I
nd
o
ne
s
ia
n J
o
urna
l o
f
E
lect
rica
l En
g
ineering
a
nd
Co
m
pu
t
er
Science
Vo
l.
4
3
,
No
.
1
,
Ju
ly
2
0
2
6
,
p
p
.
259
~
27
0
I
SS
N:
2502
-
4
7
5
2
,
DOI
: 1
0
.
1
1
5
9
1
/ijeecs.v
4
3
.i
1
.
pp
259
-
27
0
259
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ee
cs.ia
esco
r
e.
co
m
Ev
a
lua
ting o
v
ers
a
mpling
methods
for imba
la
nced
Ara
bic dialec
t
ide
ntif
ica
tion
M
a
ula
na
I
h
s
a
n Ahm
a
d,
Ain
a
M
us
d
ho
lifa
h,
Arif
Nurwid
y
a
nto
ro
D
e
p
a
r
t
me
n
t
o
f
C
o
mp
u
t
e
r
S
c
i
e
n
c
e
a
n
d
El
e
c
t
r
o
n
i
c
s
,
F
a
c
u
l
t
y
o
f
M
a
t
h
e
ma
t
i
c
s
a
n
d
N
a
t
u
r
a
l
S
c
i
e
n
c
e
s
,
U
n
i
v
e
r
si
t
a
s
G
a
d
j
a
h
M
a
d
a
,
Y
o
g
y
a
k
a
r
t
a
,
I
n
d
o
n
e
si
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
:
Feb
4
,
2
0
2
6
R
ev
is
ed
:
J
u
n
2
,
2
0
2
6
Acc
ep
ted
:
J
u
n
2
7
,
2
0
2
6
Th
is
stu
d
y
i
n
v
e
stig
a
tes
wh
e
th
e
r
o
v
e
rsa
m
p
li
n
g
is
a
re
li
a
b
le
so
l
u
ti
o
n
fo
r
se
v
e
re
c
las
s
imb
a
lan
c
e
in
Ara
b
i
c
d
iale
c
t
i
d
e
n
ti
f
ica
ti
o
n
.
Us
in
g
t
h
e
S
h
a
m
i
Co
rp
u
s
a
s
a
c
o
n
tro
ll
e
d
tes
t
b
e
d
,
we
d
e
m
o
n
stra
te
th
a
t
c
o
n
v
e
n
ti
o
n
a
l
o
v
e
rsa
m
p
li
n
g
o
fte
n
fa
il
s
in
h
ig
h
-
d
ime
n
sio
n
a
l
s
p
a
rse
tex
t
sp
a
c
e
s,
b
u
t
d
e
n
sit
y
-
b
a
se
d
c
lu
ste
r
fil
terin
g
c
a
n
e
ffe
c
ti
v
e
ly
re
so
lv
e
th
is.
We co
n
d
u
c
t
a
c
o
m
p
a
ra
ti
v
e
e
v
a
lu
a
ti
o
n
o
f
S
M
OTE
,
c
lu
ste
ri
n
g
-
g
u
i
d
e
d
v
a
rian
ts
(AST
RA
-
S
M
OTE
a
n
d
S
M
OTE
-
RAD
IAN
T),
a
n
d
a
c
o
st
-
se
n
siti
v
e
Clas
sWe
ig
h
t
a
p
p
r
o
a
c
h
u
n
d
e
r
a
n
id
e
n
ti
c
a
l
5
,
6
4
4
-
d
ime
n
sio
n
a
l
fe
a
t
u
re
-
e
n
g
i
n
e
e
rin
g
p
i
p
e
li
n
e
u
sin
g
Li
g
h
tG
BM
a
n
d
XG
Bo
o
st.
On
th
e
h
e
l
d
-
o
u
t
tes
t
se
t,
sta
n
d
a
rd
S
M
OT
E
a
n
d
c
las
s
we
ig
h
ti
n
g
fre
q
u
e
n
tl
y
d
ist
o
rted
d
e
c
isio
n
b
o
u
n
d
a
ries
,
y
ield
i
n
g
i
n
c
o
n
siste
n
t
g
a
in
s
a
c
ro
ss
m
o
d
e
ls.
In
c
o
n
t
ra
st,
S
M
OTE
-
RAD
IAN
T
y
ield
s
a
st
a
ti
stica
ll
y
sig
n
ifi
c
a
n
t
m
a
c
ro
-
F
1
imp
r
o
v
e
m
e
n
t
fo
r
Li
g
h
tG
BM
(0
.
8
5
3
9
v
s.
0
.
8
5
2
6
o
n
th
e
o
rig
i
n
a
l
d
a
ta)
with
a
larg
e
e
ffe
c
t
siz
e
(r
=
0
.
5
1
1
),
su
c
c
e
ss
fu
ll
y
re
sc
u
in
g
m
in
o
rit
y
d
iale
c
ts
with
o
u
t
d
e
g
ra
d
in
g
t
h
e
m
a
jo
rit
y
c
las
s.
T
h
e
se
fin
d
in
g
s
su
g
g
e
st
t
h
a
t
wh
il
e
o
v
e
rsa
m
p
li
n
g
is
n
o
t
u
n
iv
e
rsa
ll
y
re
li
a
b
le
in
s
p
a
rse
tex
t
sp
a
c
e
s,
c
o
u
p
li
n
g
it
with
d
e
n
sit
y
-
b
a
se
d
n
o
ise
n
e
u
traliza
ti
o
n
(R
AD
IAN
T)
p
ro
v
id
e
s
a
ro
b
u
st
a
n
d
i
n
terp
re
tab
l
e
a
lt
e
rn
a
ti
v
e
to
d
e
e
p
lea
rn
i
n
g
m
o
d
e
ls.
Th
i
s
stu
d
y
p
r
o
v
i
d
e
s
m
e
th
o
d
o
l
o
g
ica
l
c
l
a
rit
y
a
n
d
re
p
ro
d
u
c
i
b
le
g
u
id
a
n
c
e
f
o
r
fa
ir
a
n
d
in
c
lu
siv
e
Ara
b
ic NLP
sy
ste
m
s.
K
ey
w
o
r
d
s
:
Ar
ab
ic
d
ialec
t id
en
tific
atio
n
C
las
s
im
b
alan
ce
Featu
r
e
en
g
in
ee
r
i
n
g
Ov
er
s
am
p
lin
g
m
eth
o
d
s
SMOT
E
T
ex
t c
lass
if
icatio
n
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Ar
if
Nu
r
wid
y
an
t
o
r
o
Dep
ar
tm
en
t o
f
C
o
m
p
u
ter
Scie
n
ce
an
d
E
lectr
o
n
ics,
Facu
lty
o
f
Ma
th
em
atics a
n
d
Natu
r
al
Sci
en
ce
s
Un
iv
er
s
itas
Gad
jah
Ma
d
a,
Yo
g
y
ak
ar
ta,
I
n
d
o
n
esia
E
m
ail:
ar
if
n
@
u
g
m
.
ac
.
id
1.
I
NT
RO
D
UCT
I
O
N
Ar
ab
ic
is
s
p
o
k
e
n
b
y
o
v
e
r
4
0
0
m
illi
o
n
p
eo
p
le
ac
r
o
s
s
th
e
Mid
d
le
E
ast
an
d
No
r
th
Af
r
ic
a,
an
d
its
r
eg
io
n
al
d
ialec
ts
ex
h
ib
it
s
u
b
s
tan
tial
v
ar
iatio
n
in
v
o
ca
b
u
lar
y
,
m
o
r
p
h
o
lo
g
y
,
an
d
s
y
n
tax
[
1
]
.
Acc
u
r
ate
d
ialec
t
id
en
tific
atio
n
is
th
er
e
f
o
r
e
ess
en
tial
f
o
r
d
o
wn
s
tr
ea
m
n
a
tu
r
al
lan
g
u
a
g
e
p
r
o
ce
s
s
in
g
(
NL
P)
a
p
p
licatio
n
s
s
u
ch
as
s
p
ee
ch
r
ec
o
g
n
itio
n
,
m
ac
h
i
n
e
tr
an
s
latio
n
,
an
d
s
en
tim
en
t
a
n
aly
s
is
[
2
]
.
Sev
er
al
b
en
ch
m
ar
k
co
r
p
o
r
a
h
av
e
b
ee
n
d
ev
elo
p
e
d
f
o
r
th
is
task
,
in
clu
d
in
g
n
u
an
ce
d
Ar
ab
ic
d
ialec
t
id
en
tific
atio
n
(
NADI
)
[
3
]
,
M
u
lti
-
Ar
ab
ic
d
ialec
t
ap
p
licatio
n
s
an
d
r
eso
u
r
ce
s
(
MA
DAR
)
[
4
]
,
an
d
d
ialec
tal
Ar
ab
ic
r
eso
u
r
ce
o
f
twee
ts
(
DART
)
[
5
]
.
Am
o
n
g
th
em
,
th
e
Sh
am
i
C
o
r
p
u
s
[
6
]
p
r
o
v
id
e
s
a
f
o
cu
s
ed
test
b
ed
o
n
f
o
u
r
L
e
v
an
tin
e
d
ialec
ts
-
J
o
r
d
an
ian
,
L
e
b
an
ese,
Palest
in
ian
,
an
d
Sy
r
ian
-
m
ak
in
g
it p
ar
ticu
la
r
ly
s
u
itab
le
f
o
r
an
aly
zin
g
class
im
b
alan
ce
ef
f
ec
ts
.
Dialec
t
id
en
tific
atio
n
is
ch
alle
n
g
in
g
b
ec
a
u
s
e
d
ialec
ts
d
if
f
er
m
ar
k
ed
ly
f
r
o
m
Mo
d
er
n
Stan
d
ar
d
Ar
ab
ic
(
MSA)
an
d
ar
e
p
r
im
ar
ily
o
b
s
er
v
ed
in
in
f
o
r
m
al
s
o
cial
m
ed
ia
tex
t
[
7
]
-
[
1
0
]
.
T
witter
d
ata
f
u
r
th
er
in
tr
o
d
u
c
e
n
o
is
e
d
u
e
to
s
h
o
r
t
m
ess
ag
e
len
g
th
,
o
r
t
h
o
g
r
ap
h
ic
v
a
r
iatio
n
,
em
o
jis
,
h
ash
tag
s
,
an
d
c
o
d
e
-
s
witch
in
g
[
1
1
]
.
A
m
ajo
r
d
if
f
icu
lty
o
f
th
e
Sh
am
i
C
o
r
p
u
s
is
its
s
k
ewe
d
class
d
is
tr
ib
u
tio
n
:
Sy
r
ian
twee
ts
ac
co
u
n
t
f
o
r
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
4
3
,
No
.
1
,
Ju
ly
20
2
6
:
259
-
27
0
260
ap
p
r
o
x
im
ately
5
7
%
o
f
t
h
e
d
ata,
wh
er
ea
s
J
o
r
d
an
ia
n
r
e
p
r
e
s
en
ts
o
n
ly
1
0
.
6
%
[
6
]
.
Su
ch
im
b
alan
ce
b
iases
class
if
ier
s
to
war
d
th
e
m
ajo
r
ity
class
an
d
d
eg
r
ad
es
r
ec
o
g
n
itio
n
o
f
m
i
n
o
r
ity
d
ialec
ts
[
1
2
]
.
H
o
wev
er
,
it
r
em
ai
n
s
u
n
d
er
e
x
p
lo
r
e
d
h
o
w
co
m
m
o
n
l
y
u
s
ed
im
b
alan
ce
-
h
an
d
lin
g
s
tr
ateg
ies
-
r
an
g
in
g
f
r
o
m
g
en
e
r
ativ
e
o
v
er
s
am
p
lin
g
to
co
s
t
-
s
en
s
itiv
e
weig
h
tin
g
-
p
er
f
o
r
m
in
h
ig
h
ly
s
p
ar
s
e,
h
i
g
h
-
d
im
en
s
io
n
al
tex
t f
ea
tu
r
e
s
p
ac
es.
Featu
r
e
-
b
ased
r
e
p
r
esen
tatio
n
s
co
n
tin
u
e
to
p
lay
a
ce
n
tr
al
r
o
l
e
in
Ar
ab
ic
d
ialec
t
id
e
n
tific
atio
n
d
u
e
to
th
eir
in
ter
p
r
etab
ilit
y
a
n
d
co
m
p
u
tatio
n
al
ef
f
icien
cy
.
L
e
x
i
ca
l
f
ea
tu
r
es
e
f
f
ec
tiv
ely
ca
p
t
u
r
e
d
is
cr
im
in
ativ
e
v
o
ca
b
u
lar
y
p
atter
n
s
[
1
2
]
-
[
1
4
]
,
wh
ile
lex
ico
n
-
b
ased
an
d
ce
n
tr
o
id
-
s
im
ilar
ity
f
ea
tu
r
es
p
r
o
v
i
d
e
co
m
p
le
m
en
tar
y
d
ialec
tal
cu
es
[
3
]
,
[
1
5
]
-
[
1
9
]
.
Alth
o
u
g
h
r
ec
en
t
s
tu
d
ies
in
cr
e
asin
g
ly
r
ely
o
n
tr
an
s
f
o
r
m
er
-
b
ased
m
o
d
els
s
u
ch
as
Ar
aBER
T
,
AR
B
E
R
T
,
an
d
M
AR
B
E
R
T
[
2
0
]
-
[
2
3
]
,
th
ese
m
o
d
els
ar
e
c
o
m
p
u
tatio
n
ally
e
x
p
en
s
iv
e
an
d
r
a
r
ely
f
ac
ilit
ate
s
y
s
tem
atic
an
aly
s
is
o
f
im
b
alan
ce
-
h
an
d
lin
g
m
ec
h
a
n
is
m
s
.
L
ig
h
tweig
h
t
f
ea
tu
r
e
-
e
n
g
in
ee
r
ed
p
ip
elin
es
th
er
ef
o
r
e
r
em
ain
v
al
u
ab
le
f
o
r
tr
an
s
p
ar
en
t
in
v
esti
g
atio
n
o
f
m
et
h
o
d
o
lo
g
ical
ef
f
ec
ts
u
n
d
e
r
f
i
x
ed
r
ep
r
esen
tatio
n
s
an
d
ev
alu
atio
n
p
r
o
to
co
ls
[
2
4
]
-
[
2
7
]
.
T
o
m
itig
ate
class
im
b
alan
ce
,
o
v
er
s
am
p
lin
g
tech
n
iq
u
es
s
u
ch
as
th
e
s
y
n
th
etic
m
in
o
r
ity
o
v
e
r
-
s
am
p
lin
g
T
ec
h
n
iq
u
e
(
SMOT
E
)
ar
e
wid
e
ly
ad
o
p
te
d
[
2
8
]
,
[
2
9
]
.
Ho
wev
e
r
,
in
h
ig
h
-
d
im
e
n
s
io
n
al
s
p
ac
es
(
e.
g
.
,
th
o
u
s
an
d
s
o
f
ex
tr
ac
ted
lex
ical
an
d
m
o
r
p
h
o
lo
g
ical
f
ea
tu
r
es),
s
tan
d
a
r
d
SMOT
E
m
ay
in
tr
o
d
u
ce
n
o
i
s
y
o
r
o
v
er
la
p
p
in
g
in
s
tan
ce
s
th
at
d
is
to
r
t
class
b
o
u
n
d
a
r
ies.
C
lu
s
ter
in
g
-
g
u
id
e
d
v
ar
ian
ts
,
s
u
ch
as
KM
ea
n
s
a
n
d
DB
SC
AN
-
b
ased
f
ilter
s
,
attem
p
t
to
r
estric
t
s
y
n
th
etic
g
en
er
atio
n
t
o
m
in
o
r
ity
-
d
en
s
e
r
eg
io
n
s
o
r
f
ilter
o
u
tli
er
s
[
2
5
]
,
[
3
0
]
-
[
3
4
]
.
