
pe
rfor
m
ance
of
t
he
cl
assifi
ers
a
s
repor
te
d
in
F
igure
2
il
lustra
te
s
the
correct
ly
and
inco
rr
ect
ly
cl
a
ssifie
d
instanc
es.
Figure
2
ind
ic
at
es
that
J4
8
has
the
hi
gh
e
st
per
ce
ntage
of
71.67%
fo
ll
ow
ed
by
S
VM
w
it
h
68.33%
.
N
aï
ve
Ba
ye
s
ob
ta
i
ned
61.
11%
wh
ic
h
the
n
f
ol
lowed
by
k
-
N
N
by
55.56%
as the
lowest
per
ce
nt
age
ob
ta
in
ed
for co
rr
ect
ly
classi
fied
i
ns
ta
nce
s.
Fo
r
i
ncorr
ect
l
y
cl
assifi
ed
instances
,
J48
ob
ta
ine
d
the
lowest
pe
rce
ntage
of
bein
g
inc
orrectl
y
cl
assifi
ed
insta
nces
with
28.
33%.
This
the
n
fo
ll
owe
d
by
31.
67%
f
or
S
V
M,
38.
89%
f
or
naïve
Ba
ye
s.
k
-
NN
ob
ta
ine
d
t
he
highest
propo
r
ti
on
of
inc
orr
ect
ly
cl
assifi
e
d
in
sta
nces
by
44.44%
.
He
nce,
J48
ac
hie
ves
t
he
highest
pe
rce
nt
age
of
c
orrec
t
pr
e
dicti
on
c
om
par
ed
to
S
VM,
k
-
NN,
a
nd
naï
ve
Ba
ye
s
for
t
he
N
LP
-
Sn
ake
dataset
.
Figure
2. The
a
ccur
acy
of n
aï
ve
Ba
ye
s, k
-
N
N,
SV
M,
and
J
48 for N
LP
-
Snake
dataset
0
50
1
0
0
n
aïv
e Bay
es
k
-NN
SVM
(L
in
ear
Ker
n
el)
J4
8
6
1
.11
5
5
.56
6
8
.33
7
1
.67
3
8
.89
4
4
.44
3
1
.67
2
8
.33
Cla
ss
ified
Ins
tances
Co
rr
e
ctly
Class
ifi
ed
Inco
rr
ectly
Class
if
ied
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
1
3
, N
o.
3
,
Ma
rc
h
201
9
:
999
–
1006
1004
3.2.
Preci
si
on a
nd Rec
all
Re
gardin
g
pro
bab
il
ist
ic
inter
pr
et
at
io
n,
preci
sion
an
d
recall
are
no
t
i
nter
pret
ed
as
rati
os
.
I
ns
te
ad
,
they
are
inte
rpreted
as
pro
ba
bili
ti
es.
P
recisi
on
is
the
pr
ob
a
bili
ty
that
a
sel
ect
ed
data
is
releva
nt
w
hile
re
cal
l
is
th
e
pro
bab
il
it
y
that
a
sel
ec
te
d
data
is
cor
rectl
y
retrieve
d.
Pre
c
isi
on
an
d
recal
l
bo
th
are
sta
ti
sti
cal
m
easur
e
s
of
perform
ances o
f
a m
achine learn
i
ng alg
or
it
hm
. Th
e out
com
es w
e
re sh
own i
n
Fig
ure
3
a
nd Fig
ur
e
4.
Figure
3
il
lust
rates
the
preci
sion
pe
rfor
m
ances
obta
ined
in
W
e
ka
i
nter
f
aces
after
te
n
trai
ning
a
nd
cl
assifi
cat
ion
wer
e
car
ried
out.
It
show
s
th
at
the
hig
hest
pr
eci
sio
n
outc
om
e
of
m
achi
ne
le
arn
i
ng
al
gorithm
s
for
Bo
a
Co
ns
tr
ic
tor
was
k
-
N
N
by
87.
