I
nd
o
ne
s
ia
n J
o
urna
l o
f
E
lect
rica
l En
g
ineering
a
nd
Co
m
p
u
t
er
Science
Vo
l.
21
,
No
.
1
,
J
an
u
ar
y
2
0
2
1
,
p
p
.
5
7
4
~5
8
1
I
SS
N:
2
5
02
-
4
7
5
2
,
DOI
: 1
0
.
1
1
5
9
1
/i
j
ee
cs.v
2
1
.i
1
.
pp
5
7
4
-
581
574
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ee
cs.ia
esco
r
e.
co
m
Ana
ly
sis
of
m
a
c
hi
ne learning
alg
o
rith
m
s for
cha
ra
cte
r
recog
nition: a ca
s
e study o
n han
dw
ritt
en
digit
recog
nition
O
w
a
is
M
ujt
a
ba
K
ha
nd
a
y
,
D
r.
Sa
m
a
d Da
dv
a
nd
i
po
ur
In
stit
u
te
o
f
In
f
o
rm
a
ti
o
n
S
c
ien
c
e
s,
Un
iv
e
rsit
y
o
f
M
isk
o
lc,
M
isk
o
lc
,
Hu
n
g
a
ry
Art
icle
I
nfo
AB
ST
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
u
n
5
,
2
0
2
0
R
ev
i
s
ed
A
u
g
3
,
2
0
2
0
A
cc
ep
ted
A
u
g
1
6
,
2
0
2
0
T
h
is
p
a
p
e
r
c
o
v
e
rs
th
e
w
o
rk
d
o
n
e
in
h
a
n
d
w
rit
ten
d
ig
it
re
c
o
g
n
it
i
o
n
a
n
d
th
e
v
a
rio
u
s
c
las
sif
ier
s
th
a
t
h
a
v
e
b
e
e
n
d
e
v
e
lo
p
e
d
.
M
e
th
o
d
s
li
k
e
M
L
P
,
S
V
M
,
Ba
y
e
sia
n
n
e
tw
o
rk
s,
a
n
d
Ra
n
d
o
m
f
o
re
sts
we
re
d
isc
u
ss
e
d
w
it
h
th
e
i
r
a
c
c
u
ra
c
y
a
n
d
a
re
e
m
p
iri
c
a
ll
y
e
v
a
lu
a
ted
.
Bo
o
ste
d
L
e
tNe
t
4
,
a
n
e
n
se
m
b
le
o
f
v
a
rio
u
s
c
las
si
f
iers
,
h
a
s sh
o
w
n
m
a
x
i
m
u
m
e
ff
ici
e
n
c
y
a
m
o
n
g
th
e
se
m
e
th
o
d
s.
K
ey
w
o
r
d
s
:
Han
d
w
r
itte
n
d
ig
it r
ec
o
g
n
itio
n
ML
P
OC
R
R
an
d
o
m
f
o
r
ests
SVM
T
h
is
is
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
:
O
w
a
is
M
u
j
tab
a
Kh
an
d
a
y
I
n
s
tit
u
te
o
f
I
n
f
o
r
m
atio
n
Sc
ien
ce
s
,
T
h
e
Un
iv
er
s
i
t
y
o
f
Mis
k
o
l
c
E
g
y
te
m
v
ar
o
s
,
3
5
2
5
,
Misk
o
lc
,
Hu
n
g
ar
y
E
m
ail: a
ito
w
ai
s
@
u
n
i
-
m
is
k
o
lc.
h
u
1.
I
NT
RO
D
UCT
I
O
N
Sig
n
i
f
ica
n
t
ac
h
ie
v
e
m
e
n
t
s
h
av
e
b
ee
n
m
ad
e
i
n
t
h
e
o
p
tical
c
h
ar
ac
ter
r
ec
o
g
n
itio
n
(
OC
T
)
tech
n
o
lo
g
y
,
in
cl
u
d
in
g
h
an
d
w
r
itte
n
r
ec
o
g
n
i
tio
n
o
f
d
i
g
it
s
.
I
ts
r
o
le
is
u
b
iq
u
i
to
u
s
i
n
r
ea
l
-
ti
m
e
e
-
p
r
o
ce
s
s
i
n
g
o
f
th
e
d
a
ta
s
u
ch
a
s
r
ea
d
in
g
t
h
e
zip
co
d
es
a
n
d
s
o
r
tin
g
th
e
p
o
s
t
m
ail
s
,
b
an
k
ch
ec
k
p
r
o
ce
s
s
in
g
,
e
-
co
m
m
er
c
e
an
d
e
v
en
s
t
u
d
en
t
ac
h
iev
e
m
e
n
t
r
ec
o
g
n
i
tio
n
,
etc.
[
1
,
2
]
.
C
o
n
s
id
er
ab
le
p
r
o
g
r
ess
h
as
b
ee
n
ac
h
ie
v
ed
b
ec
au
s
e
o
f
th
e
d
ev
elo
p
m
e
n
t
s
an
d
ad
v
an
ce
m
e
n
ts
i
n
th
e
co
m
p
u
tat
io
n
al
p
o
w
er
o
f
co
m
p
u
t
er
s
an
d
th
e
av
a
ilab
ilit
y
o
f
m
o
r
e
m
as
s
i
v
e
d
atasets
th
at
ar
e
u
s
ed
f
o
r
tr
ain
i
n
g
a
n
d
test
in
g
p
u
r
p
o
s
es
[
3
]
.
No
w
ad
a
y
s
,
h
a
n
d
w
r
itte
n
r
ec
o
g
n
itio
n
s
ar
e
ev
en
u
s
ed
f
o
r
co
m
m
u
n
icatio
n
p
u
r
p
o
s
es.
OC
T
d
o
es
h
av
e
t
h
e
ca
p
ab
ilit
y
a
n
d
p
o
w
er
to
p
illar
th
e
p
ap
er
les
s
en
v
ir
o
n
m
e
n
t
b
y
p
r
o
ce
s
s
in
g
th
e
e
x
i
s
ti
n
g
p
ap
er
d
o
cu
m
en
t
s
[
4
]
.
Han
d
w
r
it
ten
d
ig
it
r
ec
o
g
n
itio
n
in
v
o
lv
es
id
en
ti
f
y
i
n
g
1
0
ch
ar
ac
ter
s
,
i.e
.
,
0
-
9
,
b
u
t
th
e
in
p
u
t
is
s
en
s
iti
v
e
to
th
e
en
v
ir
o
n
m
en
tal
n
o
is
e.
[
5
,
6
]
.
On
e
o
f
t
h
e
f
o
r
e
m
o
s
t
task
s
i
s
to
id
en
ti
f
y
th
e
lo
ca
l
ar
e
as
f
o
r
o
b
tain
in
g
d
is
cr
i
m
i
n
ati
n
g
f
e
atu
r
es.
Var
io
u
s
s
a
m
p
li
n
g
tec
h
n
iq
u
es
h
a
v
e
b
ee
n
d
ev
elo
p
ed
to
f
in
d
in
g
th
e
s
e
l
o
ca
l
r
eg
io
n
s
[
7
]
.
A
ls
o
,
th
e
d
atasets
ar
e
ev
e
n
v
a
g
u
e
b
ec
au
s
e
th
e
w
r
iti
n
g
an
d
o
r
ien
tatio
n
d
i
f
f
er
f
r
o
m
p
er
s
o
n
to
p
er
s
o
n
.
I
t
also
h
ap
p
en
s
s
o
m
et
i
m
e
s
;
o
n
e
ca
n
n
o
t
e
v
en
r
ec
o
g
n
ize
th
e
h
an
d
w
r
itte
n
ch
ar
ac
ter
s
w
r
itte
n
b
y
h
i
m
s
el
f
.
Ot
h
er
p
r
o
b
lem
s
in
cl
u
d
e
s
lip
p
in
g
o
f
th
e
p
en
,
letter
in
s
er
tio
n
,
o
r
o
m
is
s
io
n
,
w
h
ich
g
r
ea
tl
y
co
m
p
licate
th
e
tas
k
[
8
]
.
Ho
w
e
v
er
,
a
t
y
p
ical
r
ec
o
g
n
itio
n
s
y
s
te
m
is
b
u
ilt
to
f
o
cu
s
o
n
o
n
l
y
a
s
u
b
s
et
o
f
t
h
e
p
r
o
b
le
m
[
9
]
.
T
h
e
m
o
s
t
co
m
m
o
n
m
et
h
o
d
f
o
r
b
u
ild
in
g
t
h
e
h
a
n
d
w
r
itte
n
d
ig
it
r
ec
o
g
n
izer
is
u
s
i
n
g
a
m
u
lti
-
la
y
er
p
er
ce
p
tr
o
n
(
ML
P
)
[
1
0
,
1
1
]
k
NN
an
d
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
in
e
SVM.
Var
io
u
s
o
th
er
tech
n
iq
u
es
w
er
e
d
ev
elo
p
ed
u
s
i
n
g
d
if
f
er
en
t
tec
h
n
iq
u
es
w
it
h
M
L
P
s
tr
u
ct
u
r
e.