Desp
ite
th
eir
p
o
p
u
lar
ity
,
r
ep
o
r
ted
f
in
d
in
g
s
o
n
th
ei
r
ef
f
ec
tiv
en
ess
in
Ar
ab
ic
d
ialec
t
id
en
tific
atio
n
r
em
ain
in
co
n
s
is
ten
t,
lar
g
ely
d
u
e
to
v
ar
iatio
n
s
in
e
v
alu
atio
n
p
r
o
t
o
co
ls
an
d
th
e
lack
o
f
co
m
p
ar
is
o
n
ag
ain
s
t
n
o
n
-
g
en
er
ativ
e
co
s
t
-
s
en
s
itiv
e
lear
n
in
g
.
T
o
ad
d
r
ess
th
is
g
ap
,
th
e
p
r
e
s
en
t
s
tu
d
y
p
r
o
v
id
es
th
e
f
ir
s
t
co
n
tr
o
lled
ev
alu
atio
n
o
f
o
v
er
s
am
p
lin
g
m
eth
o
d
s
f
o
r
Ar
ab
ic
d
ialec
t
id
en
tific
atio
n
th
at
s
y
s
tem
atic
ally
co
m
p
ar
es
SMOT
E
an
d
clu
s
ter
in
g
-
g
u
id
e
d
v
ar
ian
ts
(
ASTRA
-
SMOT
E
a
n
d
SMOT
E
-
R
ADI
ANT
)
alo
n
g
s
id
e
a
co
s
t
-
s
en
s
itiv
e
C
las
s
W
eig
h
t
co
m
p
ar
ato
r
u
n
d
er
an
id
en
tical
f
ea
tu
r
e
-
en
g
in
ee
r
in
g
p
ip
elin
e.
B
y
f
i
x
in
g
t
h
e
c
o
r
p
u
s
,
p
r
e
p
r
o
ce
s
s
in
g
s
tep
s
,
f
ea
tu
r
e
r
ep
r
esen
tatio
n
,
a
n
d
ev
alu
ati
o
n
p
r
o
to
c
o
l,
th
is
wo
r
k
ef
f
ec
tiv
ely
d
is
en
tan
g
les
th
e
im
p
ac
t
o
f
i
m
b
alan
ce
h
an
d
lin
g
f
r
o
m
lear
n
er
-
s
p
ec
if
ic
in
d
u
ctiv
e
b
iases
in
s
p
ar
s
e
tex
t sp
ac
es.
Sp
ec
if
ically
,
t
h
is
s
t
u
d
y
co
n
t
r
i
b
u
tes
b
y
:
(
1
)
c
o
n
d
u
cti
n
g
a
u
n
if
i
ed
c
o
m
p
a
r
at
iv
e
ev
al
u
a
ti
o
n
i
n
a
h
i
g
h
l
y
s
p
a
r
s
e
5
,
6
4
4
-
d
im
en
s
io
n
a
l
f
ea
t
u
r
e
s
p
ac
e
u
s
i
n
g
L
i
g
h
tGB
M
a
n
d
XGB
o
o
s
t;
(
2
)
a
p
p
l
y
i
n
g
r
ig
o
r
o
u
s
b
o
o
ts
tr
a
p
p
in
g
s
tatis
t
ic
al
v
ali
d
at
io
n
t
o
m
e
asu
r
e
e
f
f
ec
t
s
i
ze
s
o
n
a
h
el
d
-
o
u
t
t
est
s
et;
(
3
)
ag
g
r
e
g
a
ti
n
g
f
e
at
u
r
e
i
m
p
o
r
t
an
ce
at
t
h
e
b
l
o
c
k
le
v
e
l
t
o
i
n
t
er
p
r
et
t
h
e
m
o
d
els'
d
ec
is
i
o
n
-
m
a
k
i
n
g
p
r
o
ce
s
s
es;
an
d
(
4
)
d
e
m
o
n
s
tr
ati
n
g
th
at
d
e
n
s
it
y
-
b
ase
d
f
ilt
e
r
i
n
g
(
R
AD
I
AN
T
)
s
ig
n
i
f
i
ca
n
tl
y
o
u
tp
e
r
f
o
r
m
s
co
n
v
e
n
t
io
n
a
l
SMOT
E
a
n
d
cl
ass
wei
g
h
ti
n
g
.
R
ath
er
th
a
n
m
e
r
el
y
p
r
o
p
o
s
i
n
g
a
n
ew
al
g
o
r
i
th
m
,
t
h
is
wo
r
k
p
r
o
v
i
d
es
m
et
h
o
d
o
l
o
g
ic
al
cl
ar
it
y
b
y
i
d
e
n
ti
f
y
i
n
g
wh
e
n
o
v
e
r
s
am
p
li
n
g
y
ie
ld
s
s
t
ab
le
m
in
o
r
it
y
im
p
r
o
v
e
m
e
n
ts
a
n
d
w
h
en
it
d
e
g
r
a
d
es
p
e
r
f
o
r
m
a
n
ce
d
u
e
to
l
ea
r
n
e
r
-
f
e
at
u
r
e
i
n
te
r
a
cti
o
n
s
.
T
h
e
r
em
ai
n
d
er
o
f
th
is
p
a
p
er
is
o
r
g
a
n
ized
as
f
o
llo
ws.
S
ec
tio
n
2
d
escr
ib
es
th
e
d
atas
et,
f
ea
tu
r
e
r
ep
r
esen
tatio
n
s
,
r
e
b
alan
cin
g
s
tr
ateg
ies,
lear
n
i
n
g
m
o
d
els,
an
d
e
v
alu
atio
n
p
r
o
to
co
l.
Sectio
n
3
p
r
esen
ts
th
e
ex
p
er
im
en
tal
r
esu
lts
,
s
tatis
tica
l
an
aly
s
is
,
an
d
in
ter
p
r
etativ
e
d
is
cu
s
s
io
n
o
f
f
in
d
in
g
s
.
Sectio
n
4
co
n
cl
u
d
es
with
k
ey
f
in
d
in
g
s
,
lim
itatio
n
s
,
an
d
d
ir
ec
tio
n
s
f
o
r
f
u
tu
r
e
r
esear
ch
.
2.
M
E
T
H
O
D
2
.1
.
P
r
o
blem
s
et
up
a
nd
da
t
a
W
e
i
n
v
e
s
t
i
g
a
t
e
m
u
l
ti
-
c
l
as
s
Ar
a
b
i
c
d
i
a
le
c
t
i
d
e
n
t
i
f
i
c
a
ti
o
n
u
n
d
e
r
s
e
v
e
r
e
i
m
b
a
l
a
n
ce
u
s
i
n
g
th
e
S
h
a
m
i
C
o
r
p
u
s
[
6
]
,
c
o
m
p
r
is
i
n
g
6
6
,
2
4
7
t
w
e
e
ts
a
c
r
o
s
s
f
o
u
r
L
e
v
a
n
t
i
n
e
d
i
a
l
e
ct
s
T
a
b
l
e
1
.
U
s
e
d
a
s
r
e
l
e
a
s
e
d
t
o
p
r
e
s
e
r
v
e
r
e
a
l
i
s
t
i
c
n
o
is
e
a
n
d
b
i
as
e
s
[
6
]
,
th
e
co
r
p
u
s
i
s
h
e
a
v
il
y
s
k
e
w
e
d
:
S
y
r
i
a
n
t
w
ee
t
s
d
o
m
i
n
at
e
(
5
7
.
0
%
)
,
w
h
i
l
e
J
o
r
d
a
n
i
a
n
a
c
c
o
u
n
t
s
f
o
r
o
n
l
y
1
0
.
6
%
.
T
o
p
r
e
v
e
n
t
d
a
t
a
l
e
a
k
a
g
e
,
w
e
s
p
l
i
t
t
h
e
d
a
t
a
i
n
t
o
a
h
e
l
d
-
o
u
t
t
e
s
t
s
e
t
(
2
5
%
)
a
n
d
a
t
r
a
i
n
i
n
g
s
e
t
(
7
5
%
)
,
s
u
b
d
i
v
i
d
i
n
g
t
h
e
l
att
e
r
(
8
0
/
2
0
)
f
o
r
m
o
d
e
l
f
i
t
t
i
n
g
a
n
d
v
a
l
i
d
a
t
i
o
n
.
O
u
r
f
i
v
e
-
s
t
a
g
e
p
i
p
e
l
i
n
e
F
i
g
u
r
e
1
e
n
c
o
m
p
a
s
s
es
t
e
x
t
p
r
e
p
r
o
c
e
s
s
i
n
g
,
h
i
g
h
-
d
i
m
e
n
s
i
o
n
a
l
f
e
a
t
u
r
e
e
n
g
i
n
e
e
r
i
n
g
,
i
m
b
a
l
a
n
c
e
h
a
n
d
l
i
n
g
(
o
r
i
g
i
n
a
l
,
d
a
t
a
-
l
e
v
e
l
o
v
e
r
s
a
m
p
l
i
n
g
,
a
n
d
a
l
g
o
r
i
t
h
m
-
l
ev
e
l
w
e
i
g
h
t
i
n
g
)
,
g
r
a
d
i
e
n
t
b
o
o
s
t
in
g
t
r
a
i
n
i
n
g
,
a
n
d
b
o
o
t
s
t
r
a
p
s
t
a
t
is
t
i
c
a
l
e
v
al
u
a
t
i
o
n
.
2
.
2
.
P
re
pro
ce
s
s
ing
T
o
r
e
d
u
c
e
o
r
th
o
g
r
a
p
h
ic
v
ar
iat
io
n
,
we
ap
p
ly
lig
h
t
n
o
r
m
aliza
tio
n
(
r
em
o
v
al
o
f
d
iacr
itics
an
d
tatwee
l)
an
d
ag
g
r
ess
iv
e
n
o
r
m
aliza
tio
n
(
e.
g
.
,
ه
→
ة
,
ي
→
ى
,
ا
→
إ/
أ/
آ
)
.
No
n
-
lin
g
u
is
tic
ar
tifa
cts
s
u
ch
a
s
UR
L
s
,
m
en
tio
n
s
,
h
ash
tag
s
,
d
ig
its
,
an
d
s
y
m
b
o
ls
ar
e
r
em
o
v
ed
u
s
in
g
r
e
g
u
l
ar
ex
p
r
ess
io
n
s
.
Sto
p
wo
r
d
h
an
d
lin
g
is
ap
p
lied
s
elec
tiv
ely
u
s
in
g
a
co
m
p
r
eh
en
s
iv
e
Ar
ab
ic
s
to
p
wo
r
d
lis
t
co
m
b
in
e
d
with
to
p
ic
-
s
p
ec
if
ic
s
to
p
wo
r
d
s
(
e.
g
.
,
co
u
n
tr
y
an
d
city
n
am
es lik
e
"
ن
ام
ع
"
,
"ن
ان
ب
ل" ,
"اير
و
س
")
to
p
r
ev
en
t
th
e
m
o
d
el
f
r
o
m
ex
p
lo
itin
g
g
e
o
g
r
ap
h
ical
b
iases
.
2
.
3
.
F
e
a
t
ure
e
ng
ineerin
g
W
e
e
n
g
i
n
e
e
r
e
d
a
co
m
p
r
eh
en
s
iv
e,
m
u
lti
-
b
lo
ck
f
ea
t
u
r
e
r
ep
r
e
s
en
tatio
n
co
m
p
r
is
in
g
5
,
6
4
4
d
i
m
en
s
io
n
s
.
T
h
is
ex
tr
em
e
d
im
en
s
io
n
al
ity
r
ef
lects
th
e
in
h
e
r
en
t
s
p
ar
s
ity
o
f
te
x
t
d
ata,
in
ten
tio
n
ally
te
s
tin
g
th
e
lim
its
o
f
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
E
va
lu
a
tin
g
o
ve
r
s
a
mp
lin
g
meth
o
d
s
fo
r
imb
a
la
n
ce
d
A
r
a
b
ic
d
ia
lect
id
en
tifi
ca
tio
n
(
Ma
u
la
n
a
I
h
s
a
n
A
h
ma
d
)
261
s
tan
d
ar
d
o
v
e
r
s
am
p
lin
g
alg
o
r
ith
m
s
an
d
ju
s
tify
in
g
th
e
n
ee
d
f
o
r
clu
s
ter
-
b
ased
f
ilte
r
in
g
.
T
o
m
an
ag
e
d
im
en
s
io
n
ality
an
d
n
o
is
e,
ad
a
p
tiv
e
f
ea
tu
r
e
s
elec
tio
n
was
ap
p
lied
.
L
ex
ical
wo
r
d
f
ea
tu
r
es
(
T
F
-
I
DF
an
d
B
M2
5
)
wer
e
f
ilter
ed
u
s
in
g
th
e
χ²
(
C
h
i
-
s
q
u
ar
e)
test
,
wh
ile
ch
ar
ac
ter
-
lev
el
n
-
g
r
am
s
wer
e
s
tr
ictly
f
ilter
ed
u
s
in
g
m
u
tu
al
in
f
o
r
m
atio
n
(
MI
)
.
Ad
d
itio
n
al
ly
,
o
u
t
-
of
-
f
o
l
d
(
OOF)
s
tack
in
g
p
r
o
b
ab
ilit
ies
f
r
o
m
L
o
g
is
tic
R
eg
r
ess
io
n
an
d
L
in
ea
r
SVC
wer
e
g
en
er
ated
u
s
in
g
a
5
-
f
o
ld
s
tr
atif
ied
c
r
o
s
s
-
v
alid
atio
n
.
T
o
en
s
u
r
e
s
tab
ilit
y
an
d
p
r
ev
e
n
t
d
ata
leak
ag
e,
th
ese
OOF
p
r
ed
icti
o
n
s
wer
e
a
v
er
ag
e
d
ac
r
o
s
s
two
d
if
f
e
r
en
t
r
an
d
o
m
s
ee
d
s
i
n
ter
n
ally
with
in
th
e
tr
ain
in
g
s
et.
T
h
e
c
o
m
p
lete
f
u
n
ctio
n
al
b
lo
ck
s
ar
e
s
u
m
m
ar
ized
in
T
ab
le
2
.
T
ab
le
1
.
Data
s
et
d
is
tr
ib
u
tio
n
i
n
th
e
s
h
am
i d
ialec
t c
o
r
p
u
s
D
i
a
l
e
c
t
Jo
r
d
a
n
Le
b
a
n
o
n
P
a
l
e
s
t
i
n
e
S
y
r
i
a
To
t
a
l
Tw
e
e
t
s
7
,
0
1
7
1
0
,
8
2
9
1
0
,
6
4
2
3
7
,
7
5
9
6
6
,
2
4
7
P
e
r
c
e
n
t
a
g
e
(
%)
1
0
.
6
1
6
.
3
1
6
.
1
5
7
.
0
1
0
0
Fig
u
r
e
1
.
Sy
s
tem
o
v
er
v
iew
o
f
th
e
p
r
o
p
o
s
ed
p
ip
elin
e
T
ab
le
2
.
Featu
r
e
b
lo
ck
s
s
u
m
m
ar
y
B
l
o
c
k
D
i
me
n
si
o
n
D
e
scri
p
t
i
o
n
Le
x
i
c
a
l
(
TF
-
I
D
F
&
B
M
2
5
)
2
4
2
0
W
o
r
d
a
n
d
c
h
a
r
_
w
b
n
-
g
r
a
ms,
a
g
g
r
e
ssi
v
e
l
y
f
i
l
t
e
r
e
d
u
s
i
n
g
χ
² se
l
e
c
t
i
o
n
t
o
r
e
t
a
i
n
t
h
e
mo
s
t
d
i
s
c
r
i
m
i
n
a
t
i
v
e
l
e
x
i
c
a
l
p
a
t
t
e
r
n
s.