2%,
th
e
second
hi
gh
e
st
is
J4
8
by
78.
9%,
the
sec
ond
lowest
preci
sion
f
or
Boa Co
ns
tric
to
r
is
62.1% f
ollow
e
d by
naïve B
ay
es as the l
owest
pr
eci
sio
n by 60%
.
Figure
3. The
pr
eci
sio
n pe
rfo
rm
ance o
f naï
ve
Bay
es, k
-
NN,
S
VM,
a
nd J
48
for
th
ree s
na
ke
s
pecies i
n N
LP
-
Sn
a
ke data
set
Figure
4. The
re
cal
l perfo
rm
a
nce
of n
aï
ve
B
ay
es, k
-
N
N,
S
VM,
a
nd J
48 for t
hr
ee
s
nak
e
sp
eci
es in
NLP
-
Sn
a
ke data
set
The
highest
preci
sion
of
m
achine
le
a
rn
i
ng
a
lgorit
hm
s
fo
r
Dog
-
To
oth
e
d
Ca
t
is
J4
8
with
69.
4%,
SVM
with
64.
6%,
fol
lowed
by
naïv
e
Ba
ye
s
with
57.
9%
as
the
se
cond
lo
west
pr
eci
sion
an
d
k
-
NN
with
56%
as
the
lowest
preci
sio
n.
F
or
Na
j
a
T
ripudian
s,
the
hi
gh
e
st
pr
eci
sio
n
is
J4
8
by
87.
2%
;
the
second
highest
pr
eci
si
on
is
SV
M
by
78.
9%
,
f
ollowe
d
by
the
se
co
nd
lowest
pr
e
ci
sio
n
f
or
Na
j
a
T
ripudian
s
is
naï
ve
Ba
ye
s
by
64.
7%.
More
ov
e
r, t
he l
ow
est
pr
eci
si
on is
obta
ined
by
k
-
NN w
it
h 5
1.6%
only
.
Figure
4
sho
w
s
that
the
hig
he
st
recall
ou
tc
om
e
of
m
achine
le
arn
in
g
al
gorithm
s
fo
r
Boa
Con
stric
to
r
was
J
48
by
75
%,
the
sec
on
d
highest
is
S
V
M
by
60%
,
the
seco
nd
lo
west
recall
f
or
Boa
Const
rict
or
is
55%
in
naïve
Ba
ye
s
f
ol
lowed b
y
k
-
N
N
as t
he
l
ow
est
r
ecal
l by
18.
3%
.
The
highest
re
cal
l
of
m
achin
e
le
arn
i
ng
al
go
rithm
s
fo
r
D
og
-
To
oth
e
d
Ca
t
i
s
J48
t
hat
goes
by
71.7%,
SV
M
an
d
k
-
N
N
sh
a
r
e
the
sa
m
e
ou
tc
om
e
by
70
%
a
nd
naïve
Ba
ye
s
by
55%
as
the
lo
w
est
recall
.
Fo
r
Naj
a
Trip
ud
ia
ns
,
s
urp
risin
gly
the
highest
recall
is
k
-
NN
by
78.
3%,
the
sec
ond
highest
rec
al
l
is
SV
M
by
75
%
,
fo
ll
owe
d
by
th
e
second
lo
we
st
recall
fo
r
N
aja
Trip
udia
ns
is
naïve
Ba
ye
s
b
y
73.3
%
.
J4
8
ob
ta
in
s
the
lowest
recall
w
it
h 6
8.3%
only
.