R
e
n
ata
F.P
Nev
es
an
d
et
al.
p
r
o
p
o
s
ed
a
m
et
h
o
d
f
o
r
h
an
d
w
r
itte
n
d
ig
it
r
ec
o
g
n
itio
n
,
w
h
ic
h
i
m
p
r
o
v
ed
t
h
e
ef
f
ici
en
c
y
r
ates
co
m
p
ar
ed
to
t
h
e
ML
P
a
n
d
h
y
b
r
id
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esia
n
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4752
A
n
a
lysi
s
o
f m
a
ch
in
e
lea
r
n
in
g
a
lg
o
r
ith
ms fo
r
ch
a
r
a
cter reco
g
n
itio
n
:
a
ca
s
e
…
(
Ow
a
is
Mu
jt
a
b
a
K
h
a
n
d
a
y
)
575
class
i
f
ier
s
o
v
er
th
e
NI
ST
SD1
9
d
ig
it
d
atab
ase
[
1
2
]
.
C
h
en
g
-
L
i
n
L
i
u
an
d
et
al.
co
m
b
in
ed
e
ig
h
t
clas
s
i
f
ie
r
s
a
n
d
w
it
h
te
n
f
ea
t
u
r
e
v
ec
to
r
s
.
T
h
e
y
C
E
NP
AR
MI
,
C
E
D
AR
,
an
d
MN
I
ST
d
atab
ases
w
er
e
tes
te
d
.
SVC
w
it
h
R
B
F
k
er
n
el
(
SV
C
-
R
B
F)
g
a
v
e
th
e
h
ig
h
e
s
t
ac
cu
r
ac
y
in
m
o
s
t
ca
s
e
s
b
u
t
o
n
th
e
co
s
t
o
f
s
to
r
ag
e
a
n
d
co
m
p
u
tatio
n
[
1
3
]
.
A
s
t
u
d
y
w
as
d
o
n
e
in
[
1
4
]
s
h
o
w
ed
th
e
b
u
n
d
le
o
f
f
ea
t
u
r
e
ex
t
r
ac
tio
n
tech
n
iq
u
es
a
n
d
w
er
e
ev
alu
a
ted
u
s
i
n
g
t
h
e
b
en
ch
m
ar
k
d
atasets
a
v
ailab
le
p
u
b
licall
y
;
t
h
e
m
et
h
o
d
s
[
1
5
]
an
d
[
1
6
]
o
u
tp
er
f
o
r
m
ed
t
h
e
o
t
h
e
r
m
et
h
o
d
s
a
v
ailab
le
in
t
h
e
liter
atu
r
e
b
y
s
h
o
w
i
n
g
th
e
ac
cu
r
ac
y
o
f
9
9
.
0
3
%
an
d
9
8
.
7
5
%
r
esp
ec
tiv
el
y
.
T
h
e
f
o
llo
w
in
g
r
esear
c
h
m
et
h
o
d
s
ar
e
ap
p
lied
in
th
e
p
ap
er
.
2.
RE
S
E
ARCH
M
E
T
H
O
D
2
.
1
.
M
ulti
-
l
a
y
er
perc
ept
ro
ns
An
M
L
P
is
a
n
ar
ti
f
icial
n
e
u
r
al
n
et
w
o
r
k
clas
s
i
f
ier
th
a
t
e
m
p
lo
y
s
t
h
e
m
o
d
eli
n
g
o
f
h
u
m
an
b
io
lo
g
ical
n
eu
r
o
n
s
[
1
7
]
.
I
t
is
a
f
ee
d
f
o
r
w
ar
d
n
et
w
o
r
k
t
h
at
co
m
p
u
tes
t
h
e
s
ig
m
o
id
f
u
n
ctio
n
o
f
t
h
e
w
ei
g
h
ted
s
u
m
o
f
all
t
h
e
in
p
u
t
n
eu
r
o
n
s
,
as
g
iv
e
n
i
n
(
1
)
.
T
h
e
ty
p
ical
M
L
P
lo
o
k
s
li
k
e
as
g
iv
e
n
i
n
Fi
g
u
r
e
1
.
(
1
)
Fig
u
r
e
1
.
A
r
ch
itectu
r
e
o
f
M
L
P
W
h
er
e
th
e
w
eig
h
t
o
f
t
h
e
in
p
u
t
is
,
is
th
e
in
p
u
t
co
m
in
g
f
r
o
m
t
h
e
n
eu
r
o
n
,
an
d
is
th
e
b
ias.
T
h
e
s
ig
m
o
id
f
u
n
c
tio
n
i
s
also
c
alled
as
th
e
ac
tiv
a
tio
n
f
u
n
ct
io
n
.
T
h
e
cl
ass
i
f
ier
co
n
s
i
s
ts
o
f
t
h
r
ee
la
y
er
s
,
t
h
e
in
p
u
t
la
y
er
,
t
h
e
o
u
tp
u
t
la
y
er
,
a
n
d
t
h
e
h
id
d
en
la
y
er
.
T
h
e
n
et
w
o
r
k
is
f
u
ll
y
co
n
n
ec
ted
,
a
n
d
ea
c
h
la
y
er
h
as
a
ce
r
tai
n
n
u
m
b
er
o
f
n
eu
r
o
n
s
.
T
h
e
n
u
m
b
er
o
f
i
n
p
u
t
an
d
o
u
tp
u
t
n
eu
r
o
n
s
d
ep
en
d
s
u
p
o
n
th
e
attr
ib
u
te
s
a
n
d
t
h
e
n
u
m
b
er
o
f
class
es
ex
i
s
ti
n
g
in
th
e
d
ata
s
et
.
I
n
th
e
ca
s
e
o
f
th
e
d
ig
it
clas
s
i
f
ier
,
th
e
o
u
tp
u
t
n
e
u
r
o
n
s
ar
e
1
0
.
Fo
r
an
y
M
-
clas
s
class
i
f
icatio
n
,
th
e
m
o
d
el
h
a
s
m
o
u
tp
u
ts
.
O
n
i
n
p
u
t
p
atter
n
o
f
T
h
e
o
u
tp
u
t
o
f
clas
s
is
co
m
p
u
ted
b
y
:
[
∑
(
)
]
=
[
∑
]
(
2
)
is
th
e
n
u
m
b
er
o
f
h
id
d
en
u
n
it
s
an
d
ar
e
th
e
co
n
n
ec
ti
n
g
w
ei
g
h
ts
o
f
t
h
e
o
u
tp
u
t
la
y
er
a
n
d
th
e
h
id
d
en
la
y
er
.
T
h
e
M
L
P
m
o
d
el
is
tr
ain
ed
a
g
ai
n
s
t
a
test
s
et,
a
n
d
it a
u
to
m
at
icall
y
lear
n
s
to
ad
j
u
s
t
t
h
e
w
ei
g
h
ts
f
o
r
ea
ch
co
n
n
ec
tio
n
[
1
8
]
.
T
h
e
lear
n
in
g
p
r
o
ce
s
s
u
s
ed
i
s
t
h
e
er
r
o
r
b
ac
k
-
p
r
o
p
ag
atio
n
al
g
o
r
ith
m
,
w
h
ic
h
ad
j
u
s
t
s
t
h
e
co
n
n
ec
ti
n
g
w
eig
h
t
s
b
y
m
in
i
m
i
zin
g
t
h
e
m
ea
n
s
q
u
ar
e
er
r
o
r
(
M
SE)
o
v
er
a
s
et
o
f
tr
ain
i
n
g
e
x
a
m
p
les [
19
].
{
[
]
}
(
3
)
I
n
p
u
t X
:
A
s
e
t o
f
ac
ti
v
a
tio
n
s
f
o
r
th
e
in
p
u
t la
y
er
,
Feed
f
o
r
w
ar
d
: Fo
r
ea
ch
l=2
,
3
,
…,
L
co
m
p
u
te
Ou
tp
u
t e
r
r
o
r
: Co
m
p
u
te
th
e
v
e
cto
r
B
ac
k
p
r
o
p
ag
ate
th
e
er
r
o
r
: Fo
r
ea
ch
co
m
p
u
te
Ou
tp
u
t: T
h
e
g
r
ad
ien
t o
f
t
h
e
co
s
t f
u
n
ctio
n
:
,
an
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4752
I
n
d
o
n
esia
n
J
E
lec
E
n
g
&
C
o
m
p
Sci,
Vo
l.
21
,
No
.
1
,
J
an
u
ar
y
2
0
2
1
:
5
7
4
-
581
576
2
.
2
.
Su
pp
o
rt
v
ec
t
o
r
m
a
chine
(
SVM
)
SVM
is
a
s
u
p
er
v
is
ed
lear
n
i
n
g
m
o
d
el
in
tr
o
d
u
ce
d
b
y
J
o
ac
h
i
m
’
s
a
n
d
s
u
b
s
eq
u
e
n
tl
y
u
s
ed
b
y
o
th
er
s
.
I
t
atte
m
p
ts
to
f
in
d
t
h
e
d
i
m
en
s
io
n
al
s
p
ac
e
a
m
o
n
g
a
ll
t
h
e
|
T
|
-
d
i
m
en
s
io
n
a
l
s
p
ac
es
th
at
s
ep
ar
at
e
th
e
n
e
g
ati
v
e
f
o
r
m
o
f
t
h
e
p
o
s
iti
v
e
tr
ain
i
n
g
e
x
a
m
p
les
[
2
0
,
2
1
]
.
I
t
is
b
ased
o
n
Str
u
ct
u
r
al
R
is
k
Mi
n
i
m
iza
tio
n
[
22
-
24
]
.