C
h
a
r
a
c
t
e
r
N
-
g
r
a
ms
(
M
I
)
3
,
1
5
0
C
h
a
r
a
c
t
e
r
n
-
g
r
a
ms (
si
z
e
s
2
-
3
a
n
d
4
-
5
)
st
r
i
c
t
l
y
f
i
l
t
e
r
e
d
u
s
i
n
g
M
u
t
u
a
l
I
n
f
o
r
mat
i
o
n
(
M
I
)
t
o
c
a
p
t
u
r
e
s
u
b
-
w
o
r
d
d
i
a
l
e
c
t
m
a
r
k
e
r
s
.
S
t
r
u
c
t
u
r
a
l
& St
a
t
s
9
D
o
c
u
me
n
t
l
e
n
g
t
h
,
t
o
k
e
n
c
o
u
n
t
s,
t
y
p
e
-
t
o
k
e
n
r
a
t
i
o
,
a
n
d
s
t
o
p
w
o
r
d
st
a
t
i
s
t
i
c
s.
M
o
r
p
h
o
l
o
g
i
c
a
l
S
i
g
n
a
l
s
34
S
h
a
m
i
-
sp
e
c
i
f
i
c
p
r
e
f
i
x
e
s/
s
u
f
f
i
x
e
s (
e
.
g
.
,
ا
م
,
م
ع
,
ب
)
,
e
x
p
l
i
c
i
t
r
u
l
e
s
,
a
n
d
mo
r
p
h
o
l
o
g
i
c
a
l
w
o
r
d
i
n
t
e
r
a
c
t
i
o
n
s.
D
i
st
r
i
b
u
t
i
o
n
a
l
P
r
i
o
r
s
5
C
o
l
l
o
c
a
t
i
o
n
d
o
mi
n
a
n
c
e
s
c
o
r
e
s
a
n
d
N
B
-
S
V
M
l
o
g
-
c
o
u
n
t
r
a
t
i
o
s.
O
O
F
S
t
a
c
k
i
n
g
&
I
n
t
e
r
a
c
t
i
o
n
s
26
O
u
t
-
of
-
f
o
l
d
(
O
O
F
)
p
r
e
d
i
c
t
i
o
n
p
r
o
b
a
b
i
l
i
t
i
e
s
f
r
o
m
Lo
g
i
st
i
c
R
e
g
r
e
ss
i
o
n
a
n
d
Li
n
e
a
r
S
V
C
,
c
o
m
b
i
n
e
d
w
i
t
h
t
h
e
i
r
i
n
t
e
r
a
c
t
i
o
n
t
e
r
ms
(
p
r
o
d
u
c
t
,
d
i
f
f
,
ma
x
)
.
2
.
4
.
I
m
ba
l
a
nce
ha
nd
lin
g
s
t
r
a
t
eg
ies
Fiv
e
im
b
alan
ce
-
h
an
d
lin
g
s
ce
n
ar
io
s
ar
e
ev
alu
ate
d
T
a
b
le
3
.
R
esam
p
lin
g
is
ex
p
licitly
ex
ec
u
ted
ex
clu
s
iv
ely
o
n
th
e
s
u
b
-
tr
ain
i
n
g
d
ata
p
r
io
r
to
class
if
ier
tr
ain
in
g
,
k
ee
p
in
g
th
e
v
alid
atio
n
a
n
d
test
s
ets co
m
p
letely
r
ep
r
esen
tativ
e
o
f
th
e
r
ea
l
-
wo
r
ld
d
is
tr
ib
u
tio
n
.
T
o
p
r
ev
en
t
ex
ce
s
s
iv
e
n
o
is
e
g
en
er
atio
n
,
th
e
s
am
p
lin
g
_
s
tr
ateg
y
f
o
r
all
g
en
er
ativ
e
m
eth
o
d
s
was
s
tr
ict
ly
tu
n
ed
v
ia
g
r
id
s
ea
r
ch
,
ca
p
p
in
g
s
y
n
th
etic
tar
g
ets
to
ap
p
r
o
x
im
ately
7
,
8
9
9
s
am
p
les
f
o
r
J
o
r
d
a
n
ian
,
an
d
1
9
,
4
0
0
f
o
r
L
eb
an
ese
an
d
Palest
in
ian
.
ASTRA
-
SMOT
E
in
tr
o
d
u
ce
s
a
p
r
e
-
f
ilter
in
g
s
tag
e,
r
estrictin
g
s
y
n
th
esis
ex
clu
s
iv
ely
to
m
in
o
r
ity
clu
s
ter
s
ex
ce
ed
i
n
g
a
5
0
%
p
u
r
ity
t
h
r
esh
o
ld
.
C
o
n
v
er
s
ely
,
SMOT
E
-
R
ADI
ANT
ac
ts
as
a
p
o
s
t
-
f
ilter
;
it
allo
ws
s
tan
d
ar
d
SMOT
E
g
en
er
atio
n
b
u
t
s
u
b
s
eq
u
e
n
tly
ap
p
lies
DB
SC
AN
ex
clu
s
iv
ely
t
o
th
e
s
y
n
th
etic
d
ata
(
clea
n
_
o
n
l
y
_
s
y
n
th
=
T
r
u
e)
to
n
eu
tr
alize
o
v
er
lap
p
in
g
n
o
is
e
at
in
ter
-
class
d
ec
is
i
o
n
b
o
u
n
d
ar
ies.
T
h
e
C
las
s
W
eig
h
t
s
ce
n
ar
io
is
in
clu
d
ed
as
a
n
o
n
-
g
en
er
ativ
e
co
m
p
ar
ato
r
to
ass
ess
wh
eth
er
p
er
f
o
r
m
an
ce
s
h
if
ts
ar
is
e
f
r
o
m
d
ata
au
g
m
e
n
tatio
n
o
r
m
e
r
ely
lo
s
s
r
ewe
ig
h
tin
g
.
Fig
u
r
e
2
illu
s
tr
ates
th
e
wo
r
k
f
lo
ws
o
f
th
e
ev
alu
ate
d
o
v
er
s
am
p
lin
g
m
et
h
o
d
s
Fig
u
r
e
2
(
a)
SMOT
E
,
Fig
u
r
e
2
(
b
)
ASTRA
-
SMOT
E
,
an
d
Fig
u
r
e
2
(
c)
SMOT
E
-
R
ADI
ANT
.
W
h
ile
all
m
eth
o
d
s
g
en
er
ate
a
b
alan
ce
d
tr
ain
in
g
d
ataset,
ASTRA
-
SMOT
E
in
tr
o
d
u
ce
s
clu
s
ter
in
g
b
ef
o
r
e
o
v
er
s
am
p
lin
g
,
a
n
d
SMOT
E
-
R
ADI
ANT
ad
d
itio
n
ally
r
em
o
v
es n
o
is
y
s
y
n
th
etic
s
am
p
les.
T
ab
le
3
.
I
m
b
alan
ce
h
an
d
lin
g
s
tr
ateg
ies ev
alu
ated
S
c
e
n
a
r
i
o
D
e
scri
p
t
i
o
n
K
e
y
p
a
r
a
me
t
e
r
s
O
r
i
g
i
n
a
l
U
ses
o
r
i
g
i
n
a
l
i
m
b
a
l
a
n
c
e
d
d
a
t
a
w
i
t
h
o
u
t
r
e
sa
mp
l
i
n
g
.
-
S
M
O
TE
I
n
t
e
r
p
o
l
a
t
e
s
mi
n
o
r
i
t
y
s
a
m
p
l
e
s
c
o
n
v
e
n
t
i
o
n
a
l
l
y
.
k
-
n
e
i
g
h
b
o
r
s =
5
A
S
TR
A
-
S
M
O
TE
A
p
p
l
i
e
s
M
i
n
i
B
a
t
c
h
K
M
e
a
n
s
c
l
u
st
e
r
i
n
g
;
g
e
n
e
r
a
t
e
s s
y
n
t
h
e
t
i
c
d
a
t
a
s
t
r
i
c
t
l
y
i
n
si
d
e
"
safe"
m
i
n
o
r
i
t
y
-
d
e
n
se
c
l
u
s
t
e
r
s.
n
_
c
l
u
st
e
r
s =
1
2
8
,
s
a
f
e
_
t
h
r
e
s
h
o
l
d
=
0
.
5
,
a
l
p
h
a
_
m
o
d
e
=
f
u
l
l
,
k
-
n
e
i
g
h
b
o
r
s
=
5
S
M
O
TE
-
R
A
D
I
A
N
T
P
e
r
f
o
r
ms S
M
O
TE
f
o
l
l
o
w
e
d
b
y
d
e
n
s
i
t
y
-
b
a
se
d
f
i
l
t
e
r
i
n
g
(
D
B
S
C
A
N
)
t
o
r
e
m
o
v
e
o
v
e
r
l
a
p
p
i
n
g
s
y
n
t
h
e
t
i
c
n
o
i
se
a
t
c
l
a
ss
b
o
u
n
d
a
r
i
e
s
.
M
e
t
r
i
c
=
c
o
s
i
n
e
,
e
p
s
=
a
u
t
o
,
mi
n
_
sa
mp
l
e
s
=
a
u
t
o
,
c
l
e
a
n
_
o
n
l
y
_
s
y
n
t
h
=
Tr
u
e
C
l
a
s
sW
e
i
g
h
t
C
o
s
t
-
se
n
si
t
i
v
e
l
e
a
r
n
i
n
g
u
s
i
n
g
s
t
a
n
d
a
r
d
i
n
v
e
r
se
-
f
r
e
q
u
e
n
c
y
c
l
a
ss
w
e
i
g
h
t
s
.
c
l
a
ss
_
w
e
i
g
h
t
=
"
b
a
l
a
n
c
e
d
"
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
4
3
,
No
.
1
,
Ju
ly
20
2
6
:
259
-
27
0
262
Fig
u
r
e
2
.
Ov
e
r
s
am
p
lin
g
s
tr
ate
g
ies (
a)
SMOT
E
,
(
b
)
ASTRA
-
SMOT
E
,
an
d
(
c)
SMOT
E
-
R
ADI
ANT
2
.
5
.
M
o
dels
,
t
ra
ini
n
g
,
a
nd
ev
a
lua
t
io
n pro
t
o
c
o
l
W
e
em
p
lo
y
two
g
r
ad
ien
t
b
o
o
s
tin
g
class
if
ier
s
:
L
ig
h
tGB
M
(
leaf
-
wis
e
g
r
o
wth
)
an
d
XGBo
o
s
t
(
lev
el
-
wis
e
g
r
o
wth
)
.
Op
tim
al
o
v
er
s
am
p
lin
g
p
ar
am
ete
r
s
wer
e
s
ele
cted
b
y
m
ax
im
izin
g
th
e
h
a
r
m
o
n
ic
m
ea
n
o
f
b
o
th
m
o
d
els'
v
alid
atio
n
m
ac
r
o
-
F1
s
co
r
es.
C
la
s
s
if
ier
h
y
p
er
p
ar
am
eter
s
wer
e
s
u
b
s
eq
u
en
tly
tu
n
ed
:
L
ig
h
tGB
M
(
n
u
m
_
lea
v
es=2
5
,
co
ls
am
p
le_
b
y
tr
ee
=0
.
4
,
m
in
_
c
h
ild
_
s
am
p
l
es=5
0
,
L
1
/L2
r
eg
u
lar
izatio
n
=
1
.
0
)
an
d
XGBo
o
s
t
(
m
ax
_
d
e
p
th
=8
,
co
ls
am
p
le_
b
y
tr
ee
=0
.
3
,
m
in
_
ch
ild
_
weig
h
t=3
0
,
L
1
/L2
r
eg
u
lar
izatio
n
=
1
.
5
)
.
B
o
th
m
o
d
els
ut
ilized
a
0
.
0
1
lear
n
in
g
r
ate
an
d
1
5
0
ea
r
ly
s
to
p
p
in
g
r
o
u
n
d
s
.
T
o
en
s
u
r
e
s
tatis
tical
r
ig
o
r
o
n
th
e
h
eld
-
o
u
t
test
s
et
(
1
6
,
5
6
2
twee
ts
)
,
we
p
er
f
o
r
m
ed
1
,
0
0
0
b
o
o
ts
tr
ap
iter
atio
n
s
(
s
am
p
lin
g
with
r
ep
lace
m
en
t)
o
n
th
e
p
r
e
d
ictio
n
s
.
Pair
ed
two
-
s
id
ed
W
ilco
x
o
n
s
i
g
n
ed
-
r
an
k
t
ests
an
d
C
o
h
en
’
s
ef
f
ec
t
s
izes
(
r
)
wer
e
ap
p
lied
to
th
e
r
esu
ltin
g
Ma
cr
o
-
F1
d
is
tr
ib
u
tio
n
s
.
T
h
is
p
r
o
t
o
c
o
l
s
tr
ictly
ev
alu
ates
wh
eth
er
p
er
f
o
r
m
an
ce
s
h
if
ts
ar
e
s
tatis
t
ically
s
ig
n
if
ican
t
an
d
p
r
ac
tically
m
ea
n
in
g
f
u
l,
en
s
u
r
in
g
o
u
t
-
of
-
s
am
p
le
v
alid
ity
an
d
p
r
ev
e
n
tin
g
P
-
h
ac
k
in
g
.
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
3
.
1
.
P
er
f
o
r
m
a
nce
a
na
ly
s
is
T
ab
le
4
r
ep
o
r
ts
th
e
h
eld
-
o
u
t
test
-
s
et
p
er
f
o
r
m
an
ce
in
ter
m
s
o
f
m
ac
r
o
-
p
r
ec
is
io
n
,
m
ac
r
o
-
r
ec
all,
m
ac
r
o
-
F1
,
an
d
ac
cu
r
ac
y
f
o
r
th
e
ev
al
u
ated
s
tr
ateg
ies.
T
h
e
m
o
d
els
tr
ain
ed
o
n
th
e
o
r
ig
in
al
d
ata
alr
ea
d
y
ac
h
iev
e
v
e
r
y
h
ig
h
p
er
f
o
r
m
an
ce
d
u
e
to
th
e
r
o
b
u
s
t 5
,
6
4
4
-
d
im
e
n
s
io
n
al
f
ea
tu
r
e
en
g
in
ee
r
in
g
p
ip
elin
e.
L
ig
h
t
GB
M
an
d
XG
B
o
o
s
t
in
itially
ac
h
iev
ed
m
ac
r
o
-
F1
s
c
o
r
es o
f
0
.
8
5
2
6
an
d
0
.
8
5
1
7
,
r
es
p
ec
tiv
ely
(
b
o
o
ts
tr
ap
m
ea
n
s
: 0
.
8
5
2
5
a
n
d
0
.
8
5
1
6
)
.
T
ab
le
4
.
T
est
-
s
et
p
er
f
o
r
m
an
ce
o
f
L
ig
h
tGB
M
an
d
XGBo
o
s
t u
n
d
er
f
i
v
e
s
ce
n
ar
io
s
M
o
d
e
l
R
e
b
a
l
a
n
c
i
n
g
s
t
r
a
t
e
g
y
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
M
a
c
r
o
-
F1
A
c
c
u
r
a
c
y
Li
g
h
t
G
B
M
O
r
i
g
i
n
a
l
0
.
8
6
2
7
0
.
8
4
3
9
0
.
8
5
2
6
0
.
8
9
6
3
Li
g
h
t
G
B
M
S
M
O
TE
0
.
8
5
8
4
0
.
8
4
6
3
0
.
8
5
1
8
0
.
8
9
6
1
Li
g
h
t
G
B
M
A
S
TR
A
-
S
M
O
TE
0
.
8
6
0
3
0
.
8
4
4
3
0
.
8
5
1
7
0
.
8
9
5
9
Li
g
h
t
G
B
M
S
M
O
TE
-
R
A
D
I
A
N
T
0
.
8
6
2
0
0
.
8
4
6
9
0
.
8
5
3
9
0
.
8
9
7
4
Li
g
h
t
G
B
M
C
l
a
s
sW
e
i
g
h
t
0
.