0
20
40
60
80
1
0
0
n
aïv
e Bay
es
k
-NN
SVM
J4
8
Prec
isio
n
Cla
ss
ifiers
Bo
a Co
n
str
icto
r
Do
g
-T
o
o
th
ed
Cat
Naj
a T
r
ip
u
d
ian
s
55
1
8
.3
60
75
55
70
70
7
1
.7
7
3
.3
7
8
.3
75
6
8
.3
0
20
40
60
80
1
0
0
n
aïv
e Bay
es
k
-NN
SVM
J4
8
Bo
a Co
n
str
icto
r
Do
g
-T
o
o
th
ed
Cat
Naj
a T
r
ip
u
d
ian
s
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Snake
sp
eci
es
i
den
ti
fi
catio
n b
y u
si
ng natur
al
lang
uage pr
oc
essing
(
Nur L
i
yana Izz
ati Ru
s
li
)
1005
In
ge
ne
ral,
for
pr
eci
sio
n
an
d
r
ecal
l
perform
a
nce
am
ong
the
f
our
al
gorith
m
s
fo
r
th
e
NL
P
-
S
nake
data
set
,
J48
sho
ws
the
m
os
t
prec
ise
per
ce
nta
ge
we
re
ob
ta
ine
d
for
eac
h
s
na
ke
s
pecies.
It
al
so
has
th
e
hi
gh
est
accuracy c
om
par
ed
to SVM,
k
-
NN an
d naï
ve
Bay
es.
4.
CONCL
US
I
O
N
The
pa
pe
r
dem
on
st
rates
the
prel
i
m
inary
resul
t
fo
r
the
r
eco
gnit
ion
of
s
nake
sp
eci
es
by
us
i
ng
natu
ral
la
nguag
e
pr
oce
ssing.
Hu
m
an
descr
i
ption
of
sn
a
ke
im
ages
f
ro
m
three
s
pec
ie
s
was
c
ollec
te
d
th
r
ough
a
s
urvey
in
so
ci
al
m
edia.
The
res
ulti
ng
raw
data
set
con
ta
ins
a
te
xtu
al
descr
ipti
on
of
sn
a
ke
char
a
ct
erist
ic
s
by
a
hum
an.
The
pre
-
proces
sing
an
d
feat
ure
sel
ect
ion
was
then
perf
or
m
e
d.
The
feature
extracti
on
ta
sk
inv
ol
ves
sto
p
wor
d
el
i
m
inati
on
,
st
e
m
m
ing
,
word
tok
e
nizer
an
d
TF
-
I
DF
tra
nsfo
rm
.
The
pro
cessed
data
set
,
nam
ed
NLP
-
Snak
e
dataset
,
c
on
sis
ts
of
34
6
at
tr
ibu
te
s
with
180
sam
ples.
Then,
the
pe
rfo
rm
ance
of
f
our
m
achine
le
arn
i
ng
al
gorithm
s
(n
aï
ve
Ba
ye
s,
k
-
N
N,
SV
M
a
nd
de
ci
sion
tree
J
48)
are
e
valuate
d
f
or
trai
ning
a
nd
cl
assifi
cat
io
n.
All
in
al
l,
the
ov
e
rall
per
f
orm
ances
sh
ow
that
the
J4
8
is
th
e
best
an
d
su
i
te
d
for
te
xt
cl
assifi
cat
ion
ta
s
k,
in
par
ti
cula
r,
t
o
id
entify
sn
a
ke
c
ha
racteri
sti
c in
natu
ral la
ngua
ge
ta
s
k.
In
t
he
fu
t
ur
e
,
we
ai
m
to
colle
ct
a
la
rg
e
r
da
ta
set
by
in
volv
ing
a
great
er
num
ber
of
s
na
ke
sp
eci
es
a
nd
m
or
e p
arti
ci
pa
nts.
By
do
i
ng s
o,
m
or
e acc
ura
te
r
esults a
re e
xp
ect
e
d for
rea
l
-
w
or
ld
im
ple
m
entat
ion
.
ACKN
OWLE
DGE
MENTS
We
would
li
ke
to
th
an
ks
t
o
th
e
ra
ndom
par
ti
ci
pan
ts
w
ho
he
lped
us
by
a
nsweri
ng
th
e
sur
vey
f
or
thi
s
stud
y.