Hig
h
d
i
m
en
s
io
n
al
f
ea
tu
r
e
s
p
ac
es,
f
e
w
ir
r
elev
a
n
t
f
ea
t
u
r
es (
d
en
s
e
c
o
n
ce
p
t v
ec
to
r
)
,
an
d
s
p
ar
s
e
in
s
t
an
ce
v
ec
to
r
s
ar
e
t
h
e
p
ar
ticu
lar
p
r
o
p
e
r
ties
o
f
th
e
tex
t
ac
k
n
o
w
led
g
ed
b
y
th
e
SV
M
[
2
4
]
.
SVMs
m
ap
d
ata
to
a
h
ig
h
d
i
m
e
n
s
io
n
al
f
ea
t
u
r
e
s
o
t
h
at
t
h
e
d
ata
p
o
in
t
s
co
u
ld
b
e
ca
te
g
o
r
ized
ev
en
th
e
d
ata
i
s
n
o
t
lin
ea
r
l
y
s
ep
ar
ab
le.
Ma
th
e
m
atica
l
f
u
n
ctio
n
s
u
s
ed
b
y
SVM
s
i
n
I
B
M®
SP
SS
®
f
o
r
tr
an
s
f
o
r
m
a
ti
o
n
ar
e
lin
ea
r
,
p
o
l
y
n
o
m
ial,
R
a
d
ial
B
asis
F
u
n
ctio
n
,
an
d
s
ig
m
o
id
.
SVMs
u
n
d
er
tak
e
t
w
o
i
m
p
o
r
ta
n
t a
d
v
a
n
tag
e
s
f
o
r
T
C
J
o
ac
h
i
m
s
[
2
1
].
a)
T
er
m
s
elec
tio
n
is
o
f
ten
n
o
t
r
eq
u
ir
ed
,
as
SVMs
ten
d
to
b
e
r
e
aso
n
ab
l
y
r
o
b
u
s
t
to
o
v
er
f
itt
in
g
an
d
ca
n
s
ca
le
u
p
to
co
n
s
id
er
ab
le
d
im
e
n
s
io
n
a
liti
es;
b)
No
h
u
m
a
n
a
n
d
m
ac
h
i
n
e
ef
f
o
r
t
in
p
ar
a
m
eter
t
u
n
in
g
o
n
a
v
alid
atio
n
s
et
i
s
n
ee
d
ed
b
ec
au
s
e
t
h
er
e
is
a
th
eo
r
etica
ll
y
m
o
ti
v
ated
,
“
d
e
f
a
u
lt”
ch
o
ice
o
f
p
ar
a
m
eter
s
etti
n
g
s
,
w
h
ich
h
as a
l
s
o
b
ee
n
ar
ch
it
ec
tu
r
e.
T
h
er
e
ar
e
f
o
u
r
t
y
p
es
o
f
m
u
l
ti
-
cla
s
s
ar
ch
itect
u
r
es
u
s
i
n
g
b
in
ar
y
cla
s
s
i
f
ier
s
:
o
n
e
-
ag
ai
n
s
t
-
r
est,
o
n
e
-
ag
ain
s
t
-
o
n
e,
ac
y
clic
d
ir
ec
t
g
r
ap
h
-
A
DG,
a
n
d
u
n
b
ala
n
ce
d
d
ec
is
io
n
tr
ee
-
U
DT
[2
5
]
.
I
n
o
n
e
a
g
ain
s
t
r
e
s
t
ar
ch
itect
u
r
e
f
o
r
d
i
s
ti
n
g
u
i
s
h
i
n
g
,
m
cla
s
s
e
s
m
c
lass
if
ier
s
ar
e
n
ee
d
ed
.
E
v
er
y
cla
s
s
i
f
ier
C
i
i
s
to
b
e
tr
ai
n
ed
f
o
r
r
ec
o
g
n
izi
n
g
c
lass
.
T
h
e
o
n
e
-
a
g
ain
s
t
-
o
n
e
class
i
f
ier
s
ar
e
n
ee
d
ed
f
o
r
ea
ch
d
if
f
er
en
t
clas
s
p
air
an
d
ar
e
ev
alu
ated
i
n
p
ar
allel.
Sa
m
p
les
o
f
o
n
l
y
an
d
ar
e
u
s
ed
to
tr
ain
th
e
clas
s
i
f
ier
.
W
h
en
th
e
c
lass
if
ie
r
r
ec
o
g
n
izes
t
h
e
s
a
m
p
le
b
elo
n
g
in
g
to
class
o
r
a
v
o
te
is
ass
i
g
n
ed
to
o
r
r
esp
ec
tiv
el
y
.
Af
ter
all
th
e
class
i
f
ier
s
h
a
v
e
class
if
ied
th
e
s
a
m
p
le
th
e
clas
s
w
h
ich
r
ec
ei
v
ed
th
e
m
o
r
e
v
o
tes
i
s
co
n
s
id
er
ed
to
b
e
th
e
class
to
w
h
ic
h
s
a
m
p
le
b
elo
n
g
s
.
Fig
u
r
e
2
s
h
o
w
s
all
th
e
f
o
u
r
co
m
b
in
atio
n
s
in
w
h
ich
SV
Ms
ca
n
b
e
ar
ch
itectu
r
e
f
o
r
th
e
cla
s
s
i
f
ier
.
T
h
e
n
u
m
b
er
o
f
class
i
f
ier
s
n
ee
d
ed
f
o
r
th
e
f
o
u
r
d
if
f
er
en
t
t
y
p
e
s
o
f
m
u
lt
i
-
cl
ass
ar
c
h
itect
u
r
es
i
s
lis
ted
in
T
ab
le
1
.
Fig
u
r
e
2
.
(
a)
o
n
e
ag
ain
s
t r
est,
(
b
)
o
n
e
ag
ain
s
t o
n
e,
(
c)
cy
cl
ic
d
ir
ec
t g
r
ap
h
,
(
d
)
u
n
b
alan
ce
d
d
ec
is
io
n
tr
ee
T
ab
le
1
.
B
in
ar
y
clas
s
if
ier
ar
ch
itectu
r
es
A
r
c
h
i
t
e
c
t
u
r
e
N
u
mb
e
r
o
f
C
l
a
ss
i
f
i
e
r
s
C
l
a
ssi
f
i
e
r
s
u
se
d
f
o
r
S
a
m
p
l
e
C
l
a
ssi
f
i
c
a
t
i
o
n
o
n
e
a
g
a
i
n
st
o
n
e
o
n
e
a
g
a
i
n
st
r
e
st
A
D
H
UDH
2
.
3
.
B
a
y
esia
n net
w
o
rk
s
A
B
a
y
esia
n
n
et
w
o
r
k
is
a
d
ir
ec
ted
ac
y
clic
g
r
ap
h
(
DA
G)
w
it
h
a
co
n
d
itio
n
al
p
r
o
b
a
b
ilit
y
d
is
tr
ib
u
tio
n
(
C
.
P
.
tab
le)
f
o
r
ea
ch
n
o
d
e,
c
o
ll
ec
tiv
e
l
y
r
ep
r
esen
ted
b
y
Θ.
E
ac
h
n
o
d
e
r
ep
r
esen
ts
a
d
o
m
ain
v
ar
iab
le,
an
d
ea
ch
ar
c
b
et
w
ee
n
n
o
d
es
r
ep
r
esen
ts
a
p
r
o
b
a
b
ilis
tic
d
ep
en
d
en
c
y
[
2
6
]
.
T
h
is
r
ec
o
g
n
itio
n
m
o
d
el
lear
n
s
t
h
e
co
n
d
itio
n
al
p
r
o
b
ab
ilit
y
o
f
e
ac
h
attr
ib
u
te
A
i
f
r
o
m
t
h
e
cla
s
s
lab
el
C
f
r
o
m
th
e
tr
ain
i
n
g
d
ata.
I
t
s
p
er
f
o
r
m
a
n
ce
h
as
b
ee
n
p
r
o
v
ed
to
b
e
co
m
p
etitiv
e
w
it
h
s
tate
-
of
-
th
e
-
ar
t
cl
ass
i
f
ier
s
[
2
7
]
.
T
h
e
ad
v
an
ta
g
es
o
f
t
h
e
s
e
m
o
d
els
ar
e
th
e
y
ca
n
f
i
t
co
m
p
lex
p
r
o
b
lem
s
i
n
an
y
d
o
m
ai
n
,
w
h
e
th
er
c
o
n
tin
u
o
u
s
,
d
is
cr
ete,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esia
n
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4752
A
n
a
lysi
s
o
f m
a
ch
in
e
lea
r
n
in
g
a
lg
o
r
ith
ms fo
r
ch
a
r
a
cter reco
g
n
itio
n
:
a
ca
s
e
…
(
Ow
a
is
Mu
jt
a
b
a
K
h
a
n
d
a
y
)
577
o
r
m
ix
ed
,
w
i
th
u
n
d
eter
m
i
n
ed
lab
els,
p
ar
t
ial
lab
els,
o
r
m
a
n
y
cla
s
s
v
ar
iab
les
to
b
e
s
i
m
u
lt
an
eo
u
s
l
y
p
r
ed
icted
[2
8
]
.
T
h
e
B
ay
e
s
ian
cla
s
s
i
f
ier
r
ep
r
esen
ted
b
y
a
B
a
y
e
s
ia
n
n
et
w
o
r
k
i
s
d
ef
i
n
ed
as f
o
llo
w
s
:
(
4
)
Ass
u
m
in
g
th
a
t
all
th
e
attr
ib
u
tes
ar
e
in
d
ep
en
d
en
t
g
i
v
en
t
h
e
class
.