8
4
3
1
0
.
8
5
7
8
0
.
8
5
0
2
0
.
8
9
3
1
X
G
B
o
o
st
O
r
i
g
i
n
a
l
0
.
8
6
2
0
0
.
8
4
2
9
0
.
8
5
1
7
0
.
8
9
6
0
X
G
B
o
o
st
S
M
O
TE
0
.
8
6
0
2
0
.
8
4
5
4
0
.
8
5
2
4
0
.
8
9
6
4
X
G
B
o
o
st
A
S
TR
A
-
S
M
O
TE
0
.
8
6
0
4
0
.
8
4
3
0
0
.
8
5
1
1
0
.
8
9
5
6
X
G
B
o
o
st
S
M
O
TE
-
R
A
D
I
A
N
T
0
.
8
6
1
8
0
.
8
4
4
5
0
.
8
5
2
5
0
.
8
9
6
4
X
G
B
o
o
st
C
l
a
s
sW
e
i
g
h
t
0
.
8
4
2
7
0
.
8
5
8
7
0
.
8
5
0
4
0
.
8
9
3
4
I
n
th
is
h
ig
h
-
d
im
en
s
io
n
al
s
p
ac
e,
th
e
ad
d
itio
n
o
f
s
tan
d
ar
d
SMOT
E
o
r
clu
s
ter
in
g
-
r
estricte
d
ASTRA
-
SM
OT
E
g
en
er
ally
d
eg
r
ad
ed
th
e
m
ac
r
o
-
F1
p
e
r
f
o
r
m
an
ce
f
o
r
L
ig
h
tGB
M.
Gen
er
atin
g
s
y
n
th
etic
tex
t
in
s
tan
ce
s
in
s
u
ch
a
s
p
ar
s
e
s
p
a
ce
o
f
ten
i
n
tr
o
d
u
ce
s
o
v
er
la
p
p
in
g
n
o
is
e
th
at
d
is
to
r
ts
p
r
ec
is
e
d
ec
is
io
n
b
o
u
n
d
ar
ies.
Ho
wev
er
,
SMOT
E
-
R
ADI
AN
T
s
u
cc
ess
f
u
lly
b
r
o
k
e
th
is
b
o
ttlen
ec
k
,
ac
h
iev
in
g
th
e
h
ig
h
est
n
u
m
er
ical
m
ac
r
o
-
F1
(
0
.
8
5
3
9
)
a
n
d
ac
c
u
r
ac
y
(
8
9
.
7
4
%)
ac
r
o
s
s
all
L
ig
h
tGB
M
co
n
f
ig
u
r
atio
n
s
.
B
y
tr
ea
tin
g
s
y
n
th
eti
c
d
ata
g
en
er
atio
n
as
a
two
-
s
tep
p
r
o
ce
s
s
-
wh
er
e
DB
S
C
AN
ac
ts
as
a
p
o
s
t
-
f
ilter
to
n
eu
tr
alize
o
v
er
la
p
p
in
g
a
n
o
m
alies
-
R
ADI
ANT
allo
ws th
e
m
o
d
el
to
s
af
ely
e
x
p
lo
it a
u
g
m
en
ted
m
in
o
r
it
y
r
eg
io
n
s
with
o
u
t b
o
u
n
d
ar
y
d
e
g
r
ad
ati
o
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
E
va
lu
a
tin
g
o
ve
r
s
a
mp
lin
g
meth
o
d
s
fo
r
imb
a
la
n
ce
d
A
r
a
b
ic
d
ia
lect
id
en
tifi
ca
tio
n
(
Ma
u
la
n
a
I
h
s
a
n
A
h
ma
d
)
263
Fo
r
XGBo
o
s
t,
th
e
im
p
ac
t
o
f
o
v
er
s
am
p
lin
g
was
m
o
r
e
n
u
an
ce
d
.
W
h
ile
SMOT
E
an
d
SMOT
E
-
R
ADI
ANT
b
o
th
y
ield
ed
id
en
tical
m
ar
g
in
al
g
ain
s
in
m
ac
r
o
-
F1
(
0
.
8
5
2
5
)
,
th
e
o
v
er
all
p
er
f
o
r
m
an
ce
o
f
XGBo
o
s
t
r
em
ain
ed
c
o
n
s
is
ten
tly
lo
wer
t
h
an
th
at
o
f
L
ig
h
tGB
M.
T
h
is
s
u
g
g
ests
th
at
th
e
lev
el
-
wis
e
tr
e
e
g
r
o
wth
s
tr
ateg
y
o
f
XGBo
o
s
t
is
g
en
er
ally
m
o
r
e
s
en
s
itiv
e
an
d
less
o
p
tim
al
at
h
an
d
lin
g
s
y
n
th
etic
p
er
tu
r
b
atio
n
s
in
h
ig
h
ly
s
p
a
r
s
e
lex
ical
s
p
ac
es c
o
m
p
ar
ed
t
o
L
i
g
h
tGB
M'
s
leaf
-
wis
e
g
r
o
wth
.
I
n
ter
esti
n
g
ly
,
th
e
c
o
s
t
-
s
en
s
it
iv
e
ap
p
r
o
ac
h
(
C
lass
W
eig
h
t
)
ex
h
ib
ited
th
e
lo
west
m
ac
ro
-
F1
an
d
ac
cu
r
ac
y
ac
r
o
s
s
b
o
th
lear
n
er
s
.
W
h
ile
it
s
u
cc
ess
f
u
lly
f
o
r
ce
d
th
e
m
o
d
els
to
p
ay
atten
tio
n
t
o
m
in
o
r
ity
class
es
(
as
ev
id
en
ce
d
b
y
th
e
h
ig
h
est
r
ec
all
s
co
r
es),
it
s
ev
er
ely
c
o
m
p
r
o
m
is
ed
m
ac
r
o
-
p
r
ec
is
io
n
.
T
h
is
h
ig
h
lig
h
ts
a
cr
itical
lim
itatio
n
o
f
n
o
n
-
g
en
er
a
tiv
e
r
ewe
ig
h
tin
g
in
h
ig
h
ly
o
v
er
lap
p
in
g
lex
ical
s
p
ac
es:
i
t
s
im
p
ly
s
h
if
ts
th
e
d
ec
is
io
n
th
r
esh
o
ld
,
in
cr
ea
s
in
g
m
in
o
r
ity
r
ec
o
g
n
itio
n
at
th
e
d
i
r
ec
t e
x
p
en
s
e
o
f
th
e
m
ajo
r
it
y
cla
s
s
.
3
.
2
.
E
rr
o
r
pa
t
t
er
ns
a
nd
per
-
cla
s
s
perf
o
rm
a
nce
T
o
e
v
a
l
u
a
t
e
d
i
al
e
c
t
-
le
v
e
l
p
e
r
f
o
r
m
a
n
c
e
s
h
i
f
ts
,
w
e
a
n
a
l
y
z
e
t
h
e
c
o
n
f
u
s
i
o
n
m
a
t
r
i
c
es
Fi
g
u
r
e
3
a
lo
n
g
s
i
d
e
t
h
e
p
e
r
-
c
l
a
s
s
F
1
-
s
c
o
r
e
s
T
a
b
l
e
5
a
n
d
F
i
g
u
r
e
4.
I
n
t
h
e
L
i
g
h
t
GB
M
(
o
r
i
g
i
n
a
l
)
c
o
n
f
i
g
u
r
a
t
i
o
n
,
t
h
e
m
o
d
e
l
p
e
r
f
o
r
m
s
r
o
b
u
s
t
l
y
b
u
t
n
a
t
u
r
a
ll
y
f
a
v
o
r
s
t
h
e
m
a
j
o
r
i
t
y
S
y
r
i
a
n
cl
a
s
s
,
y
i
e
l
d
i
n
g
l
o
w
e
r
t
r
u
e
p
o
s
it
i
v
es
f
o
r
t
h
e
s
e
v
e
r
el
y
u
n
d
e
r
r
e
p
r
e
s
e
n
t
e
d
J
o
r
d
a
n
i
a
n
d
ia
l
e
c
t
(
1
,
3
1
9
o
u
t
o
f
1
,
7
5
4
)
.
A
c
r
i
t
i
c
al
f
i
n
d
i
n
g
e
m
e
r
g
e
s
w
h
e
n
a
p
p
l
y
i
n
g
t
h
e
c
o
s
t
-
s
e
n
s
i
t
i
v
e
C
l
as
s
W
ei
g
h
t
s
t
r
a
t
e
g
y
.
W
h
i
l
e
it
s
u
c
ce
s
s
f
u
l
l
y
i
n
c
r
e
as
es m
i
n
o
r
i
t
y
r
e
c
a
l
l
-
J
o
r
d
a
n
i
a
n
t
r
u
e
p
o
s
i
t
i
v
es
r
i
s
e
f
r
o
m
1
,
3
1
9
t
o
1
,
3
9
1
-
i
t
i
n
d
is
c
r
i
m
i
n
ate
l
y
s
h
i
f
t
s
t
h
e
g
l
o
b
a
l
d
ec
i
s
i
o
n
b
o
u
n
d
a
r
y
.
T
h
i
s
s
e
v
e
r
e
l
y
d
a
m
a
g
es
t
h
e
m
a
j
o
r
it
y
c
la
s
s
,
d
r
o
p
p
i
n
g
c
o
r
r
e
c
t
S
y
r
i
a
n
p
r
e
d
i
ct
i
o
n
s
f
r
o
m
9
,
0
5
9
(
o
r
i
g
i
n
a
l
)
t
o
8
,
8
3
0
.
C
o
n
s
e
q
u
e
n
t
l
y
,
t
h
is
i
n
f
l
u
x
o
f
f
a
l
s
e
p
o
s
it
i
v
es
c
a
u
s
es
a
p
r
e
c
is
i
o
n
c
o
l
la
p
s
e
,
d
r
a
s
t
ic
a
l
l
y
d
r
o
p
p
i
n
g
t
h
e
J
o
r
d
a
n
i
an
F
1
-
s
c
o
r
e
f
r
o
m
0
.
7
9
4
1
t
o
0
.
7
8
5
0
i
n
L
i
g
h
t
G
B
M
.
T
h
i
s
d
e
m
o
n
s
t
r
a
t
es
t
h
at
n
o
n
-
g
e
n
e
r
a
t
i
v
e
c
o
s
t
-
s
e
n
s
it
i
v
e
le
a
r
n
i
n
g
m
e
r
e
l
y
c
a
n
n
i
b
a
li
z
es
t
h
e
m
a
j
o
r
i
t
y
c
l
ass
r
a
t
h
e
r
t
h
a
n
i
m
p
r
o
v
i
n
g
t
r
u
e
s
e
p
a
r
a
b
i
li
t
y
.
I
n
c
o
n
t
r
a
s
t
,
g
e
n
e
r
a
t
i
v
e
a
p
p
r
o
a
c
h
e
s
s
h
o
w
d
i
s
t
i
n
c
t
m
o
d
e
l
-
d
e
p
e
n
d
e
n
t
b
e
h
a
v
i
o
r
s
.
F
o
r
L
i
g
h
t
G
B
M
,
u
n
c
o
n
s
t
r
a
i
n
e
d
s
t
a
n
d
a
r
d
S
MO
T
E
a
n
d
c
l
u
s
te
r
-
r
e
s
t
r
i
c
t
e
d
AS
T
R
A
-
S
MO
T
E
i
n
t
r
o
d
u
c
e
o
v
e
r
l
a
p
p
i
n
g
s
y
n
t
h
e
ti
c
a
n
o
m
a
l
i
e
s
t
h
a
t
s
l
i
g
h
t
l
y
d
e
g
r
a
d
e
t
h
e
o
v
e
r
a
l
l
p
e
r
-
c
l
as
s
b
a
la
n
c
e.
H
o
w
e
v
e
r
,
S
M
OT
E
-
R
A
D
I
ANT
p
r
o
v
i
d
e
s
a
m
u
c
h
s
a
f
e
r
a
n
d
m
o
r
e
e
f
f
e
ct
i
v
e
m
i
n
o
r
i
t
y
-
m
a
j
o
r
i
t
y
t
r
a
d
e
-
o
f
f
.
B
y
u
ti
l
i
z
i
n
g
D
B
SC
A
N
t
o
f
i
lt
e
r
o
u
t
o
v
e
r
l
a
p
p
i
n
g
n
o
i
s
e
,
R
A
D
I
A
N
T
p
r
o
v
i
d
e
s
l
o
c
a
li
z
ed
s
y
n
t
h
e
ti
c
s
u
p
p
o
r
t
.
T
h
is
al
l
o
w
s
L
i
g
h
tG
B
M
t
o
es
t
a
b
li
s
h
t
i
g
h
t
e
r
d
e
c
is
i
o
n
b
o
u
n
d
a
r
i
e
s
,
i
m
p
r
o
v
i
n
g
J
o
r
d
a
n
i
a
n
a
n
d
L
e
b
a
n
e
s
e
r
e
c
o
g
n
i
t
i
o
n
w
i
t
h
o
u
t
a
g
g
r
e
s
s
i
v
el
y
c
a
n
n
ib
a
l
i
z
i
n
g
t
h
e
S
y
r
i
a
n
m
a
j
o
r
i
t
y
c
l
ass
.
F
i
g
u
r
e
3
(
a
)
s
h
o
w
s
t
h
e
o
r
i
g
i
n
al
L
i
g
h
t
GB
M
cl
as
s
i
f
i
c
a
ti
o
n
p
a
t
te
r
n
.
F
i
g
u
r
e
3
(
b
)
h
i
g
h
l
i
g
h
ts
t
h
e
b
a
l
a
n
c
e
d
i
m
p
r
o
v
e
m
e
n
t
a
c
h
i
e
v
e
d
b
y
S
MO
T
E
-
R
A
D
I
A
NT
.
F
i
g
u
r
e
3
(
c
)
il
l
u
s
t
r
at
es
t
h
e
t
r
a
d
e
-
o
f
f
i
n
t
r
o
d
u
ce
d
b
y
C
l
as
s
W
e
i
g
h
t.
F
i
g
u
r
e
3
(
d
)
s
h
o
w
s
t
h
e
o
r
i
g
i
n
a
l
X
G
B
o
o
s
t
c
l
ass
i
f
i
ca
t
i
o
n
p
a
t
t
e
r
n
.
F
i
g
u
r
e
3
(
e
)
d
e
m
o
n
s
t
r
a
t
e
s
i
m
p
r
o
v
e
d
c
l
a
s
s
b
a
l
a
n
ce
w
i
t
h
SM
O
T
E
-
R
A
D
I
A
N
T
.
F
i
g
u
r
e
3
(
f
)
i
l
l
u
s
t
r
at
e
s
t
h
e
s
t
r
o
n
g
e
r
b
i
a
s
i
n
t
r
o
d
u
c
e
d
b
y
C
l
a
s
s
W
e
i
g
h
t
.
(
a
)
(
b
)
(
c
)
(
d
)
(
e
)
(f)
F
i
g
u
r
e
3
.
C
o
n
f
u
s
i
o
n
m
a
t
r
i
c
es
co
m
p
a
r
i
n
g
d
i
a
l
e
c
t
c
la
s
s
i
f
ic
a
t
i
o
n
p
e
r
f
o
r
m
a
n
c
e
u
n
d
e
r
d
i
f
f
e
r
e
n
t
b
a
l
a
n
c
i
n
g
s
t
r
at
e
g
i
es
(
a
)
L
i
g
h
t
G
B
M
(
O
r
i
g
i
n
al
)
,
(
b
)
L
i
g
h
t
G
B
M
w
it
h
S
MO
T
E
-
R
A
D
I
A
N
T
,
(
c
)
L
i
g
h
tG
B
M
w
it
h
c
l
ass
w
e
i
g
h
t
,
(
d
)
X
G
B
o
o
s
t
(
O
r
i
g
i
n
al
)
,
(
e
)
XG
B
o
o
s
t
w
it
h
S
MO
T
E
-
R
A
D
I
AN
T
,
a
n
d
(
f
)
X
G
B
o
o
s
t
wi
t
h
c
la
s
s
w
e
i
g
h
t
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
4
3
,
No
.