W
e
also
wou
l
d
li
ke
t
o
t
hanks
Tam
an
Ular
Perlis f
or
the s
nak
e
p
ic
tu
res.
REFERE
NCE
S
[1]
W
HO
blood
produc
ts
and
relat
ed
Biol
ogi
ca
ls a
ni
m
al
sera
Ant
ivenom
s
fra
m
es
page
.
R
et
ri
eve
d
Oc
tobe
r
7
,
2016
,
fr
om
htt
p://apps.
who.
i
nt/
bloodproducts/snake
an
ti
venom
s/dat
ab
ase
/
[2]
Chew,
K.S.
,
Kh
or,
H.W.,
Ahm
ad,
R.
,
Rahman
,
N.A.H.
N
(2011)
.
A
Five
-
y
e
ar
r
et
rospec
ti
ve
rev
i
ew
of
snake
b
ite
pat
i
ent
s a
dm
itte
d
to a ter
t
ia
r
y
un
ive
rsit
y
hospital
in
Mal
a
y
s
ia.
Int
e
rna
ti
on
al
Journ
al of
E
m
erg
ency
Medic
in
e
4(1)
,
1
-
6
[3]
Kang,
S.H.
,
Song,
S.H.,
Lee,
S
.
H.
(2012).
Id
entificat
ion
of
butterfl
y
spe
ci
es
wit
h
a
single
neur
a
l
net
work
s
y
s
tem
.
Journal
of
As
i
a
-
Paci
fi
c
En
tomolog
y
15(3), 431
–
435.
[4]
Zha
o,
P.,
Dou,
G.,
Chen,
G.S
.
(2014)W
ood
Speci
es
ide
nt
ifica
t
ion
using
f
e
at
ure
-
le
ve
l
fusio
n
sche
m
e.
Opti
k
-
Inte
rna
ti
ona
l
Jou
rna
l
for
L
ight
an
d
Elec
tron
Opti
c
s 125(3),
1144
–
1148.
[5]
Zha
o,
P.
,
Dou,
G.,
Chen,
G.S.
(2014).
W
ood
s
pec
i
es
ide
ntific
a
ti
on
using
improved
ac
t
ive
sha
pe
m
odel
.
Optik
-
Inte
rna
ti
ona
l
Jou
rna
l
for
L
ight
an
d
Elec
tron
Opti
c
s 125(18),
5212
–
5217
[6]
Za
m
ri,
M
.
I.
P.
,
C
ordova
,
F.
,
Kha
i
ruddin,
A.S
.
M.,
Mokhtar,
N.
,
Yu
sof,
R.
(2016)
T
ree
spe
cies
class
ifi
c
at
ion
base
d
o
n
image
anal
y
s
is
using
improved
-
basic
gre
y
le
ve
l
aur
a
m
at
rix
.
Co
m
pute
rs
and
El
e
ct
roni
cs
in
Agri
cul
tur
e
124,
227
–
233.
[7]
Faria
,
F.A
.
,
Al
m
ei
da,
J.
,
Alber
ton,
B.
,
Morel
lato,
L
.
P.C.
,
Roc
ha,
A.
,
da
S.
T
orre
s,
R.
(2015)
Ti
m
e
serie
s
-
b
ase
d
cl
assifi
er
fusion
for
fine
-
g
rai
n
ed pla
nt
spec
i
es
re
c
ognit
ion.
Pa
tt
er
n
Recogni
t
ion
L
et
t
ers
pp.
[8]
Christi
anse
n
,
P.,
Stee
n,
K
.
A.
,
Jrgensen,
R
.
N.,
Karstoft
,
H.(2014)
Autom
at
ed
de
t
ec
t
ion
and
r
ec
o
gnit
ion
of
w
il
dlif
e
using t
her
m
al ca
m
era
s.
Sensors
1
4(8),
13778
[9]
Yu,
X.,
W
ang
,
J.,
Ka
y
s,
R.