T
h
e
co
n
d
itio
n
al
in
d
ep
en
d
en
ce
ass
u
m
p
tio
n
is
[
3
0
]
:
∏
(
5
)
T
h
e
r
esu
ltin
g
n
aiv
e
B
a
y
e
s
ia
n
class
i
f
ier
is
:
∏
(
6
)
T
h
e
m
ai
n
p
r
o
b
lem
in
co
n
f
r
o
n
tin
g
n
ai
v
e
B
ay
e
s
is
its
co
n
d
iti
o
n
al
attr
ib
u
te
ass
u
m
p
t
io
n
;
Na
ïv
e
B
a
y
es
class
i
f
icatio
n
i
m
p
r
o
v
e
s
th
e
a
cc
u
r
ac
y
i
n
th
e
d
o
m
ai
n
s
w
it
h
ir
r
elev
an
t
o
r
r
ed
u
n
d
an
t
attr
i
b
u
tes
b
u
t
n
o
t
i
n
th
e
o
th
er
s
.
T
h
u
s
,
a
v
ar
ie
t
y
o
f
m
eth
o
d
s
w
er
e
d
ev
e
lo
p
ed
f
o
r
im
p
r
o
v
in
g
i
ts
e
f
f
icie
n
c
y
;
f
o
r
ex
a
m
p
le,
th
e
tr
ee
au
g
m
e
n
ted
n
aï
v
e
B
a
y
e
s
[
29
]
lead
s
to
ac
ce
p
tab
le
co
m
p
u
tat
i
o
n
al
co
m
p
lex
i
t
y
.
T
h
e
o
th
er
a
p
p
r
o
ac
h
e
s
u
s
ed
b
y
L
ia
n
g
x
ia
a
n
d
et
al.
f
o
r
i
m
p
r
o
v
in
g
it
s
e
f
f
icien
c
y
ar
e
f
ea
tu
r
e
s
elec
tio
n
,
s
tr
u
ct
u
r
e
ex
te
n
s
io
n
,
lo
ca
l
lear
n
in
g
,
an
d
d
ata
ex
p
an
s
io
n
.
T
h
e
i
m
p
r
o
v
ed
alg
o
r
ith
m
s
u
s
ed
ar
e
E
NB
,
SP
-
T
A
N,
L
W
NB
(
K=
5
0
)
,
an
d
L
NB
.
E
NB
an
d
SP
-
T
A
N
o
u
tp
er
f
o
r
m
ed
N
.
B
.
in
cr
e
asin
g
t
h
e
e
f
f
icie
n
c
y
f
r
o
m
8
2
.
4
1
%
to
8
3
.
2
2
an
d
8
4
.
7
6
%
,
r
esp
ec
tiv
el
y
,
in
th
e
3
6
d
ata
s
ets th
at
w
er
e
te
s
ted
.
[
3
0
].
2
.
4
.
Ra
nd
o
m
f
o
re
s
t
s
R
an
d
o
m
Fo
r
est
is
a
ter
m
f
o
r
class
i
f
ier
co
m
b
in
a
tio
n
s
,
also
k
n
o
w
n
as
C
las
s
i
f
ier
E
n
s
e
m
b
les,
i.e
.
,
a
co
m
b
i
n
atio
n
o
f
M
u
ltip
le
C
las
s
if
ier
S
y
s
te
m
s
(
M
C
S)
to
i
m
p
r
o
v
e
th
e
r
eliab
ili
t
y
in
co
m
p
ar
i
s
o
n
w
it
h
i
n
d
iv
id
u
a
l
class
i
f
ier
s
[
3
1
]
.
T
h
e
class
if
ier
u
s
es
L
tr
ee
-
s
tr
u
ctu
r
ed
class
i
f
i
er
s
,
w
h
er
e
is
th
e
in
p
u
t
,
an
d
Θ_
k
ar
e
i
n
d
ep
en
d
en
tl
y
id
en
tical
ly
d
i
s
tr
ib
u
ted
r
a
n
d
o
m
v
ec
to
r
s
.
T
h
u
s
,
i
t i
s
s
a
id
a
f
a
m
il
y
o
f
m
et
h
o
d
s
h
a
v
i
n
g
v
ar
io
u
s
al
g
o
r
ith
m
s
.
T
h
e
f
o
u
r
ap
p
r
o
ac
h
es
p
r
o
p
o
s
ed
f
o
r
b
u
ild
in
g
t
h
e
MCP
s
ar
e
th
e
d
esig
n
lev
el,
th
e
class
if
ie
r
lev
el,
t
h
e
f
ea
tu
r
e
le
v
el,
a
n
d
th
e
d
ata
le
v
el
[
3
2
]
.
T
h
e
last
t
w
o
ap
p
r
o
ac
h
es
h
av
e
b
ee
n
u
s
ed
an
d
p
r
o
v
e
n
ex
ce
p
tio
n
all
y
s
u
cc
es
s
f
u
l
i
n
v
o
lv
in
g
th
e
B
ag
g
i
n
g
tech
n
iq
u
e,
b
o
o
s
tin
g
tec
h
n
iq
u
e,
an
d
R
an
d
o
m
s
u
b
s
p
ac
e
p
r
in
cip
les [
32
-
34
]
.
I
n
th
e
R
an
d
o
m
S
u
b
s
p
ac
e
p
r
i
n
cip
le,
ea
ch
tr
ee
is
g
r
o
w
n
as b
elo
w
:
a)
I
f
th
er
e
ar
e
tr
ain
in
g
ex
a
m
p
les
in
a
tr
ain
in
g
s
e
t,
th
en
s
a
m
p
le
ca
s
es
f
o
r
th
e
r
esu
lti
n
g
tr
ai
n
in
g
s
et
o
f
th
e
tr
ee
at
r
an
d
o
m
w
i
th
o
u
t r
ep
lace
m
en
t.
b)
A
num
ber
w
her
e
i
s
t
he
n
um
ber
of
f
ea
t
ur
es
i
s
se
t
at
ea
ch
node.
A
su
b
se
t
o
f
t
he
i
s
chos
en
at
r
an
dom
, and a
m
o
ng
t
hem
, t
he be
s
t
sp
l
i
t
i
s s
e
l
ec
t
e
d.
c)
E
ac
h
tr
ee
is
g
r
o
w
n
to
it
s
m
a
x
i
m
u
m
s
ize
an
d
u
n
p
r
u
n
ed
.
T
h
e
alg
o
r
ith
m
w
o
r
k
s
o
n
t
w
o
p
ar
a
m
eter
s
an
d
;
:
n
u
m
b
er
o
f
tr
ee
s
in
t
h
e
f
o
r
est
a
n
d
f
ea
tu
r
es
th
at
ar
e
p
r
eselecte
d
f
o
r
th
e
s
p
litt
i
n
g
p
r
o
ce
s
s
.
Ma
n
y
r
e
s
ea
r
ch
er
s
h
av
e
u
s
ed
R
.
F.
B
er
m
in
[
3
5
]
s
p
lit
at
ea
ch
n
o
d
e
is
d
o
n
e
ac
co
r
d
in
g
to
th
e
lin
ea
r
c
o
m
b
i
n
atio
n
s
o
f
th
e
f
ea
t
u
r
es
i
n
s
tead
o
f
a
s
i
n
g
le
o
n
e.
R
o
b
n
i
k
[
3
6
]
im
p
r
o
v
ed
th
e
co
m
b
i
n
atio
n
p
r
o
ce
s
s
b
y
in
tr
o
d
u
cin
g
th
e
w
ei
g
h
ted
v
o
tin
g
m
et
h
o
d
.
B
o
in
ee
et
al.
in
tr
o
d
u
ce
d
Me
ta
R
an
d
o
m
Fo
r
ests
;
it
co
n
s
is
ted
o
f
R
.
F.
a
s
b
ase
class
i
f
ier
s
f
o
r
co
m
b
i
n
a
tio
n
tech
n
iq
u
e
s
.
I
n
al
m
o
s
t
m
a
n
y
al
g
o
r
ith
m
s
,
t
h
e
n
u
m
b
er
;
n
o
o
f
tr
ee
s
in
t
h
e
f
o
r
est
h
as
b
ee
n
ar
b
itra
r
y
ch
o
s
e
n
eq
u
al
to
1
0
0
.
B
er
m
ain
c
h
o
s
e
w
it
h
o
u
t e
x
p
lai
n
i
n
g
t
h
e
r
ea
s
o
n
.
Si
m
o
n
B
er
n
ar
d
et
al.
[
3
1
]
ex
p
er
i
m
en
ted
o
n
MN
I
ST
d
ataset
an
d
tr
ied
to
ex
p
lain
th
e
p
ar
a
m
etr
izatio
n
in
f
lu
e
n
ce
f
o
r
th
e
R
.
F.
T
h
e
n
u
m
b
er
o
f
f
ea
t
u
r
es
r
an
d
o
m
l
y
s
e
lecte
d
in
th
e
Fo
r
est
-
R
I
p
r
o
ce
s
s
h
as
to
b
e
g
r
ea
ter
th
an
1
,
b
u
t
it
s
v
a
lu
e
ca
n
n
o
t
b
e
to
o
h
ig
h
as
th
e
ac
cu
r
ac
y
ten
d
s
to
co
n
v
er
g
e
w
it
h
i
n
cr
ea
s
in
g
b
u
t
s
to
p
s
a
s
.