1
,
Ju
ly
20
2
6
:
259
-
27
0
264
Fo
r
XGBo
o
s
t
,
all
r
eb
alan
cin
g
s
tr
ateg
ies
-
wh
eth
er
g
en
er
ativ
e
o
r
co
s
t
-
s
en
s
itiv
e
-
s
tr
u
g
g
le
to
c
o
n
s
is
ten
tly
im
p
r
o
v
e
th
e
s
e
v
e
r
e
l
y
im
b
ala
n
ce
d
J
o
r
d
an
ia
n
class
.
E
v
en
with
R
ADI
ANT
,
XG
B
o
o
s
t’
s
lev
el
-
wis
e
g
r
o
wth
s
h
o
ws
h
ig
h
er
s
en
s
itiv
ity
to
s
y
n
th
etic
p
er
tu
r
b
atio
n
s
,
f
ailin
g
to
f
u
lly
d
is
en
tan
g
le
th
e
b
o
u
n
d
ar
ies
in
th
e
5
,
6
4
4
-
d
im
en
s
io
n
al
s
p
ac
e.
Acr
o
s
s
all
m
o
d
els
an
d
s
tr
ateg
ies,
co
n
f
u
s
io
n
b
etwe
en
L
eb
an
ese
an
d
Palest
in
ian
r
em
ai
n
s
th
e
m
o
s
t
f
r
eq
u
e
n
t
b
id
ir
ec
tio
n
al
e
r
r
o
r
.
T
h
is
r
ef
lects
th
e
s
tr
o
n
g
le
x
ical,
m
o
r
p
h
o
lo
g
ical,
an
d
g
eo
g
r
ap
h
ical
o
v
er
l
a
p
b
etwe
en
th
ese
two
L
e
v
an
tin
e
d
ialec
ts
,
co
u
p
le
d
with
th
e
lim
ited
co
n
tex
t
u
al
cu
es
av
aila
b
le
in
s
h
o
r
t
T
witter
tex
ts
.
Ov
er
all,
th
is
co
m
b
in
ed
an
aly
s
is
co
n
f
ir
m
s
th
at
wh
ile
C
las
s
W
eig
h
t
r
ed
is
tr
ib
u
t
es
m
is
cla
s
s
if
icatio
n
p
atter
n
s
b
y
s
ac
r
if
icin
g
m
a
jo
r
ity
-
class
p
r
ec
i
s
io
n
,
d
en
s
i
ty
-
g
u
id
e
d
o
v
e
r
s
am
p
lin
g
(
S
MO
T
E
-
R
ADI
ANT
)
g
en
u
in
ely
r
ef
in
es in
ter
-
class
d
ec
is
io
n
b
o
u
n
d
ar
ies f
o
r
leaf
-
wi
s
e
g
r
ad
ien
t b
o
o
s
tin
g
m
o
d
els.
T
ab
le
5
.
Per
-
class
F1
-
s
co
r
e
o
n
th
e
test
s
et
(
Δ
r
elativ
e
to
o
r
ig
i
n
al)
M
o
d
e
l
D
i
a
l
e
c
t
O
r
i
g
i
n
a
l
S
M
O
TE
(
Δ)
A
S
TR
A
-
S
M
O
TE
(
Δ)
S
M
O
TE
-
R
A
D
I
A
N
T
(
Δ)
C
l
a
s
sW
e
i
g
h
t
(
Δ)
Li
g
h
t
G
B
M
Jo
r
d
a
n
0
.
7
9
4
1
0
.
7
9
0
7
(
−
0
.
0
0
3
4
)
0
.
7
9
1
3
(
−
0
.
0
0
2
8
)
0
.
7
9
4
6
(
+
0
.
0
0
0
5
)
0
.
7
8
5
0
(
−
0
.
0
0
9
1
)
Le
b
a
n
o
n
0
.
8
3
8
4
0
.
8
4
2
2
(
+
0
.
0
0
3
8
)
0
.
8
3
9
1
(
+
0
.
0
0
0
7
)
0
.
8
4
5
9
(
+
0
.
0
0
7
5
)
0
.
8
3
9
0
(
+
0
.
0
0
0
6
)
P
a
l
e
s
t
i
n
e
0
.
8
2
8
7
0
.
8
2
8
9
(
+
0
.
0
0
0
2
)
0
.
8
2
6
5
(
−
0
.
0
0
2
2
)
0
.
8
2
6
4
(
−
0
.
0
0
2
3
)
0
.
8
2
8
0
(
−
0
.
0
0
0
7
)
S
y
r
i
a
0
.
9
4
9
2
0
.
9
4
5
3
(
−
0
.
0
0
3
9
)
0
.
9
4
9
8
(
+
0
.
0
0
0
6
)
0
.
9
4
8
5
(
−
0
.
0
0
0
7
)
0
.
9
4
8
9
(
−
0
.
0
0
0
3
)
M
a
c
r
o
-
F1
0
.
8
5
2
6
0
.
8
5
1
8
(
−
0
.
0
0
0
8
)
0
.
8
5
1
7
(
−
0
.
0
0
0
9
)
0
.
8
5
3
9
(
+
0
.
0
0
1
3
)
0
.
8
5
0
2
(
−
0
.
0
0
2
4
)
X
G
B
o
o
st
Jo
r
d
a
n
0
.
7
9
0
7
0
.
7
9
3
1
(
+
0
.
0
0
2
4
)
0
.
7
8
9
5
(
−
0
.
0
0
1
2
)
0
.
7
9
2
2
(
+
0
.
0
0
1
5
)
0
.
7
8
4
5
(
−
0
.
0
0
6
2
)
Le
b
a
n
o
n
0
.
8
3
9
4
0
.
8
4
0
8
(
+
0
.
0
0
1
4
)
0
.
8
3
9
5
(
+
0
.
0
0
0
1
)
0
.
8
4
2
5
(
+
0
.
0
0
3
1
)
0
.
8
4
0
3
(
+
0
.
0
0
0
9
)
P
a
l
e
s
t
i
n
e
0
.
8
2
7
5
0
.
8
2
7
6
(
+
0
.
0
0
0
1
)
0
.
8
2
5
8
(
−
0
.
0
0
1
7
)
0
.
8
2
6
4
(
−
0
.
0
0
1
1
)
0
.
8
2
7
3
(
−
0
.
0
0
0
2
)
S
y
r
i
a
0
.
9
4
9
2
0
.
9
4
7
6
(
−
0
.
0
0
1
6
)
0
.
9
4
9
5
(
+
0
.
0
0
0
3
)
0
.
9
4
8
8
(
−
0
.
0
0
0
4
)
0
.
9
4
9
4
(
+
0
.
0
0
0
2
)
M
a
c
r
o
-
F1
0
.
8
5
1
7
0
.
8
5
2
4
(
+
0
.
0
0
0
7
)
0
.
8
5
1
1
(
−
0
.
0
0
0
6
)
0
.
8
5
2
5
(
+
0
.
0
0
0
8
)
0
.
8
5
0
4
(
−
0
.
0
0
1
3
)
Fig
u
r
e
4
.
Per
-
class
F1
-
s
co
r
e
o
f
L
ig
h
tGB
M
an
d
XGBo
o
s
t u
n
d
er
d
if
f
e
r
en
t
im
b
ala
n
ce
-
h
a
n
d
li
n
g
s
tr
ateg
ies
.
L
ig
h
tGB
M
ex
h
ib
its
b
alan
ce
d
i
m
p
r
o
v
e
m
en
ts
ac
r
o
s
s
m
in
o
r
ity
d
ialec
ts
u
n
d
er
SMOT
E
-
R
ADI
ANT
,
s
u
cc
ess
f
u
lly
av
o
id
in
g
th
e
p
r
ec
is
io
n
co
llap
s
e
s
ee
n
in
C
lass
W
eig
h
t
an
d
XGBo
o
s
t stru
g
g
les with
b
o
u
n
d
ar
y
d
is
to
r
tio
n
ac
r
o
s
s
all
g
en
er
ativ
e
an
d
c
o
s
t
-
s
en
s
itiv
e
m
eth
o
d
s
3.
3
.
B
lo
c
k
-
lev
el
f
e
a
t
ure
im
p
o
rt
a
nce
Giv
en
th
e
ex
tr
e
m
e
s
p
ar
s
ity
o
f
th
e
5
,
6
4
4
-
d
im
e
n
s
io
n
al
s
p
ac
e,
an
aly
zin
g
in
d
iv
id
u
al
f
ea
tu
r
es
is
u
n
in
ter
p
r
etab
le.
I
n
s
tead
,
we
a
g
g
r
eg
ated
th
e
f
ea
t
u
r
e_
im
p
o
r
t
an
ce
s
_
(
m
ea
s
u
r
ed
b
y
th
e
to
ta
l
n
u
m
b
e
r
o
f
s
p
lit
s
)
f
r
o
m
th
e
tr
ee
m
o
d
els
in
to
f
u
n
ctio
n
al
b
l
o
ck
s
.
Fig
u
r
e
5
illu
s
tr
ates
th
ese
ag
g
r
eg
at
ed
p
atter
n
s
ac
r
o
s
s
r
ep
r
esen
tativ
e
co
n
f
ig
u
r
ati
o
n
s
.
Fig
u
r
e
5
(
a)
s
h
o
ws
th
e
o
r
ig
i
n
al
f
ea
tu
r
e
h
ier
ar
ch
y
o
f
L
ig
h
tGB
M.
Fig
u
r
e
5
(
b
)
d
em
o
n
s
tr
ates
th
at
SMOT
E
-
R
ADI
ANT
p
r
eser
v
es
th
is
h
ier
ar
ch
y
with
o
n
ly
m
in
o
r
ch
an
g
es.
Fig
u
r
e
5
(
c
)
r
ev
ea
l
s
th
at
C
las
s
W
eig
h
t
m
ar
k
ed
ly
i
n
cr
ea
s
es
s
p
lit
co
u
n
ts
ac
r
o
s
s
f
ea
tu
r
e
b
lo
ck
s
.
Fi
g
u
r
e
5
(
d
)
p
r
esen
ts
th
e
o
r
i
g
in
al
XGBo
o
s
t
f
ea
tu
r
e
d
is
tr
ib
u
tio
n
,
Fig
u
r
e
5
(
e
)
s
h
o
w
s
th
at
SMOT
E
-
R
ADI
ANT
m
a
in
tain
s
a
s
im
ilar
p
atter
n
,
an
d
F
ig
u
r
e
5
(
f
)
i
n
d
icate
s
th
at
C
lass
W
eig
h
t a
g
ain
s
u
b
s
tan
tially
in
f
lates sp
lit co
u
n
ts
.
A
d
is
tin
ct
h
ier
ar
ch
y
em
er
g
es
b
etwe
en
f
ea
tu
r
e
v
o
lu
m
e
a
n
d
f
ea
tu
r
e
ef
f
icien
c
y
.
Acr
o
s
s
all
Or
ig
in
al
m
o
d
els,
th
e
c
h
ar
ac
ter
-
lev
el
s
u
b
-
wo
r
d
m
ar
k
er
s
(
ch
a
r
_
m
i2
3
)
a
n
d
s
p
ar
s
e
wo
r
d
-
lev
el
r
e
p
r
e
s
en
tatio
n
s
(
lex
ical)
d
o
m
in
ate
th
e
to
tal
im
p
o
r
tan
c
e
d
u
e
to
th
eir
m
ass
iv
e
d
im
e
n
s
io
n
ality
(
5
0
0
an
d
2
,
4
2
0
co
l
u
m
n
s
,
r
esp
ec
tiv
ely
)
.
No
te
th
at
ch
ar
_
m
i2
3
r
e
f
er
s
s
p
ec
if
ically
to
th
e
ch
ar
ac
ter
2
–
3
-
g
r
am
s
u
b
-
b
lo
ck
with
in
t
h
e
b
r
o
ad
e
r
3
,
1
5
0
-
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
E
va
lu
a
tin
g
o
ve
r
s
a
mp
lin
g
meth
o
d
s
fo
r
imb
a
la
n
ce
d
A
r
a
b
ic
d
ia
lect
id
en
tifi
ca
tio
n
(
Ma
u
la
n
a
I
h
s
a
n
A
h
ma
d
)
265
d
im
en
s
io
n
al
C
h
ar
ac
ter
N
-
g
r
a
m
s
(
MI
)
b
lo
ck
r
e
p
o
r
ted
i
n
T
ab
le
2
.
Ho
wev
er
,
in
ter
m
s
o
f
a
v
e
r
ag
e
im
p
o
r
ta
n
ce
p
er
co
lu
m
n
,
th
e
Naiv
e
B
ay
es
-
SVM
lo
g
-
co
u
n
t
r
atio
s
(
n
b
s
v
m
)
an
d
th
e
Ou
t
-
of
-
Fo
ld
s
tack
in
g
p
r
o
b
ab
ilit
ies
(
o
o
f
_
s
v
m
,
o
o
f
_
lr
,
o
o
f
_
in
te
r
ac
tio
n
s
)
ar
e
ex
ce
p
tio
n
ally
p
o
w
er
f
u
l.
Fo
r
in
s
tan
ce
,
in
L
ig
h
tGB
M
(
Or
ig
in
al)
,
th
e
4
co
lu
m
n
s
o
f
th
e
n
b
s
v
m
b
l
o
ck
av
er
ag
e
3
,
7
1
0
s
p
lits
p
er
co
l
u
m
n
,
m
a
k
in
g
th
em
t
h
e
m
o
s
t
s
tr
u
ctu
r
ally
cr
itical
s
ig
n
als f
o
r
th
e
f
ir
s
t f
ew
lev
els
o
f
th
e
d
ec
is
io
n
tr
ee
s
.
W
h
en
SMOT
E
-
R
ADI
ANT
is
ap
p
lied
,
th
e
r
elativ
e
co
n
tr
ib
u
ti
o
n
an
d
r
a
n
k
in
g
s
tr
u
ctu
r
e
o
f
th
ese
f
ea
tu
r
e
b
lo
ck
s
r
em
ain
h
ig
h
ly
s
tab
le.
I
n
L
ig
h
tGB
M
R
A
DI
ANT
,
ch
ar
_
m
i2
3
(
3
5
,
4
7
0
s
p
lits
)
an
d
lex
ical
(
2
8
,
4
4
0
s
p
lits
)
r
etain
th
eir
p
o
s
itio
n
s
,
with
a
s
lig
h
t
r
ein
f
o
r
ce
m
en
t
in
l
o
n
g
e
r
ch
ar
ac
ter
s
eq
u
e
n
ce
s
(
ch
a
r
_
m
i
4
5
r
is
in
g
to
1
8
,
3
3
3
s
p
lits
)
.
T
h
is
s
tab
ilit
y
is
a
s
tr
o
n
g
in
d
icato
r
th
at
d
en
s
ity
-
b
a
s
ed
f
ilter
in
g
s
u
cc
ess
f
u
lly
au
g
m
en
ts
th
e
m
in
o
r
it
y
r
eg
io
n
s
with
o
u
t
d
is
r
u
p
tin
g
o
r
co
n
f
u
s
in
g
t
h
e
alg
o
r
ith
m
'
s
r
elia
n
ce
o
n
its
m
o
s
t tr
u
s
ted
lin
g
u
is
tic
cu
es.
I
n
s
h
ar
p
c
o
n
tr
ast,
ap
p
l
y
in
g
co
s
t
-
s
en
s
itiv
e
lear
n
in
g
(
C
lass
W
eig
h
t)
d
r
asti
ca
lly
alter
s
tr
ee
g
r
o
wth
b
eh
av
io
r
.