,
Jan
sen,
P.A.
,
W
ang
,
T
.
,
Huang,
T
.
(
2013).
Autom
ated
id
ent
if
icati
on
of
an
imal
spe
cies
in
c
amera
tra
p
i
m
age
s.
EUR
ASI
P
Journal
on
Im
age
and
Vi
d
eo
P
roce
ss
ing
2013
(
1),
1
–
10
[10]
Am
ir
A.,
Za
hri
N.A.H.
,
Yaa
kob
N.,
Ahm
ad
R.
B.
(2017)
Im
age
Cla
ss
ifi
c
at
ion
for
Snake
Species
Us
ing
Mac
hine
Le
arn
ing
Techni
ques.
In:
Phon
-
Am
nuai
suk
S.,
Au
TW.,
Om
ar
S.
(ed
s)
Com
put
at
ion
al
Intelli
g
e
nce
in
Inform
at
i
on
S
y
stems
.
CIIS 2
016.
Advan
ce
s i
n
Intelli
g
ent S
y
s
te
m
s a
nd
Com
puti
ng,
vol
5
32.
Springer, Cha
m
[11]
Jam
es,
A.P.,
Ma
the
ws
,
B.
,
Suga
t
han,
S.,
R
aveen
dra
n,
D.K.
(201
4)
Discriminat
iv
e
histogra
m
ta
xo
nom
y
f
ea
tu
res
f
or
snake
spec
ie
s
id
ent
ific
a
ti
on.
Hu
man
-
ce
ntric Co
mputing
and
In
f
orm
ati
on
Scienc
es
4(1), 1
–
11
[12]
M
Im
ambi,
S.,
&
Sudha
(2011).
Pre
-
Proce
ss
in
g
of
m
edi
ca
l
d
ocuments
and
r
educ
ing
Dim
ens
iona
lit
y
.
Ad
vanced
Computing:
An
I
nte
rnational
Jou
rnal
,
2
(5)
,
15
–
24
[13]
Za
idi,
N.
A.
,
Pet
it
jean,
F.
,
&
W
e
bb
,
G.
I.
(2016).
Prec
ondit
ion
ing
an
Artifi
c
ia
l
Ne
ura
l
Network
Us
ing
Naive
Ba
y
es
.
Proce
ed
ings
of
t
he
20th
Pac
i
fic
-
Asia
Confe
ren
ce
on
Adv
an
ce
s
in
Knowle
dge
Discov
ery
and
Data
Mini
ng
,
P
AKDD
2016
,
pp
.
341
-
3
53
[14]
Cao
J,
Fang
Z,
Qu
G,
Sun
H,
Zha
ng
D
(2017).
An
ac
cu
ra
te
tr
aff
ic
c
la
ss
ifica
t
i
on
m
odel
base
d
on
support
vec
tor
m
ac
hine
s.
Int J Network
Mgmt
.
2017;27:
e1962
.
[15]
P.
Gu,
R.
Kh
at
o
un,
Y.
B
egr
ic
h
e
and
A.
Serhrou
c
hni.
(2017)
.
k
-
N
ea
rest
Neighbou
rs
cl
assificat
ion
base
d
S
y
b
il
at
t
a
ck
det
e
ct
ion
in
Vehic
u
la
r
Netwo
rks.
2017
Thir
d
Inte
rnational
Confe
renc
e
o
n
Mobil
e
and
Sec
ure
Serv
ices
(
MobiSe
cSe
rv)
,
Miami
Beach, FL,
2017
,
pp
.
1
-
6
.
[16]
M.
Aashkaa
r,
P.
Sharm
a
and
N.
Garg.
(2017).
Perform
anc
e
an
aly
s
is
using
J48
dec
ision
tree
fo
r
India
n
cor
pora
te
world.