A
ll
cu
r
v
e
s
f
o
r
v
alu
e
s
b
eg
in
to
r
is
e
an
d
a
r
e
c
o
n
s
tan
t
till
t
h
e
v
alu
e
o
f
an
d
th
en
s
tar
t
to
d
ec
r
ea
s
e
an
d
r
ea
ch
a
m
in
i
m
u
m
at
,
w
h
ich
co
u
ld
b
e
b
ec
au
s
e
to
o
m
a
n
y
f
ea
t
u
r
es
p
r
eselecte
d
m
a
k
e
s
th
e
d
i
v
er
s
it
y
d
ec
r
ea
s
e
b
et
w
ee
n
tr
ee
s
i
n
f
o
r
ests
.
T
h
e
m
o
r
e
t
h
e
f
ea
t
u
r
es
ar
e
r
a
n
d
o
m
l
y
s
ele
cted
,
th
e
m
o
r
e
t
h
e
tr
e
e
s
ar
e
id
en
tical
to
ea
ch
o
th
e
r
.
R
ec
o
g
n
itio
n
r
ates
w
it
h
r
esp
ec
t
to
an
d
ar
e
s
h
o
w
n
i
n
Fi
g
u
r
e
3
,
an
d
w
e
ca
n
co
n
clu
d
e
t
h
at
i
n
t
h
e
tr
ee
in
d
u
cin
g
p
r
o
ce
s
s
,
r
ath
er
t
h
an
r
a
n
d
o
m
l
y
c
h
o
o
s
i
n
g
th
e
s
p
litt
i
n
g
te
s
ts
,
t
h
en
w
e
s
h
o
u
ld
i
m
p
le
m
en
t
s
o
m
e
s
elec
t
io
n
m
e
asu
r
e.
T
h
e
r
ec
o
g
n
itio
n
r
ate
m
ax
i
m
a
in
t
h
e
ex
p
er
i
m
e
n
t
co
n
d
u
cted
b
y
th
e
Si
m
o
n
[3
1
]
in
ca
s
e
o
f
h
an
d
w
r
itte
n
d
ig
it
r
ec
o
g
n
itio
n
,
t
h
e
m
a
x
i
m
a
r
e
co
g
n
itio
n
r
ates
ar
e
r
ea
ch
ed
in
th
e
ar
ea
d
ef
in
ed
b
y
th
e
in
ter
v
al
s
[
1
0
0
,
3
0
0
]
f
o
r
an
d
[
5
,
2
0
]
f
o
r
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4752
I
n
d
o
n
esia
n
J
E
lec
E
n
g
&
C
o
m
p
Sci,
Vo
l.
21
,
No
.
1
,
J
an
u
ar
y
2
0
2
1
:
5
7
4
-
581
578
(
a)
(
b
)
Fig
u
r
e
3
.
R
ec
o
g
n
i
tio
n
r
ates
w
i
th
r
esp
ec
t to
an
d
[3
1
]
a)
R
ec
o
g
n
itio
n
R
ate
s
w
r
t
,
b)
R
ec
o
g
n
itio
n
R
ate
s
w
r
t
3.
RE
SU
L
T
S
A
ND
D
IS
CU
SS
I
O
N
ML
P
s
ar
e
ef
f
icien
t
cla
s
s
i
f
ier
s
f
o
r
th
e
m
u
lti
-
cla
s
s
p
r
o
b
lem
,
b
u
t
u
s
in
g
a
b
ac
k
-
p
r
o
p
ag
atio
n
alg
o
r
ith
m
as
t
h
e
lear
n
i
n
g
alg
o
r
it
h
m
i
s
a
d
i
s
ad
v
an
tag
e
b
ec
au
s
e
t
h
e
al
g
o
r
ith
m
co
u
ld
s
to
p
at
a
lo
ca
l
m
i
n
i
m
u
m
.
Ho
w
ev
er
,
a
m
o
m
e
n
t
u
m
s
tr
ateg
y
co
u
ld
b
e
u
s
ed
f
o
r
escap
i
n
g
t
h
e
lo
ca
l
m
i
n
i
m
u
m
,
b
u
t
t
h
e
n
it
o
v
er
f
i
t
s
t
h
e
w
ei
g
h
ts
,
t
h
u
s
d
ec
r
ea
s
in
g
t
h
e
g
e
n
er
alizin
g
c
ap
ab
ilit
y
[
3
7
]
.
Fig
u
r
e
4
s
h
o
ws
th
e
v
ar
io
u
s
t
y
p
es
o
f
M
L
P
s
u
s
ed
b
y
th
e
v
ar
io
u
s
r
esear
ch
er
s
[
3
]
f
o
r
d
ig
it
r
ec
o
g
n
itio
n
.
T
h
e
y
-
a
x
i
s
is
th
e
er
r
o
r
p
er
ce
n
tag
e,
an
d
t
h
e
x
-
a
x
i
s
r
ep
r
esen
ts
th
e
d
i
f
f
er
en
t
t
y
p
es o
f
cla
s
s
i
f
ier
s
u
s
ed
.
C
o
m
p
ar
is
o
n
o
f
d
if
f
er
en
t c
la
s
s
i
f
ier
s
,
as sh
o
w
n
in
T
ab
le
2
.
Fig
u
r
e
4
.
C
lass
if
ier
s
w
it
h
th
eir
tr
ain
in
g
ti
m
e,
er
r
o
r
,
an
d
r
ec
o
g
n
itio
n
r
ate
s
I
t
is
ev
id
en
t
t
h
at
to
ac
ce
s
s
a
m
ea
g
er
er
r
o
r
r
ate,
an
d
th
er
e
is
a
n
ee
d
f
o
r
lo
ts
a
m
o
u
n
t
o
f
ti
m
e
t
o
b
e
s
p
en
t
b
y
th
e
m
o
d
el
f
o
r
tr
ain
i
n
g
,
i.e
.
,
lo
ts
o
f
d
ata
w
a
s
p
r
o
v
id
ed
f
o
r
th
e
tr
ai
n
in
g
p
u
r
p
o
s
e.
I
t
is
s
ee
n
w
it
h
i
n
cr
ea
s
ed
ac
cu
r
ac
y
,
a
n
d
th
er
e
is
a
p
r
o
p
o
r
tio
n
al
in
cr
ea
s
e
i
n
tr
ain
in
g
ti
m
e.
I
t
f
o
llo
w
s
t
h
e
s
i
m
p
le
r
u
le
w
it
h
t
h
e
m
o
r
e
ex
p
er
ien
ce
t
h
at
y
o
u
lear
n
m
o
r
e
[
3
8
].
T
ab
le
2
.
C
o
m
p
ar
is
o
n
o
f
d
if
f
er
en
t c
las
s
i
f
ier
s
M
e
t
h
o
d
Er
r
o
r
R
a
t
e
(
%
)
T
r
a
i
n
i
n
g
(
W
e
e
k
s)
R
e
c
o
g
n
i
t
i
o
n
T
i
me
(
S
)
B
a
se
l
i
n
e
l
i
n
e
a
r
C
l
a
ssi
f
i
e
r
8
.
4
0
.
0
7
0
.
0
1
B
a
se
l
i
n
e
n
e
a
r
e
st
n
e
i
g
h
b
o
r
(
k
=
3
)
2
.
4
0
1
L
e
t
N
e
t
1
1
.
7
0
.
5
0
.
0
1
5
L
e
t
N
e
t
4
1
.
1
2
0
.
0
3
F
u
l
l
y
c
o
n
n
e
c
t
e
d
mu
l
t
i
-
l
a
y
e
r
n
e
u
r
a
l
n
e
t
w
o
r
k
s
1
.
6
1
0
.
1
B
o
o
st
e
d
L
e
t
N
e
t
4
0
.
7
5
0
.
0
5
B
est
m
ar
g
i
n
s
o
f
s
ep
ar
atio
n
b
et
w
ee
n
an
y
t
w
o
cla
s
s
es
ar
e
f
o
u
n
d
b
y
t
h
e
SVM
th
u
s
,
co
n
s
id
er
ed
as
th
e
b
est
b
in
ar
y
class
i
f
ier
.
I
t
f
i
n
d
s
th
e
b
est
o
r
o
p
tim
al
li
n
er
f
u
n
ctio
n
th
at
s
ep
ar
ates
t
h
e
class
e
s
.
R
en
ata
F.P
et
al.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esia
n
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4752
A
n
a
lysi
s
o
f m
a
ch
in
e
lea
r
n
in
g
a
lg
o
r
ith
ms fo
r
ch
a
r
a
cter reco
g
n
itio
n
:
a
ca
s
e
…
(
Ow
a
is
Mu
jt
a
b
a
K
h
a
n
d
a
y
)
579
u
s
ed
a
n
SVM
b
ased
r
ec
o
g
n
iz
er
th
at
u
s
es
4
5
SVM
s
f
o
r
clas
s
if
ica
tio
n
b
ec
au
s
e
t
h
e
SVM
i
s
a
b
in
ar
y
c
lass
if
ier
,
an
d
th
er
e
ar
e
te
n
o
u
tp
u
t
cla
s
s
es
f
o
r
th
i
s
p
r
o
b
le
m
,
s
o
t
h
e
ar
ch
itect
u
r
e
m
u
s
t
b
e
co
m
p
o
s
ed
o
f
4
5
SVMs
to
co
m
p
ar
e
f
o
r
all
th
e
co
m
b
i
n
at
io
n
s
.