I
n
b
o
th
L
ig
h
tGB
M
an
d
XGBo
o
s
t,
th
e
to
tal
n
u
m
b
er
o
f
s
p
lits
ac
r
o
s
s
alm
o
s
t a
ll b
lo
ck
s
n
ea
r
ly
d
o
u
b
les.
Fo
r
ex
am
p
le,
XGBo
o
s
t'
s
r
eli
an
ce
o
n
ch
a
r
_
m
i2
3
in
f
lates
f
r
o
m
5
7
,
6
0
4
s
p
lits
(
Or
ig
in
al)
to
1
2
9
,
9
4
6
s
p
lits
(
C
lass
W
eig
h
t)
.
T
h
is
ex
tr
em
e
in
f
latio
n
r
e
v
ea
ls
th
at
th
e
h
ea
v
y
p
en
aliza
tio
n
weig
h
ts
f
o
r
ce
th
e
b
o
o
s
tin
g
alg
o
r
ith
m
to
o
v
er
-
f
r
ag
m
e
n
t
th
e
d
ec
is
io
n
s
p
ac
e
in
a
d
esp
er
ate
attem
p
t
to
s
ep
ar
ate
o
v
er
lap
p
in
g
class
es,
d
ir
ec
tly
ex
p
lain
in
g
t
h
e
p
r
ec
is
io
n
c
o
ll
ap
s
e
an
d
b
o
u
n
d
ar
y
d
is
to
r
tio
n
o
b
s
er
v
ed
in
Sectio
n
3
.
2
.
(
a
)
(
b
)
(
c
)
(
d
)
(
e
)
(f)
F
i
g
u
r
e
5
.
A
g
g
r
e
g
a
t
e
d
f
e
a
t
u
r
e
im
p
o
r
t
a
n
c
e
p
e
r
b
l
o
c
k
.
S
M
O
T
E
-
R
A
D
I
A
N
T
p
r
es
e
r
v
e
s
t
h
e
s
t
a
b
le
f
e
a
t
u
r
e
h
i
e
r
a
r
c
h
y
o
f
t
h
e
o
r
i
g
i
n
a
l
d
a
t
a
,
w
h
e
r
e
a
s
C
l
ass
W
e
i
g
h
t
f
o
r
c
e
s
a
g
g
r
e
s
s
i
v
e
t
r
e
e
f
r
a
g
m
e
n
t
at
i
o
n
,
v
i
s
i
b
l
y
i
n
f
l
a
t
i
n
g
t
o
t
al
s
p
l
i
t
c
o
u
n
ts
a
c
r
o
s
s
a
l
l
f
ea
t
u
r
e
b
l
o
c
k
s
(
a
)
L
i
g
h
t
GB
M
(
O
r
i
g
i
n
a
l
)
,
(
b
)
L
i
g
h
t
GB
M
w
i
t
h
S
M
O
T
E
-
R
AD
I
A
N
T
,
(
c
)
L
i
g
h
tG
B
M
w
i
t
h
C
l
ass
W
e
i
g
h
t
,
(
d
)
XG
B
o
o
s
t
(
O
r
i
g
i
n
al
)
,
(
e
)
X
GB
o
o
s
t
wi
t
h
S
M
O
T
E
-
R
AD
I
A
N
T
,
a
n
d
(
f
)
X
G
B
o
o
s
t
w
it
h
C
l
a
s
s
W
e
i
g
h
t
3.
4
.
St
a
t
is
t
ica
l
v
a
lid
a
t
io
n
T
o
en
s
u
r
e
o
b
s
er
v
ed
test
-
s
et
f
lu
ctu
atio
n
s
ar
e
n
o
t
ar
tif
ac
ts
o
f
s
am
p
lin
g
v
ar
ia
b
ilit
y
,
we
em
p
lo
y
ed
a
B
o
o
ts
tr
ap
p
in
g
p
r
o
ce
d
u
r
e.
W
e
p
er
f
o
r
m
ed
1
,
0
0
0
b
o
o
ts
tr
ap
iter
atio
n
s
(
s
am
p
lin
g
wit
h
r
ep
lace
m
en
t)
o
n
p
r
ed
ictio
n
s
to
g
en
er
ate
h
ig
h
ly
s
tab
le
Ma
cr
o
-
F1
d
is
tr
ib
u
tio
n
s
.
Pair
ed
two
-
s
id
ed
W
ilc
o
x
o
n
s
ig
n
ed
-
r
an
k
test
s
an
d
C
o
h
e
n
’
s
ef
f
ec
t
s
izes
(
r
)
wer
e
th
en
ca
lcu
lated
t
o
m
e
asu
r
e
th
e
p
r
ac
tical
m
ag
n
itu
d
e
o
f
th
e
d
if
f
e
r
en
ce
s
.
Usi
n
g
N=
1
,
0
0
0
g
u
a
r
an
tees
p
-
v
alu
es
th
at
a
r
e
ex
ce
p
tio
n
ally
r
o
b
u
s
t
ag
ain
s
t
r
a
n
d
o
m
s
ee
d
v
ar
iatio
n
s
.
T
ab
le
6
s
u
m
m
ar
izes th
ese
f
in
d
in
g
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
4
3
,
No
.
1
,
Ju
ly
20
2
6
:
259
-
27
0
266
T
ab
le
6
.
B
o
o
ts
tr
ap
Ma
cr
o
-
F
1
Me
an
s
(
N=
1
,
0
0
0
)
.
All c
o
m
p
ar
is
o
n
s
ag
ain
s
t th
e
Or
ig
in
al
co
n
f
ig
u
r
atio
n
ar
e
h
ig
h
ly
s
ig
n
if
ican
t (
p
<
0
.
0
0
1
)
.
T
h
e
ef
f
ec
t size
(
C
o
h
e
n
'
s
r
)
is
d
en
o
ted
i
n
p
ar
e
n
th
eses
,
alo
n
g
s
id
e
th
e
d
i
r
ec
tio
n
o
f
th
e
p
er
f
o
r
m
an
ce
s
h
if
t (
+
f
o
r
in
cr
e
ase,
-
f
o
r
d
ec
r
ea
s
e)
M
o
d
e
l
O
r
i
g
i
n
a
l
S
M
O
TE
A
S
TR
A
-
S
M
O
TE
S
M
O
TE
-
R
A
D
I
A
N
T
C
l
a
s
sW
e
i
g
h
t
Li
g
h
t
G
B
M
0
.
8
5
2
5
0
.
8
5
1
7
(
r
=
0
.
3
6
)
(
-
)
0
.
8
5
1
5
(
r
=
0
.
4
4
)
(
-
)
0
.
8
5
3
7
(
r
=
0
.
5
1
)
(
+
)
0
.
8
5
0
2
(
r
=
0
.
5
8
)
(
-
)
X
G
B
o
o
st
0
.
8
5
1
6
0
.
8
5
2
2
(
r
=
0
.
2
9
)
(
+
)
0
.
8
5
0
9
(
r
=
0
.
3
3
)
(
-
)
0
.
8
5
2
3
(
r
=
0
.
3
3
)
(
+
)
0
.
8
5
0
4
(
r
=
0
.
3
8
)
(
-
)
T
h
e
s
tatis
tica
l
ev
id
en
ce
co
m
p
letely
r
ef
u
tes
th
e
ass
u
m
p
tio
n
th
at
all
o
v
er
s
am
p
lin
g
m
eth
o
d
s
p
r
o
v
id
e
eq
u
iv
alen
t
o
r
n
e
g
lig
ib
le
im
p
a
cts
in
s
p
ar
s
e
tex
t
s
p
ac
es.
Fo
r
L
ig
h
tGB
M,
th
e
ap
p
licatio
n
o
f
s
tan
d
ar
d
SMOT
E
,
clu
s
ter
in
g
-
r
estricte
d
ASTRA
-
SMOT
E
,
an
d
C
las
s
W
eig
h
t
al
l
r
esu
lted
in
s
tati
s
tically
s
ig
n
if
ican
t
p
er
f
o
r
m
an
ce
d
eg
r
ad
atio
n
(
Sig
.
Dec
r
ea
s
e,
p
<
0
.
0
0
1
)
with
m
ed
iu
m
to
lar
g
e
ef
f
ec
t
s
izes
(
r
r
an
g
in
g
f
r
o
m
0
.
3
6
to
0
.
5
8
)
.
T
h
is
p
r
o
v
es
th
at
u
n
co
n
s
tr
ain
e
d
s
y
n
th
etic
g
en
e
r
atio
n
a
n
d
al
g
o
r
ith
m
ic
r
ewe
ig
h
tin
g
ar
e
ac
tiv
ely
h
ar
m
f
u
l
wh
en
d
ea
lin
g
with
h
ig
h
ly
o
v
er
lap
p
i
n
g
5
,
6
4
4
-
d
im
en
s
io
n
al
le
x
ical
f
ea
tu
r
es.
I
n
s
tar
k
co
n
tr
ast,
SMOT
E
-
R
ADI
ANT
is
th
e
o
n
ly
co
n
f
ig
u
r
at
io
n
th
at
y
ield
s
a
s
tatis
tically
s
ig
n
if
ican
t
p
er
f
o
r
m
an
ce
i
n
cr
ea
s
e
f
o
r
L
i
g
h
tGB
M,
ac
co
m
p
an
ied
b
y
a
lar
g
e
p
o
s
itiv
e
ef
f
ec
t
s
ize
(
r
=
0
.
5
1
1
)
.
T
h
is
estab
lis
h
es
th
at
th
e
p
e
r
f
o
r
m
an
ce
g
ain
s
ac
h
iev
ed
b
y
a
p
p
ly
in
g
DB
SC
AN
-
b
ased
n
o
is
e
n
eu
tr
aliza
tio
n
ar
e
n
o
t
d
u
e
to
r
a
n
d
o
m
ch
an
ce
,
b
u
t r
ep
r
esen
t a
s
tr
u
ctu
r
ally
s
o
u
n
d
an
d
h
ig
h
ly
r
ep
r
o
d
u
cib
le
r
ef
in
em
e
n
t o
f
t
h
e
d
ec
is
io
n
b
o
u
n
d
a
r
ies.
Fo
r
XGBo
o
s
t,
wh
ile
s
tan
d
a
r
d
SMOT
E
a
n
d
SMOT
E
-
R
ADI
ANT
b
o
th
r
eg
is
ter
ed
s
tatis
tically
s
ig
n
if
ican
t
in
cr
ea
s
es,
th
eir
ef
f
ec
t
s
izes
wer
e
n
o
tab
ly
s
m
aller
(
r
=
0
.
2
8
8
an
d
r
=
0
.
3
2
7
,
r
esp
ec
tiv
ely
)
co
m
p
ar
e
d
to
th
e
g
ain
s
s
ee
n
in
L
ig
h
tGB
M.
Fu
r
th
er
m
o
r
e,
b
ec
au
s
e
XGBo
o
s
t
'
s
in
itia
l
p
er
f
o
r
m
an
ce
w
as
in
h
er
en
tly
lo
wer
th
an
L
ig
h
tGB
M'
s
,
th
ese
m
in
o
r
s
tatis
tica
l g
ain
s
wer
e
in
s
u
f
f
icien
t to
m
ak
e
XGBo
o
s
t th
e
s
u
p
e
r
io
r
m
o
d
el.
Ov
er
all,
th
is
r
ig
o
r
o
u
s
1
,
0
0
0
-
iter
atio
n
b
o
o
ts
tr
ap
p
in
g
v
a
lid
atio
n
co
n
f
ir
m
s
th
at
d
e
n
s
ity
-
f
ilter
ed
o
v
er
s
am
p
lin
g
(
R
ADI
ANT
)
p
r
o
v
id
es
a
p
r
ac
t
ically
m
ea
n
in
g
f
u
l
im
p
r
o
v
em
en
t
o
v
e
r
b
o
th
g
e
n
er
ativ
e
an
d
co
s
t
-
s
en
s
itiv
e
ap
p
r
o
ac
h
es,
s
p
e
cif
ically
wh
en
p
air
e
d
with
leaf
-
wis
e
g
r
ad
ien
t b
o
o
s
tin
g
ar
ch
it
ec
tu
r
es.
3.
5
.
Co
m
pa
riso
n wit
h
t
ra
ns
f
o
rm
er
-
ba
s
ed
a
pp
ro
a
ches
T
o
co
n
tex
tu
alize
th
e
p
ea
k
p
er
f
o
r
m
a
n
ce
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
o
lo
g
y
,
T
ab
le
7
co
m
p
ar
es
o
u
r
ch
am
p
io
n
m
o
d
el
(
L
ig
h
tGB
M
SMOT
E
-
R
ADI
ANT
)
ag
ain
s
t
p
r
ev
io
u
s
ly
r
ep
o
r
ted
n
e
u
r
al
an
d
tr
an
s
f
o
r
m
er
-
b
ased
ap
p
r
o
ac
h
es o
n
th
e
Sh
am
i d
ataset.
Fo
r
co
n
s
is
ten
cy
,
th
e
p
er
-
class
F1
-
s
co
r
es o
f
th
e
r
ec
u
r
r
en
t,
co
n
v
o
lu
tio
n
al,
an
d
tr
an
s
f
o
r
m
er
m
o
d
els we
r
e
r
ec
o
m
p
u
ted
as p
e
r
ce
n
tag
es b
ased
o
n
th
e
c
o
n
f
u
s
io
n
m
atr
ices r
ep
o
r
ted
[
2
3
]
.
T
ab
le
7
.
Per
-
class
F1
-
s
co
r
e
(
%)
co
m
p
ar
is
o
n
a
g
ain
s
t d
ee
p
le
ar
n
in
g
a
r
ch
itectu
r
es o
n
th
e
Sh
am
i d
ataset
A
p
p
r
o
a
c
h
e
s
M
o
d
e
l
Jo
r
d
a
n
Le
b
a
n
o
n
P
a
l
e
s
t
i
n
e
S
y
r
i
a
M
a
c
r
o
-
F1
R
e
c
u
r
r
e
n
t
G
R
U
6
4
.
0
7
7
9
.
9
3
7
3
.
6
5
9
3
.
2
3
7
7
.
7
2
R
e
c
u
r
r
e
n
t
LSTM
5
8
.
5
6
7
7
.
3
3
7
0
.
6
6
9
1
.
8
4
7
4
.
6
0
C
o
n
v
o
l
u
t
i
o
n
a
l
C
N
N
5
7
.
2
3
7
8
.
0
2
7
0
.
2
6
9
1
.
4
5
7
4
.
2
4
Tr
a
n
sf
o
r
mer
-
b
a
se
d
B
a
se
-
A
r
a
b
e
r
t
7
2
.
1
4
8
3
.
2
8
8
0
.
9
4
9
4
.
8
2
8
2
.
7
9
Tr
a
n
sf
o
r
mer
-
b
a
se
d
A
r
a
b
i
c
-
X
LM
-
R
-
B
a
s
e
7
4
.
2
1
8
3
.
5
5
8
1
.
5
6
9
4
.
8
2
8
3
.
5
3
Tr
a
n
sf
o
r
mer
-
b
a
se
d
S
t
a
c
k
i
n
g
-
Tr
a
n
sf
o
r
mer
7
8
.
7
3
8
3
.
9
8
8
2
.
6
3
9
5
.
1
3
8
5
.
1
2
G
r
a
d
i
e
n
t
B
o
o
st
i
n
g
(
t
h
i
s s
t
u
d
y
)
Li
g
h
t
G
B
M
(
O
r
i
g
i
n
a
l
)
7
9
.
4
1
8
3
.
8
4
8
2
.
8
7
9
4
.
9
2
8
5
.
2
6
G
r
a
d
i
e
n
t
B
o
o
st
i
n
g
(
t
h
i
s s
t
u
d
y
)
Li
g
h
t
G
B
M
S
M
O
TE
-
R
A
D
I
A
N
T
7
9
.
4
6
8
4
.