2016
In
ter
nati
onal
Conf
ere
nce
on
R
ese
arc
h
Adv
an
ce
s
in
In
te
grated
Navi
ga
t
ion
Syst
ems
(
RA
INS)
,
Banga
lor
e,
2016,
pp
.
1
-
5.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
1
3
, N
o.
3
,
Ma
rc
h
201
9
:
999
–
1006
1006
BIOGR
AP
HI
ES OF
A
UTH
ORS
Nur
Liy
ana
Ru
sli
gra
duated
with
Bac
h
el
or
of
Com
pute
r
Network
Engi
nee
ri
ng
from
Univer
sit
y
Ma
lay
sia
Perli
s
in
20
17.
She
is
now
working
in
E
rics
on
as
a
NO
C
E
ngine
er
.
Her
intere
st
incl
ude
m
ac
hin
e lea
rning
and
compu
te
r
n
et
works
.
Am
iz
a
Am
ir
is a seni
or le
ct
ure
r
i
n
School
of
Com
pute
r and
Com
m
unic
a
ti
on
Engi
ne
eri
ng
at
Univ
ersit
i
Ma
lay
s
ia Perl
is.
Sh
e
re
ceive
d
her
P
h.
D. in
Inform
ati
on
Technol
og
y
,
on
distri
bute
d
ar
ti
fi
ci
a
l
in
te
l
li
gen
ce,
from
Monash Unive
rsit
y
,
Aus
tralia i
n
2015
.
H
er curr
ent
rese
arc
h
in
te
rest
s inc
lud
e
m
ac
h
in
e
l
ea
rn
ing
,
d
istributed
s
y
stem,
m
et
a
heur
ist
ic
opti
m
iz
ation,
da
t
a
an
aly
t
ic
s
and
s
oftwa
re
-
def
ine
d
net
work (SDN
).
She
teac
h
es
cour
ses i
n
dat
a
an
aly
t
ic
s a
n
d
artificia
l
in
te
l
ligence.
NI
K
AD
ILA
H
HANI
N
Z
AHR
I
is a seni
or le
cturer
in
School
of
Com
pute
r and
Com
m
unic
at
ion Engi
ne
eri
ng
at
Univer
siti
Malays
ia
Perl
is.
She
re
ce
iv
ed
h
er
Ph.D.
in
Medic
a
l
Eng
inee
ring,
on
Com
m
unic
a
ti
on
and
Inf
orm
at
ion
S
y
s
te
m
,
from
Univer
si
t
y
of
Yam
ana
shi,
Jap
a
n
in
2013
.
He
r
c
urre
nt
r
ese
ar
ch i
nte
rests
inc
lud
e nat
ura
l
l
angua
g
e
proc
essing,
m
a
c
hin
e le
arn
ing, data
m
ini
ng
and
da
ta
ana
l
y
t
i
cs.
She
te
a
che
s progr
amm
ing,
software
engi
ne
e
ring
and
da
ta
an
aly
tics c
ourse
R.
Prof.
Dr.
R
.
B
adl
ishah
Ahm
ad is a
Profess
or in
Malay
s
ia.
He
is
Deput
y
Vic
e
Ch
anc
e
ll
or
(Rese
arc
h
and
In
novat
ion)
,
Univ
e
rsiti
Sult
an
Za
in
al
Abid
in
(UniS
ZA)
since 15
Ma
rch
2017.
Gradua
te
d
PhD
(1999)
fro
m
Univer
sit
y
of
Strat
hcly
d
e
(Sco
tl
and
,
UK
).
H
e
h
as
supervise
d
m
ore
tha
n
40
PhD
an
d
MS
c
student
s.
Speci
a
li
z
ed and Expe
rt
ise
in
Co
m
pute
r
and
T
el
e
comm
unic
a
ti
on
Ne
twor
k
Modelling,
Embedde
d
S
y
s
te
m
Design
and
Op
e
n
Source
Software
.
Evaluation Warning : The document was created with Spire.PDF for Python.