T
h
e
h
y
p
er
b
o
lic
ta
n
g
e
n
t
f
u
n
c
tio
n
is
u
s
ed
as
a
n
ac
tiv
atio
n
f
u
n
ctio
n
a
n
d
g
r
ad
ien
t
d
esce
n
t
al
g
o
r
ith
m
f
o
r
lear
n
in
g
.
T
h
e
m
et
h
o
d
s
h
o
w
e
d
b
etter
p
er
f
o
r
m
an
ce
o
n
t
h
e
NSI
D
SD1
9
d
ataset
th
an
M
L
P
,
1
0
SVMs,
an
d
B
ellilli
-
M
L
P
+1
5
/4
5
SVM.
T
h
e
er
r
o
r
r
ate
is
r
e
d
u
ce
d
to
2
.
0
6
%.
C
o
m
p
ar
is
o
n
o
f
er
r
o
r
r
ates o
f
d
if
f
er
e
n
t c
las
s
i
f
i
er
s
,
as sh
o
w
n
in
Fig
u
r
e
5
.
Fig
u
r
e
5
.
C
o
m
p
ar
is
o
n
o
f
er
r
o
r
r
ates o
f
d
if
f
er
en
t c
la
s
s
i
f
ier
s
Kan
d
m
i
n
g
L
i
u
p
r
o
p
o
s
ed
a
m
o
d
el
o
n
a
B
a
y
e
s
ian
n
et
w
o
r
k
w
it
h
t
h
e
co
n
ca
v
it
y
f
ea
tu
r
e
a
n
d
class
i
f
icatio
n
an
d
a
s
s
o
ciatio
n
r
u
les
f
o
r
d
ig
it
s
to
s
elec
t
m
o
r
e
f
ea
t
u
r
es
t
h
at
e
f
f
ec
t
iv
el
y
ac
h
ie
ve
a
m
o
r
e
ac
c
u
r
ate
r
esu
lt.
T
h
e
tr
ai
n
i
n
g
s
a
m
p
le
co
n
s
i
s
ted
o
f
6
0
0
s
in
g
le
d
i
g
it
s
a
m
p
les
s
elec
ti
n
g
2
0
n
o
n
-
o
v
er
la
p
p
in
g
d
i
g
it
s
a
m
p
le
s
f
o
r
test
an
d
atta
in
ed
t
h
e
ac
cu
r
ac
y
o
f
6
5
.
0
7
to
tal
p
er
ce
n
t
an
d
3
4
.
9
3
%
o
f
f
alse
r
ate
o
n
2
0
0
s
a
m
p
les.
B
r
ei
m
a
n
p
r
o
p
o
s
ed
a
m
o
d
el
in
[
3
5
]
ca
lled
Fo
r
est
-
R
C
,
w
h
er
e
a
t
e
ac
h
n
o
d
e,
a
s
p
lit
i
s
d
o
n
e
b
ased
o
n
t
h
e
li
n
ea
r
co
m
b
i
n
atio
n
o
f
f
ea
t
u
r
es,
w
h
i
ch
allo
w
s
th
e
d
ea
lin
g
w
it
h
t
h
e
ca
s
es
h
a
v
i
n
g
o
n
l
y
a
f
e
w
i
n
p
u
t
s
s
u
p
p
lied
.
Fo
r
s
i
m
ilar
it
y
ev
al
u
atio
n
,
t
h
e
ac
cu
r
ac
ies
o
n
s
i
m
ilar
i
n
p
u
ts
w
er
e
ex
a
m
in
ed
,
s
o
to
r
e
m
o
v
e
f
r
o
m
th
e
v
o
te
t
h
at
s
h
o
w
s
th
e
lo
w
est
v
al
u
es.
R
o
b
er
t
in
[
3
6
]
i
m
p
r
o
v
ed
t
h
e
co
m
b
in
at
io
n
p
r
o
ce
s
s
d
u
e
to
o
r
ig
in
al
Fo
r
est
-
R
I
b
y
i
n
tr
o
d
u
cin
g
th
e
w
ei
g
h
ted
v
o
ti
n
g
m
et
h
o
d
to
tak
e
in
to
ac
co
u
n
t
a
r
estricte
d
s
u
b
s
et
o
f
t
h
e
clas
s
i
f
ier
o
u
t
p
u
ts
b
ased
o
n
th
e
in
d
iv
id
u
al
ac
cu
r
ac
ie
s
o
f
th
e
s
i
m
ilar
in
s
ta
n
ce
s
.
A
n
o
t
h
er
m
et
h
o
d
f
o
r
R
.
F.
is
ca
lled
Me
ta
R
a
n
d
o
m
.
B
o
n
iee
et
al.
p
r
o
p
o
s
ed
Fo
r
ests
[
39
]
in
w
h
i
ch
R
.
F.
as
a
b
ase
clas
s
i
f
ier
o
f
co
m
b
in
atio
n
tec
h
n
iq
u
es
w
er
e
u
s
ed
.
No
n
e
o
f
th
e
m
et
h
o
d
s
ca
n
b
e
p
r
o
v
en
to
b
e
s
u
p
er
io
r
to
o
th
er
s
in
ter
m
s
o
f
ac
cu
r
ac
y
,
p
lu
s
th
e
y
h
av
e
n
o
t
g
iv
e
n
e
x
p
er
i
m
e
n
tal
w
o
r
k
f
r
o
m
a
p
r
ac
tical
p
o
in
t
o
f
v
ie
w
.
I
t
co
u
ld
b
e
s
ee
n
in
t
h
ese
m
et
h
o
d
s
s
o
m
e
o
f
t
h
e
p
ar
am
eter
v
al
u
es
w
er
e
co
m
m
o
n
l
y
u
s
ed
lik
e
L
.
T
h
e
n
u
m
b
er
o
f
tr
ee
s
in
f
o
r
est
s
w
a
s
f
ix
ed
to
1
0
0
.
B
er
n
ar
d
e
t
al.
[
4
0
]
tr
ied
to
u
n
d
er
s
ta
n
d
th
e
r
o
les
o
f
th
e
p
a
r
a
m
eter
v
al
u
es
o
n
th
e
b
e
h
av
io
r
o
f
R
.
F.
C
o
n
ce
r
n
i
n
g
t
h
e
i
n
cr
ea
s
e
in
t
h
e
n
u
m
b
er
o
f
tr
ee
s
,
th
e
ac
c
u
r
ac
y
o
f
R
.
F.
co
n
v
er
g
e
s
,
an
d
t
h
e
in
cr
ea
s
e
i
n
th
e
r
ec
o
g
n
i
tio
n
r
ate
s
lo
w
s
d
o
w
n
af
ter
1
0
0
tr
ee
s
,
an
d
th
e
in
cr
e
m
e
n
t
w
a
s
lo
g
ar
i
th
m
ic.
T
h
e
y
co
n
f
ir
m
ed
t
h
at
th
e
tr
ee
in
d
u
cin
g
p
r
o
ce
s
s
s
h
o
u
ld
i
m
p
le
m
en
t
a
s
p
litt
i
n
g
s
elec
t
io
n
m
ea
s
u
r
e
r
ath
er
th
a
n
r
an
d
o
m
l
y
c
h
o
o
s
in
g
s
p
litt
i
n
g
test
s
.
Ho
w
e
v
er
,
it
ca
n
b
e
s
ee
n
th
at
th
e
r
ec
o
g
n
itio
n
r
ate
d
o
es n
o
t in
cr
e
ase
b
e
y
o
n
d
9
3
.
5
1
%.
4.
CO
NCLU
SI
O
N
Han
d
w
r
itte
n
d
i
g
it
r
ec
o
g
n
it
io
n
i
s
a
s
i
g
n
i
f
ican
t
f
ield
,
m
u
c
h
p
r
o
g
r
es
s
h
as
b
ee
n
m
ad
e,
an
d
m
an
y
tech
n
iq
u
es
h
a
v
e
b
ee
n
d
ev
elo
p
ed
.
T
h
e
b
o
o
s
ted
L
eNe
t
4
m
et
h
o
d
p
er
f
o
r
m
s
th
e
b
est
w
i
th
th
e
ac
cu
r
ac
y
o
f
9
9
.
3
%
an
d
is
t
h
e
b
est
a
m
o
n
g
t
h
e
m
et
h
o
d
s
th
at
h
a
v
e
b
ee
n
s
tu
d
ied
i
n
th
is
p
ap
er
.
T
h
e
o
n
l
y
tr
ad
eo
f
f
i
s
th
e
tr
ai
n
i
n
g
ti
m
e,
w
h
ic
h
is
v
er
y
lar
g
e
a
n
d
is
ab
o
u
t f
iv
e
w
ee
k
s
.
T
h
e
o
p
er
atio
n
al/ac
tu
al
r
ec
o
g
n
itio
n
t
i
m
e
i
s
0
.
0
5
m
s
.