5
9
8
2
.
6
4
9
4
.
8
5
8
5
.
3
9
As
s
h
o
wn
in
T
ab
le
7
,
r
e
cu
r
r
e
n
t
an
d
co
n
v
o
lu
tio
n
al
ar
ch
itectu
r
es
s
tr
u
g
g
le
s
ig
n
if
ican
tly
,
p
ar
t
icu
lar
ly
in
id
en
tify
in
g
th
e
s
ev
er
ely
u
n
d
er
r
ep
r
esen
ted
J
o
r
d
a
n
ian
d
ialec
t.
W
h
ile
in
d
iv
id
u
al
tr
an
s
f
o
r
m
er
-
b
ased
m
o
d
els
lik
e
Ar
aBER
T
im
p
r
o
v
e
o
v
er
all
p
e
r
f
o
r
m
a
n
ce
,
th
e
y
s
till
ex
h
ib
it
n
o
tab
le
im
b
alan
ce
ac
r
o
s
s
d
ialec
ts
,
with
J
o
r
d
an
ia
n
F1
r
em
ain
in
g
r
elativ
ely
lo
w
(
≈
7
2
.
14
%)
[
2
3
]
.
W
ith
in
o
u
r
c
o
n
tr
o
lled
e
x
p
er
im
e
n
tal
f
r
am
e
wo
r
k
,
th
e
lig
h
tweig
h
t
L
ig
h
tGB
M
SMOT
E
-
R
ADI
A
NT
p
ip
elin
e
n
u
m
er
ically
o
u
tp
er
f
o
r
m
s
h
ea
v
y
in
d
iv
id
u
al
tr
an
s
f
o
r
m
er
ar
c
h
itectu
r
es
an
d
is
h
ig
h
ly
co
m
p
etitiv
e
with
co
m
p
lex
Stack
in
g
-
T
r
an
s
f
o
r
m
er
m
o
d
els.
Mo
s
t
n
o
tab
ly
,
o
u
r
p
ip
elin
e
ac
h
iev
es
s
u
b
s
tan
tially
h
ig
h
er
r
ec
o
g
n
itio
n
f
o
r
th
e
m
in
o
r
ity
J
o
r
d
an
ia
n
d
ialec
t
(
7
9
.
4
6
%).
H
o
wev
er
,
b
e
ca
u
s
e
ex
p
er
im
e
n
tal
p
ip
elin
es
d
if
f
er
in
p
r
ep
r
o
ce
s
s
in
g
an
d
o
p
tim
izatio
n
s
tr
ateg
ies,
th
is
co
m
p
ar
is
o
n
s
er
v
es
as
a
c
o
n
tex
tu
al
r
ef
er
e
n
ce
p
o
in
t
r
at
h
er
th
a
n
a
claim
o
f
ab
s
o
lu
te
ar
ch
itectu
r
al
s
u
p
er
io
r
ity
.
Ultim
ately
,
r
ath
er
th
an
co
m
p
etin
g
d
ir
ec
tly
with
lar
g
e
p
r
etr
ain
ed
lan
g
u
ag
e
m
o
d
els,
th
is
s
tu
d
y
p
r
o
v
es
th
at
r
ig
o
r
o
u
s
f
ea
tu
r
e
en
g
i
n
ee
r
i
n
g
co
m
b
in
e
d
with
in
tellig
en
t
d
en
s
ity
-
f
ilter
ed
o
v
er
s
am
p
lin
g
o
f
f
er
s
a
co
m
p
u
tatio
n
ally
ef
f
icien
t,
in
ter
p
r
etab
le,
an
d
h
ig
h
l
y
co
m
p
etitiv
e
alter
n
ativ
e
f
o
r
p
r
o
ce
s
s
in
g
s
k
ewe
d
Ar
ab
ic
tex
ts
.
3.
6
.
I
nte
rpre
t
a
t
io
n,
da
t
a
s
et
bia
s
,
a
nd
pra
ct
ica
l im
pli
ca
t
io
ns
T
h
e
ex
ec
u
tio
n
lo
g
s
an
d
clu
s
ter
m
etr
ics
p
r
o
v
id
e
p
r
o
f
o
u
n
d
i
n
s
ig
h
ts
in
to
wh
y
d
en
s
ity
-
b
ased
f
ilter
in
g
(
SMOT
E
-
R
ADI
ANT
)
s
u
cc
ee
d
s
wh
er
e
clu
s
ter
in
g
-
g
u
id
e
d
g
en
er
atio
n
(
ASTRA
-
SMOT
E
)
f
ails
in
th
e
s
p
ar
s
e,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
E
va
lu
a
tin
g
o
ve
r
s
a
mp
lin
g
meth
o
d
s
fo
r
imb
a
la
n
ce
d
A
r
a
b
ic
d
ia
lect
id
en
tifi
ca
tio
n
(
Ma
u
la
n
a
I
h
s
a
n
A
h
ma
d
)
267
5
,
6
4
4
-
d
im
en
s
io
n
al
s
p
ac
e.
ASTRA
-
SMOT
E
s
u
cc
ess
f
u
lly
ac
h
iev
ed
1
0
0
%
o
f
its
s
y
n
th
etic
g
en
er
atio
n
tar
g
ets
b
y
s
tr
ictly
in
ter
p
o
latin
g
in
s
id
e
"
s
af
e"
m
in
o
r
ity
clu
s
ter
s
,
r
e
s
u
l
tin
g
in
ex
ce
p
tio
n
ally
clea
n
i
n
ter
-
class
m
etr
ics
(
e.
g
.
,
Dav
ies
-
B
o
u
ld
in
in
d
e
x
d
r
o
p
p
ed
s
ig
n
if
ican
tly
to
1
.
6
4
)
.
Ho
wev
er
,
th
is
o
v
er
ly
co
n
s
er
v
ativ
e
g
en
er
atio
n
f
ailed
to
im
p
r
o
v
e
th
e
class
if
ier
.
Gr
ad
ien
t
b
o
o
s
tin
g
tr
ee
s
r
ely
o
n
f
in
d
in
g
o
p
tim
al
s
p
lits
at
th
e
b
o
u
n
d
ar
ies
b
etwe
en
class
es.
Gen
er
atin
g
d
ata
ex
clu
s
iv
ely
d
ee
p
in
s
id
e
s
af
e
clu
s
ter
s
p
r
o
v
i
d
es
n
o
n
ew
s
tr
u
ctu
r
al
in
f
o
r
m
atio
n
to
th
e
m
o
d
el'
s
d
ec
is
io
n
b
o
u
n
d
ar
y
.
C
o
n
v
er
s
ely
,
SMOT
E
-
R
ADI
ANT
allo
wed
s
tan
d
ar
d
in
ter
p
o
la
tio
n
ac
r
o
s
s
b
o
u
n
d
ar
ies
an
d
s
u
b
s
eq
u
en
tly
u
tili
ze
d
DB
S
C
AN
to
au
to
n
o
m
o
u
s
ly
d
etec
t
an
d
f
ilter
o
u
t
o
v
er
lap
p
in
g
n
o
is
e.
T
h
e
lo
g
s
r
ev
ea
l
th
at
R
ADI
ANT
in
ten
tio
n
ally
d
r
o
p
p
ed
5
,
3
3
6
s
y
n
th
etic
s
am
p
les
(
an
1
8
%
d
r
o
p
r
ate)
ac
r
o
s
s
th
e
m
in
o
r
ity
class
es
b
ec
au
s
e
th
ey
v
io
lated
d
en
s
ity
h
e
u
r
is
tics
.
B
y
s
ac
r
if
icin
g
tar
g
et
v
o
lu
m
e
to
p
u
r
g
e
b
o
u
n
d
ar
y
-
d
is
to
r
tin
g
a
n
o
m
alies,
R
ADI
AN
T
m
ain
tain
ed
th
e
n
ec
ess
ar
y
s
tr
u
ctu
r
al
ten
s
io
n
at
th
e
d
ec
is
io
n
b
o
u
n
d
a
r
y
,
d
ir
ec
tly
lea
d
in
g
to
th
e
s
tatis
tically
s
ig
n
if
ican
t p
er
f
o
r
m
an
ce
i
n
cr
ea
s
e
o
b
s
er
v
ed
in
L
i
g
h
tGB
M.
T
h
is
d
iv
er
g
en
ce
also
h
ig
h
lig
h
t
s
th
at
th
e
ef
f
ec
t
iv
en
ess
o
f
o
v
er
s
am
p
lin
g
is
s
tr
o
n
g
ly
m
o
d
el
-
d
ep
en
d
en
t.
L
ig
h
tGB
M’
s
leaf
-
wis
e
g
r
o
wth
en
ab
les
lo
ca
lized
s
p
litt
in
g
in
m
in
o
r
ity
-
d
en
s
e
r
eg
i
o
n
s
,
allo
w
in
g
it
to
s
elec
tiv
ely
ex
p
lo
it
th
e
clea
n
s
y
n
th
etic
b
o
u
n
d
a
r
y
s
am
p
les
p
r
o
v
id
ed
b
y
R
ADI
ANT
.
Me
an
wh
ile,
X
GB
o
o
s
t’
s
lev
el
-
wis
e
g
r
o
wth
ten
d
s
to
p
r
o
p
a
g
ate
s
y
n
th
etic
p
er
tu
r
b
atio
n
s
m
o
r
e
u
n
if
o
r
m
ly
ac
r
o
s
s
th
e
t
r
ee
,
r
en
d
er
in
g
it
h
ig
h
ly
s
en
s
itiv
e
to
b
o
u
n
d
ar
y
d
is
to
r
ti
o
n
ev
e
n
wh
en
d
en
s
ity
f
ilter
in
g
is
ap
p
lied
.
Similar
m
o
d
el
-
s
p
ec
if
ic
s
en
s
itiv
itie
s
h
av
e
b
ee
n
r
ep
o
r
ted
i
n
p
r
i
o
r
wo
r
k
[
2
5
]
,
[
2
9
]
,
[
3
5
]
.
Fu
r
th
e
r
m
o
r
e,
w
h
ile
clu
s
ter
in
g
-
g
u
id
e
d
m
eth
o
d
s
aim
to
r
estrict
s
y
n
th
esis
o
r
f
ilter
o
u
tlier
s
[
3
0
]
,
[
3
2
]
,
[
3
3
]
,
th
eir
a
d
v
an
tag
es
d
o
n
o
t
alwa
y
s
tr
an
s
late
s
ea
m
le
s
s
ly
to
ex
tr
em
ely
s
p
ar
s
e
T
F
-
I
DF
s
p
a
ce
s
,
wh
er
e
s
tan
d
ar
d
d
en
s
ity
esti
m
atio
n
ca
n
b
e
u
n
r
eliab
le
with
o
u
t
in
tellig
en
t
h
eu
r
is
tic
tu
n
in
g
[
2
5
]
,
[
3
3
]
,
[
3
4
]
.
Featu
r
e
-
lev
el
an
aly
s
is
s
u
p
p
o
r
ts
th
is
in
ter
p
r
etatio
n
.
L
ex
ical
f
ea
tu
r
es
(
T
F
–
I
DF
an
d
B
M2
5
)
ca
p
tu
r
e
d
ialec
tal
d
is
tin
ctio
n
s
ef
f
ec
tiv
ely
[
1
2
]
,
[
1
3
]
,
wh
ile
th
e
n
ewly
en
g
in
ee
r
e
d
s
tatis
tical
f
ea
tu
r
es
-
s
u
ch
as
NB
-
SVM
p
r
io
r
s
an
d
OOF
s
tack
in
g
p
r
o
b
a
b
ilit
ies
-
p
r
o
v
id
e
p
o
wer
f
u
l,
co
m
p
lem
e
n
tar
y
d
ec
is
io
n
cu
es
[
1
7
]
,
[
1
9
]
.
T
h
ese
f
ea
tu
r
es
in
ter
ac
t
c
o
n
s
tr
u
ctiv
ely
with
SMOT
E
-
R
ADI
ANT
in
L
i
g
h
tGB
M,
s
o
lid
if
y
in
g
th
e
f
ea
tu
r
e
h
ier
ar
ch
y
with
o
u
t a
m
p
lify
in
g
s
en
s
itiv
ity
to
o
v
er
lap
p
in
g
lex
i
ca
l n
o
is
e
[
2
5
]
,
[
2
9
]
.
I
t
is
im
p
o
r
tan
t
t
o
ac
k
n
o
wle
d
g
e
th
at
lex
ic
o
n
-
b
ased
an
d
f
r
eq
u
e
n
cy
-
d
r
iv
en
r
ep
r
esen
tatio
n
s
m
ay
in
ad
v
er
ten
tly
en
c
o
d
e
t
o
p
ical,
s
tan
ce
-
r
elate
d
,
o
r
s
en
tim
en
t
-
b
ea
r
in
g
p
atter
n
s
i
n
ad
d
itio
n
to
p
u
r
ely
s
tr
u
ctu
r
al
d
ialec
tal
m
ar
k
er
s
.
T
h
is
is
p
ar
t
icu
lar
ly
tr
u
e
in
s
o
cial
m
ed
ia
co
r
p
o
r
a
,
wh
er
e
d
ialec
t
u
s
ag
e
o
f
ten
co
-
v
ar
ies
with
af
f
ec
tiv
e
ex
p
r
ess
io
n
an
d
d
is
cu
s
s
io
n
th
em
es
[
1
6
]
,
[
2
4
]
.
Fo
r
in
s
tan
ce
,
ce
r
tain
d
ialec
ts
m
ay
b
e
d
is
p
r
o
p
o
r
tio
n
ately
ass
o
ciate
d
with
s
p
ec
if
ic
p
o
liti
ca
l,
c
u
ltu
r
al,
o
r
s
o
cial
to
p
ics,
i
n
tr
o
d
u
cin
g
laten
t
to
p
ic
b
ias
in
to
th
e
f
ea
tu
r
e
d
is
tr
ib
u
tio
n
s
.
Ho
wev
er
,
b
ec
au
s
e
o
u
r
5
,
6
4
4
-
d
im
e
n
s
io
n
al
f
ea
tu
r
e
ex
t
r
a
ctio
n
p
ip
elin
e
an
d
v
alid
atio
n
s
p
lits
wer
e
s
tr
ictly
f
ix
ed
ac
r
o
s
s
all
r
eb
alan
ci
n
g
s
ce
n
ar
io
s
,
a
n
y
s
u
ch
b
ias
r
em
ain
ed
c
o
n
s
tan
t
th
r
o
u
g
h
o
u
t
th
e
e
x
p
er
im
e
n
ts
.
C
o
n
s
eq
u
en
tly
,
th
e
co
m
p
ar
ativ
e
ef
f
ec
ts
o
b
s
er
v
ed
am
o
n
g
th
e
Or
ig
in
al,
SMOT
E
,
ASTRA
-
SM
OT
E
,
SMOT
E
-
R
ADI
ANT
,
an
d
C
lass
W
eig
h
t
co
n
f
ig
u
r
atio
n
s
ca
n
b
e
r
eliab
l
y
attr
ib
u
ted
to
t
h
e
im
b
alan
ce
-
h
an
d
lin
g
s
tr
ateg
ies
th
em
s
elv
es r
ath
er
th
an
u
n
co
n
t
r
o
lled
s
h
if
ts
in
lex
ical
b
ias.
Fro
m
a
p
r
ac
tical
s
tan
d
p
o
in
t,
ac
h
iev
in
g
a
s
tatis
tically
s
ig
n
if
ican
t
im
p
r
o
v
em
e
n
t
in
m
in
o
r
ity
d
ialec
t
r
ec
o
g
n
itio
n
with
o
u
t
d
eg
r
ad
in
g
th
e
m
ajo
r
ity
class
h
as
p
r
o
f
o
u
n
d
r
ea
l
-
wo
r
ld
im
p
licatio
n
s
.
R
eliab
le
d
ialec
t
id
en
tific
atio
n
is
a
cr
itical
u
p
s
tr
ea
m
co
m
p
o
n
en
t
f
o
r
d
o
wn
s
tr
e
am
NL
P
ap
p
licatio
n
s
s
u
ch
as
s
en
tim
en
t
an
aly
s
is
,
m
ac
h
in
e
tr
an
s
latio
n
,
an
d
s
p
ee
ch
tech
n
o
lo
g
ies.
B
y
s
u
cc
ess
f
u
lly
liftin
g
th
e
r
ec
all
o
f
u
n
d
er
r
ep
r
esen
ted
d
ialec
ts
(
e.
g
.
,
J
o
r
d
a
n
ian
)
s
af
ely
,
m
eth
o
d
o
lo
g
ies
lik
e
R
ADI
ANT
en
s
u
r
e
th
at
d
ia
lecta
l
m
is
clas
s
if
i
ca
tio
n
er
r
o
r
s
d
o
n
o
t
p
r
o
p
a
g
ate
th
r
o
u
g
h
p
r
o
ce
s
s
in
g
p
ip
elin
es,
th
er
eb
y
p
r
ev
e
n
tin
g
AI
s
y
s
tem
s
f
r
o
m
in
ad
v
er
te
n
tly
m
ar
g
in
alizin
g
s
p
ec
if
ic
Ar
ab
ic
-
s
p
ea
k
in
g
r
e
g
io
n
s
.
R
ath
er
th
a
n
d
i
r
ec
tly
co
m
p
etin
g
with
lar
g
e
p
r
etr
ain
e
d
m
o
d
els
s
u
ch
a
s
Ar
aBER
T
,
AR
B
E
R
T
,
an
d
MA
R
B
E
R
T
[
2
0
]
-
[
2
3
]
,
th
is
s
tu
d
y
p
r
o
v
i
d
es m
eth
o
d
o
lo
g
ical
clar
i
ty
b
y
ch
a
r
ac
ter
izin
g
o
v
er
s
am
p
lin
g
b
eh
a
v
io
r
with
in
a
tr
an
s
p
ar
en
t,
in
te
r
p
r
etab
le
f
ea
tu
r
e
-
b
ased
f
r
am
ewo
r
k
[
2
5
]
,
[
3
5
]
.
Sev
er
al
lim
itatio
n
s
r
em
ain
.
T
h
e
an
aly
s
is
is
r
estricte
d
to
th
e
Sh
am
i
C
o
r
p
u
s
an
d
f
o
c
u
s
es
o
n
s
p
ec
if
ic
g
r
ad
ien
t
-
b
o
o
s
tin
g
ar
c
h
itectu
r
e
s
.
Fu
tu
r
e
wo
r
k
s
h
o
u
ld
ex
te
n
d
ev
alu
atio
n
to
b
r
o
ad
er
Ar
a
b
ic
d
ialec
t
b
en
ch
m
ar
k
s
(
e.
g
.
,
NADI
[
3
]
an
d
MA
D
AR
[
4
]
)
,
ex
p
lo
r
e
em
b
ed
d
in
g
-
awa
r
e
ad
ap
tiv
e
o
v
er
s
am
p
l
in
g
m
eth
o
d
s
,
an
d
in
v
esti
g
ate
h
y
b
r
i
d
co
s
t
-
s
en
s
itiv
e
tr
an
s
f
o
r
m
er
-
b
ased
a
p
p
r
o
ac
h
es.
4.
CO
NCLU
SI
O
N
T
h
is
s
tu
d
y
p
r
esen
ted
a
r
ig
o
r
o
u
s
ly
co
n
tr
o
lled
,
leak
a
g
e
-
f
r
e
e
co
m
p
ar
ativ
e
ev
alu
atio
n
o
f
im
b
alan
ce
m
itig
atio
n
s
tr
ateg
ies
f
o
r
Ar
ab
i
c
d
ialec
t
id
en
tific
atio
n
o
n
th
e
Sh
am
i
C
o
r
p
u
s
.
Op
er
atin
g
wit
h
in
a
h
ig
h
l
y
s
p
ar
s
e
5
,
6
4
4
-
d
im
en
s
io
n
al
tex
t
r
ep
r
e
s
en
tatio
n
,
we
d
em
o
n
s
tr
ated
th
at
co
n
v
en
tio
n
al
m
itig
atio
n
tech
n
iq
u
es
-
s
u
ch
as
u
n
co
n
s
tr
ain
e
d
SMOT
E
,
clu
s
ter
-
r
estricte
d
ASTRA
-
SMOT
E
,
an
d
co
s
t
-
s
en
s
itiv
e
class
we
ig
h
tin
g
-
lar
g
ely
f
ailed
to
r
ef
in
e
d
ec
is
io
n
b
o
u
n
d
ar
ies
.
T
h
ese
m
eth
o
d
s
f
r
eq
u
e
n
tly
s
ac
r
if
iced
m
ajo
r
ity
-
class
p
r
ec
is
io
n
o
r
in
tr
o
d
u
ce
d
b
o
u
n
d
ar
y
-
d
is
to
r
tin
g
n
o
is
e
in
a
n
attem
p
t to
ar
tific
ially
in
f
late
m
in
o
r
ity
r
ec
all.
Ho
wev
er
,
SMOT
E
-
R
ADI
AN
T
-
a
s
tr
ateg
y
th
at
co
u
p
les
g
en
er
ativ
e
o
v
er
s
am
p
lin
g
with
d
e
n
s
ity
-
b
ased
n
o
is
e
n
eu
tr
aliza
tio
n
(
DB
SC
AN)
-
s
u
cc
ess
f
u
lly
b
r
o
k
e
t
h
is
d
im
en
s
io
n
ality
b
o
ttlen
ec
k
.
B
y
a
u
to
n
o
m
o
u
s
ly
f
ilter
in
g
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
4
3
,
No
.
1
,
Ju
ly
20
2
6
:
259
-
27
0
268
o
u
t
1
8
%
o
f
o
v
e
r
lap
p
in
g
s
y
n
t
h
etic
an
o
m
alies
at
t
h
e
in
ter
-
cl
ass
b
o
u
n
d
ar
ies,
R
ADI
ANT
p
r
o
v
id
ed
s
tr
u
ctu
r
ally
s
o
u
n
d
d
ata
a
u
g
m
e
n
tatio
n
.
Val
id
ated
th
r
o
u
g
h
a
r
o
b
u
s
t
1
,
0
0
0
-
iter
atio
n
b
o
o
ts
t
r
ap
p
in
g
p
r
o
ce
d
u
r
e
o
n
th
e
h
el
d
-
o
u
t
test
s
et,
SMOT
E
-
R
ADI
ANT
y
ield
ed
a
s
tatis
tically
s
ig
n
if
ican
t
m
ac
r
o
-
F1
im
p
r
o
v
em
e
n
t
f
o
r
L
ig
h
tGB
M
(
p
<
0
.
0
0
1
)
ac
co
m
p
an
ied
b
y
a
lar
g
e
ef
f
ec
t
s
ize
(
r
=
0
.
5
1
1
)
.
T
h
is
p
r
o
v
id
es
co
m
p
ellin
g
e
v
id
en
ce
th
at
wh
ile
g
en
er
atin
g
s
y
n
th
etic
t
ex
t
d
ata
in
s
p
ar
s
e
s
p
ac
es
is
in
h
er
en
tly
r
is
k
y
,
s
tr
ateg
ically
n
e
u
tr
alizin
g
o
v
e
r
lap
p
in
g
n
o
is
e
ca
n
s
af
ely
r
escu
e
m
in
o
r
ity
d
ial
ec
ts
with
o
u
t m
ajo
r
ity
-
class
d
eg
r
ad
atio
n
.
T
h
ese
f
in
d
in
g
s
clar
if
y
th
e
c
o
m
p
lex
tr
a
d
e
-
o
f
f
s
o
f
im
b
ala
n
ce
m
itig
atio
n
,
h
ig
h
lig
h
tin
g
th
at
m
o
d
el
ar
ch
itectu
r
e
(
leaf
-
wis
e
v
s
.
le
v
el
-
wis
e)
,
f
ea
tu
r
e
d
esig
n
,
an
d
o
v
e
r
s
am
p
lin
g
s
tr
ateg
ies
in
t
er
ac
t
in
n
o
n
-
tr
i
v
ial
way
s
.
B
ey
o
n
d
d
ialec
t
id
en
tifi
ca
tio
n
,
th
is
s
tu
d
y
u
n
d
er
s
co
r
e
s
th
e
n
ec
ess
ity
o
f
ca
r
ef
u
lly
v
alid
ated
im
b
alan
ce
m
itig
atio
n
f
o
r
b
u
ild
in
g
f
air
an
d
r
o
b
u
s
t
Ar
ab
ic
NL
P
s
y
s
tem
s
,
en
s
u
r
in
g
th
at
r
eg
io
n
al
in
clu
s
iv
ity
is
m
ain
tain
ed
in
d
o
wn
s
tr
ea
m
a
p
p
licatio
n
s
.
ACK
NO
WL
E
DG
M
E
N
T
S
T
h
is
wo
r
k
was
s
u
p
p
o
r
ted
b
y
th
e
Dep
ar
tm
e
n
t
o
f
C
o
m
p
u
te
r
Scien
ce
an
d
E
lectr
o
n
ics,
U
n
iv
er
s
itas
Gad
jah
Ma
d
a
u
n
d
er
t
h
e
Pu
b
lic
atio
n
Fu
n
d
in
g
Yea
r
2
0
2
6
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
T
h
is
wo
r
k
was
s
u
p
p
o
r
ted
b
y
th
e
Dep
ar
tm
e
n
t
o
f
C
o
m
p
u
te
r
Scien
ce
an
d
E
lectr
o
n
ics,
U
n
iv
er
s
itas
Gad
jah
Ma
d
a
u
n
d
er
t
h
e
Pu
b
lic
atio
n
Fu
n
d
in
g
Yea
r
2
0
2
6
.
AUTHO
R
CO
NT
RI
B
UT
I
O
NS ST
A
T
E
M
E
N
T
T
h
is
s
tu
d
y
f
o
llo
ws
th
e
C
o
n
tr
ib
u
to
r
R
o
les
T
ax
o
n
o
m
y
(
C
R
ed
iT)
.
T
h
e
co
n
tr
i
b
u
tio
n
s
o
f
ea
c
h
au
th
o
r
ar
e
d
etailed
as f
o
llo
ws:
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
Ma
u
lan
a
I
h
s
an
Ah
m
ad
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
Ain
a
Mu
s
d
h
o
lifa
h
✓
✓
✓
✓
✓
✓
✓
✓
✓
Ar
if
Nu
r
wid
y
an
t
o
r
o
✓
✓
✓
✓
✓
✓
✓
C
:
C
o
n
c
e
p
t
u
a
l
i
z
a
t
i
o
n
M
:
M
e
t
h
o
d
o
l
o
g
y
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
d
a
t
i
o
n
Fo
:
Fo
r
mal
a
n
a
l
y
s
i
s
I
:
I
n
v
e
s
t
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
su
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
s
i
o
n
P
:
P
r
o
j
e
c
t
a
d
mi
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
si
t
i
o
n
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
T
h
e
au
t
h
o
r
s
d
ec
lar
e
th
at
th
e
y
h
av
e
n
o
k
n
o
wn
c
o
m
p
etin
g
f
in
an
cial
in
ter
ests
o
r
p
er
s
o
n
al
r
el
atio
n
s
h
ip
s
th
at
co
u
ld
h
av
e
ap
p
ea
r
ed
t
o
in
f
lu
en
ce
th
e
wo
r
k
r
e
p
o
r
te
d
in
t
h
is
p
ap
er
.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
d
ata
th
at
s
u
p
p
o
r
t
th
e
f
i
n
d
in
g
s
o
f
th
is
s
tu
d
y
ar
e
p
u
b
licly
av
ailab
le
as
th
e
Sh
am
i
C
o
r
p
u
s
[
6
]
.
T
o
s
u
p
p
o
r
t
tr
an
s
p
ar
e
n
t
an
d
r
e
p
r
o
d
u
cib
le
r
esear
ch
,
t
h
e
co
m
p
lete
s
o
u
r
ce
co
d
e
an
d
e
x
p
er
i
m
en
tal
p
ip
elin
e
ar
e
p
u
b
licly
av
ailab
le
at:
h
tt
p
s
://g
ith
u
b
.
co
m
/m
o
el
-
ih
s
an
/ar
ab
ic
-
d
ialec
t
-
id
en
tific
atio
n
.
RE
F
E
R
E
NC
E
S
[
1
]
Y
.
M
a
t
r
a
n
e
,
F
.
B
e
n
a
b
b
o
u
,
a
n
d
N
.
S
a
e
l
,
“
A
s
y
s
t
e
m
a
t
i
c
l
i
t
e
r
a
t
u
r
e
r
e
v
i
e
w
o
f
A
r
a
b
i
c
d
i
a
l
e
c
t
se
n
t
i
m
e
n
t
a
n
a
l
y
si
s
,
”
J
o
u
r
n
a
l
o
f
K
i
n
g
S
a
u
d
U
n
i
v
e
rsi
t
y
-
C
o
m
p
u
t
e
r
a
n
d
I
n
f
o
r
m
a
t
i
o
n
S
c
i
e
n
c
e
s
,
v
o
l
.
3
5
,
n
o
.
6
,
p
.
1
0
1
5
7
0
,
Ju
n
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
j
k
s
u
c
i
.
2
0
2
3
.
1
0
1
5
7
0
.
[
2
]
O
.
F
.
Za
i
d
a
n
a
n
d
C
.
C
a
l
l
i
s
o
n
-
B
u
r
c
h
,
“
A
r
a
b
i
c
d
i
a
l
e
c
t
i
d
e
n
t
i
f
i
c
a
t
i
o
n
,”
C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
st
i
c
s
,
v
o
l
.
4
0
,
n
o
.
1
,
p
p
.
1
7
1
–
2
0
2
,
M
a
r
.
2
0
1
4
,
d
o
i
:
1
0
.
1
1
6
2
/
C
O
LI
_
a
_
0
0
1
6
9
.
[
3
]
M
.
A
b
d
u
l
-
M
a
g
e
e
d
,
C
.
Z
h
a
n
g
,
A
.
E
l
mad
a
n
y
,
H
.
B
o
u
a
mo
r
,
a
n
d
N
.
H
a
b
a
sh
,
“
N
A
D
I
2
0
2
2
:
Th
e
T
h
i
r
d
N
u
a
n
c
e
d
A
r
a
b
i
c
d
i
a
l
e
c
t
i
d
e
n
t
i
f
i
c
a
t
i
o
n
s
h
a
r
e
d
t
a
s
k
,
”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
Fi
f
t
h
Ar
a
b
i
c
N
a
t
u
ra
l
L
a
n
g
u
a
g
e
Pr
o
c
e
ss
i
n
g
Wo
r
k
sh
o
p
,
S
t
r
o
u
d
sb
u
r
g
,
P
A
,
U
S
A
:
A
sso
c
i
a
t
i
o
n
f
o
r
C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
st
i
c
s
,
2
0
2
0
,
p
p
.
9
7
–
1
1
0
.
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
2
.
w
a
n
l
p
-
1
.
9
.
[
4
]
H
.
B
o
u
a
mo
r
e
t
a
l
.
,
“
T
h
e
m
a
d
a
r
A
r
a
b
i
c
d
i
a
l
e
c
t
c
o
r
p
u
s
a
n
d
l
e
x
i
c
o
n
,
”
i
n
L
RE
C
2
0
1
8
-
1
1
t
h
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
L
a
n
g
u
a
g
e
Re
so
u
r
c
e
s
a
n
d
Ev
a
l
u
a
t
i
o
n
,
2
0
1
8
,
p
p
.
3
3
8
7
–
3
3
9
6
.
Evaluation Warning : The document was created with Spire.PDF for Python.