ACK
NO
WL
E
D
G
E
M
E
NT
S
T
h
e
task
w
a
s
p
er
f
o
r
m
ed
u
n
d
er
th
e
f
r
a
m
e
w
o
r
k
o
f
FIK
P
h
ig
h
er
ed
u
ca
tio
n
in
s
tit
u
tio
n
a
l
ex
ce
lle
n
ce
p
r
o
g
r
am
o
n
th
e
o
p
ti
m
izatio
n
o
f
n
at
u
r
al
r
eso
u
r
ce
s
b
ased
o
n
m
o
d
er
n
tech
n
o
lo
g
ie
s
w
it
h
in
t
h
e
th
e
m
e
“
ap
p
l
y
i
n
g
in
telli
g
e
n
t i
n
f
o
r
m
atio
n
to
o
ls
to
in
v
o
l
v
e,
r
ed
u
ce
an
d
r
ed
u
ce
p
r
o
f
ess
io
n
al
lo
s
s
e
s
”
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4752
I
n
d
o
n
esia
n
J
E
lec
E
n
g
&
C
o
m
p
Sci,
Vo
l.
21
,
No
.
1
,
J
an
u
ar
y
2
0
2
1
:
5
7
4
-
581
580
RE
F
E
R
E
NC
E
S
[1
]
P
e
d
ry
c
z
,
W
it
o
ld
,
A
n
d
rz
e
j
S
k
o
w
r
o
n
,
a
n
d
V
lad
ik
Kre
in
o
v
ich
,
“
A
F
u
z
z
y
R
e
g
re
ss
io
n
A
p
p
ro
a
c
h
to
A
c
q
u
isit
io
n
o
f
L
in
g
u
isti
c
Ru
les
,
”
Ha
n
d
b
o
o
k
o
f
G
ra
n
u
lar Co
m
p
u
ti
n
g
.
Jo
h
n
W
il
e
y
&
S
o
n
s,
p
p
.
7
1
9
-
7
3
2
,
2
0
0
8
.
[2
]
S
h
u
y
in
g
,
Ya
n
g
.
“
I
m
a
g
e
Re
c
o
g
n
i
ti
o
n
a
n
d
P
ro
jec
t
P
ra
c
ti
c
e
,
”
Be
ij
in
g
:
P
u
b
li
s
h
in
g
Ho
u
se
o
f
El
e
c
tro
n
ics
In
d
u
stry
.
2
0
1
4
.
[3
]
Bo
tt
o
u
,
L
é
o
n
,
Co
r
in
n
a
Co
rtes
,
J
o
h
n
S
De
n
k
e
r,
Ha
rris
Dru
c
k
e
r,
Isa
b
e
ll
e
G
u
y
o
n
,
L
a
rry
D
Ja
c
k
e
l,
Ya
n
n
L
e
Cu
n
,
e
t
a
l.
,
“
Co
m
p
a
riso
n
o
f
Clas
si
f
ier
M
e
th
o
d
s:
A
Ca
se
S
tu
d
y
in
Ha
n
d
w
rit
t
e
n
Dig
it
Re
c
o
g
n
it
io
n
,
”
P
a
p
e
r
p
r
e
se
n
ted
a
t
th
e
In
ter
n
a
t
io
n
a
l
c
o
n
fer
e
n
c
e
o
n
p
a
tt
e
rn
re
c
o
g
n
i
ti
o
n
,
1
9
9
4
.
[4
]
S
h
a
m
i
m
,
S
M
,
M
o
h
a
m
m
a
d
B
a
d
ru
l
A
la
m
M
iah
,
M
a
su
d
Ra
n
a
A
n
g
o
n
a
S
a
rk
e
r,
A
b
d
u
ll
a
h
A
l
Jo
b
a
ir
,
“
Ha
n
d
w
rit
ten
Dig
it
Re
c
o
g
n
it
io
n
Us
in
g
M
a
c
h
i
n
e
Lea
rn
in
g
A
l
g
o
rit
h
m
s,
”
Glo
b
a
l
J
o
u
rn
a
l
o
f
C
o
mp
u
ter
S
c
ien
c
e
a
n
d
T
e
c
h
n
o
lo
g
y
,
v
o
l.
1
9
,
n
o
.
1
,
p
p
.
1
6
-
2
3
,
2
0
1
8.
[5
]
Ba
su
,
S
.
,
Da
s,
N.,
S
a
rk
a
r,
R.
,
Ku
n
d
u
,
M
.
,
Na
sip
u
ri,
M
.
,
&
Ba
su
,
D
.
K.,
“
Re
c
o
g
n
it
i
o
n
o
f
n
u
m
e
ric
p
o
sta
l
c
o
d
e
s
f
ro
m
m
u
lt
i
-
sc
rip
t
p
o
sta
l
a
d
d
re
ss
b
lo
c
k
s,
”
In
I
n
ter
n
a
ti
o
n
a
l
C
o
n
fer
e
n
c
e
o
n
Pa
tt
e
rn
Rec
o
g
n
it
io
n
a
n
d
M
a
c
h
i
n
e
In
telli
g
e
n
c
e
,
S
p
rin
g
e
r,
p
p
.
3
8
1
-
3
8
6
,
2
0
0
9
.
[6
]
Ku
m
a
r,
V
ik
a
s,
“
On
li
n
e
Ha
n
d
w
r
it
in
g
Re
c
o
g
n
it
io
n
P
ro
b
lem
:
Iss
u
e
s
a
n
d
T
e
c
h
n
iq
u
e
s,
”
M
IT
In
t.
J
.
Co
mp
u
t.
S
c
i.
In
fo
rm
.
T
e
c
h
n
o
l
,
v
o
l
.
4
,
n
o
.
1
,
p
p
.
16
-
2
4
,
2
0
1
4
.
[7
]
Da
s,
Nib
a
ra
n
,
Ra
m
S
a
rk
a
r,
S
u
b
h
a
d
ip
Ba
su
,
M
a
h
a
n
tap
a
s
Ku
n
d
u
,
M
it
a
Na
sip
u
ri,
a
n
d
Dip
a
k
Ku
m
a
r
Ba
su
,
“
A
G
e
n
e
ti
c
A
lg
o
rit
h
m
B
a
se
d
Re
g
io
n
S
a
m
p
li
n
g
f
o
r
S
e
lec
ti
o
n
o
f
L
o
c
a
l
F
e
a
tu
re
s
in
Ha
n
d
w
rit
ten
Di
g
it
Re
c
o
g
n
it
io
n
A
p
p
li
c
a
ti
o
n
,
”
A
p
p
l
ied
S
o
ft
Co
mp
u
ti
n
g
,
v
o
l
.
1
2
,
n
o
.
5
,
p
p
.
1
5
9
2
-
6
0
6
,
2
0
1
2
.
[8
]
P
lam
o
n
d
o
n
,
Ré
jea
n
,
a
n
d
S
a
rg
u
r
N
.
S
rih
a
ri
,
“
On
li
n
e
a
n
d
Of
f
-
L
in
e
Ha
n
d
w
rit
in
g
Re
c
o
g
n
it
io
n
:
A
Co
m
p
re
h
e
n
siv
e
S
u
rv
e
y
,
”
IEE
E
T
ra
n
sa
c
ti
o
n
s
o
n
p
a
tt
e
rn
a
n
a
lys
is
a
n
d
m
a
c
h
in
e
in
tell
ig
e
n
c
e
,
v
o
l.
2
2
,
n
o
.
1
,
p
p
.
6
3
-
8
4
,
2
0
0
0
.
[9
]
Co
n
n
e
ll
,
S
c
o
tt
D,
“
On
li
n
e
Ha
n
d
w
rit
in
g
Re
c
o
g
n
it
i
o
n
Us
i
n
g
M
u
l
ti
p
le
P
a
tt
e
rn
Clas
s M
o
d
e
ls,
”
c
it
e
se
e
r
,
200
0
.
[1
0
]
Bish
o
p
,
Ch
r
isto
p
h
e
r
M
.
,
“
Ne
u
ra
l
Ne
t
w
o
rk
s f
o
r
P
a
tt
e
rn
Re
c
o
g
n
it
io
n
,
”
Ox
f
o
rd
u
n
iv
e
rsity
p
re
ss
,
1
9
9
5
.
[1
1
]
Ha
y
k
in
,
S
im
o
n
,
“
S
e
lf
-
Or
g
a
n
izin
g
M
a
p
s,
”
Ne
u
ra
l
n
e
tw
o
rk
s
-
A
c
o
m
p
re
h
e
n
siv
e
f
o
u
n
d
a
ti
o
n
,
2
n
d
e
d
it
io
n
,
P
re
n
ti
c
e
-
Ha
ll
,
1
9
9
9
.
[1
2
]
Ne
v
e
s,
R
e
n
a
ta
F
.
P
.
,
A
lb
e
rto
N
.
G
.
L
o
p
e
s
F
il
h
o
,
Ca
rl
o
s
A
.
B
.
M
e
l
lo
,
a
n
d
Cle
b
e
r
Zan
c
h
e
tt
i
n
,
“
A
n
S
V
M
Ba
se
d
Off
-
L
in
e
Ha
n
d
w
rit
ten
Dig
it
Re
c
o
g
n
ize
r,
”
2
0
1
1
IE
EE
I
n
ter
n
a
t
io
n
a
l
C
o
n
fer
e
n
c
e
o
n
S
y
ste
ms
,
M
a
n
,
a
n
d
Cy
b
e
rn
e
ti
c
s
,
p
p
.
5
1
0
-
5
1
5
,
2
0
1
1
.
[1
3
]
L
iu
,
Ch
e
n
g
-
L
in
,
Ka
z
u
k
i
Na
k
a
sh
im
a
,
Hiro
sh
i
S
a
k
o
,
a
n
d
Hiro
m
ich
i
F
u
ji
sa
w
a
,
“
H
a
n
d
w
rit
ten
Dig
it
Re
c
o
g
n
it
io
n
:
Be
n
c
h
m
a
r
k
in
g
o
f
S
tate
-
of
-
th
e
-
A
rt
T
e
c
h
n
iq
u
e
s,
”
Pa
tt
e
rn
Rec
o
g
n
it
io
n
,
v
o
l.
3
6
,
n
o
.
1
0
,
p
p
.
2
2
7
1
-
22
85
,
2
0
0
3
.
[1
4
]
S
o
o
ra
,
Na
ra
sim
h
a
Re
d
d
y
,
a
n
d
P
a
ra
g
S
.
De
sh
p
a
n
d
e
,
“
Re
v
iew
o
f
F
e
a
tu
re
Ex
trac
ti
o
n
T
e
c
h
n
i
q
u
e
s
f
o
r
Ch
a
ra
c
ter
Re
c
o
g
n
it
io
n
,
”
IET
E
J
o
u
rn
a
l
o
f
Re
se
a
rc
h
, v
o
l.
6
4
,
n
o
.
2
,
p
p
.
2
8
0
-
2
9
5
,
2
0
1
8
.
[1
5
]
S
o
o
ra
,
Na
ra
si
m
h
a
Re
d
d
y
,
a
n
d
P
a
ra
g
S
.
De
sh
p
a
n
d
e
,
“
No
v
e
l
G
e
o
m
e
tri
c
a
l
S
h
a
p
e
F
e
a
tu
re
Ex
trac
ti
o
n
T
e
c
h
n
iq
u
e
s
f
o
r
M
u
lt
il
in
g
u
a
l
C
h
a
ra
c
ter Rec
o
g
n
it
io
n
,
”
IE
T
E
T
e
c
h
n
ica
l
Rev
iew
,
v
o
l.
3
4
,
n
o
.
6
,
p
p
.
6
1
2
-
6
2
1
,
2
0
1
7
.
[1
6
]
S
o
o
ra
,
Na
ra
sim
h
a
Re
d
d
y
,
a
n
d
P
a
ra
g
S
De
sh
p
a
n
d
e
,
“
Ro
b
u
st
F
e
a
tu
re
-
Ex
trac
ti
o
n
T
e
c
h
n
iq
u
e
f
o
r
L
ic
e
n
se
P
late
Ch
a
ra
c
ters
Re
c
o
g
n
it
io
n
,
”
IE
T
E
J
o
u
rn
a
l
o
f
Res
e
a
rc
h
,
v
o
l.
6
1
,
n
o
.
1
,
p
p
.
7
2
-
7
9
,
2
0
1
5
.
[1
7
]
Da
s,
Nib
a
ra
n
,
Ay
a
tu
ll
a
h
F
a
ru
k
M
o
ll
a
h
,
S
u
d
ip
S
a
h
a
,
a
n
d
S
y
e
d
S
a
h
id
u
l
Ha
q
u
e
,
“
Ha
n
d
w
rit
ten
A
ra
b
ic
Nu
m
e
r
a
l
Re
c
o
g
n
it
io
n
Us
in
g
a
M
u
lt
i
L
a
y
e
r
P
e
rc
e
p
tro
n
,
”
Pro
c
.
Na
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
Rec
e
n
t
T
re
n
d
s
in
In
fo
rm
a
ti
o
n
S
y
ste
ms
,
p
p
.
2
0
0
-
2
0
3
,
2
0
0
6
.
[1
8
]
Kru
se
,
Ru
d
o
lf
,
Ch
risti
a
n
Bo
rg
e
lt
,
Ch
risti
a
n
Bra
u
n
e
,
S
a
n
a
z
M
o
sta
g
h
im
,
a
n
d
M
a
tt
h
ias
S
tein
b
re
c
h
e
r,
“
Co
m
p
u
tatio
n
a
l
In
t
e
ll
ig
e
n
c
e
:
A
M
e
th
o
d
o
lo
g
ica
l
In
tro
d
u
c
ti
o
n
,
”
S
p
rin
g
e
r
,
p
p
.
4
7
-
8
1
,
2
0
1
6
.
[1
9
]
D.
E.
Ru
m
e
lh
a
rt,
G
.
E.
Hin
to
n
,
R
.
J.
W
il
li
a
m
s,
“
Lea
rn
in
g
re
p
re
se
n
tatio
n
s
b
y
b
a
c
k
-
p
ro
p
a
g
a
ti
o
n
e
rro
rs,
”
Na
tu
re
,
v
o
l.
3
2
3
,
n
o
.
9
,
p
p
.
5
3
3
–
5
3
6
,
1
9
8
6
.
[2
0
]
Ro
b
b
in
s,
He
rb
e
rt,
a
n
d
S
u
tt
o
n
M
o
n
r
o
,
“
A
S
to
c
h
a
stic
A
p
p
r
o
x
ima
ti
o
n
M
e
t
h
o
d
,
”
T
h
e
a
n
n
a
ls
o
f
ma
th
e
m
a
ti
c
a
l
sta
ti
stics
,
v
o
l.
2
2
,
n
o
.
3
,
p
p
.
4
0
0
-
4
0
7
,
1
9
5
1
.
[2
1
]
S
e
b
a
stian
i,
F
a
b
rizi
o
.
Co
n
sig
li
o
Na
z
io
n
a
le
De
ll
e
Ric
e
rc
h
e
,
“
M
a
c
h
in
e
lea
rn
in
g
in
a
u
t
o
m
a
ted
te
x
t
c
a
teg
o
riza
ti
o
n
,
”
ACM
Co
mp
u
t
in
g
S
u
rv
e
y
s
,
v
o
l.
3
4
,
n
o
.
1
,
p
p
.
1
-
4
7
,
2
0
0
2
.
[2
2
]
V
a
p
n
ik
,
V
lad
im
ir
N
.
,
“
Co
n
str
u
c
ti
n
g
L
e
a
rn
in
g
A
l
g
o
rit
h
m
s,
”
in
th
e
Na
tu
re
o
f
S
t
a
ti
stic
a
l
L
e
a
r
n
in
g
T
h
e
o
ry
,
S
p
ri
n
g
e
r
,
p
p
.
1
1
9
-
6
6
,
1
9
9
5
.
[
2
3
]
C
o
r
t
e
s
,
C
o
r
i
n
n
a
,
V
l
a
d
i
m
i
r
V
a
p
n
i
k
,
“
S
u
p
p
o
r
t
V
e
c
t
o
r
N
e
t
w
o
r
k
s
“
M
a
c
h
i
n
e
l
e
a
r
n
i
n
g
,
”
v
o
l
.
2
0
,
n
o
.
3
,
p
p
.
2
7
3
-
97
,
1
9
9
5
.
[2
4
]
Jo
a
c
h
im
s,
T
h
o
rste
n
,
“
T
e
x
t
Ca
te
g
o
riza
ti
o
n
w
it
h
S
u
p
p
o
r
t
V
e
c
t
o
r
M
a
c
h
in
e
s:
L
e
a
rn
in
g
w
it
h
M
a
n
y
R
e
le
v
a
n
t
F
e
a
tu
re
s,
”
P
a
p
e
r
p
re
se
n
ted
a
t
th
e
E
u
ro
p
e
a
n
c
o
n
fer
e
n
c
e
o
n
m
a
c
h
i
n
e
lea
rn
i
n
g
,
1
9
9
8
.
[2
5
]
Ha
ss
a
n
,
A
li
,
a
n
d
Ro
b
e
rt
I
.
Da
m
p
e
r,
“
Clas
sif
ica
ti
o
n
o
f
Em
o
ti
o
n
a
l
S
p
e
e
c
h
Us
in
g
3
d
e
c
Hie
ra
rc
h
ica
l
Clas
sif
ier,
”
S
p
e
e
c
h
Co
mm
u
n
ica
ti
o
n
,
v
o
l
.
5
4
,
n
o
.
7
,
p
p
.
9
0
3
-
1
6
,
2
0
1
2
.
[2
6
]
Ch
e
n
g
,
Jie
,
a
n
d
Ru
ss
e
ll
G
re
in
e
r,
“
Co
m
p
a
rin
g
Ba
y
e
sia
n
Ne
t
w
o
rk
Cl
a
ss
if
i
e
rs,
”
P
a
p
e
r
p
re
se
n
ted
a
t
th
e
Pro
c
e
e
d
in
g
s
o
f
th
e
Fi
ft
e
e
n
t
h
Co
n
fer
e
n
c
e
o
n
U
n
c
e
rta
in
ty
i
n
a
rt
i
fi
c
ia
l
in
telli
g
e
n
c
e
,
p
p
.
1
0
1
-
1
0
8
,
1
9
9
9
,
i
n
h
tt
p
s:/
/arx
iv
.
o
rg
/f
tp
/arx
iv
/p
a
p
e
rs/
1
3
0
1
/
1
3
0
1
.
6
6
8
4
.
p
d
f
.
[2
7
]
S
e
e
wa
ld
,
A
lex
a
n
d
e
r
K.,
“
On
th
e
Brit
tl
e
n
e
ss
o
f
Ha
n
d
w
rit
ten
Dig
i
t
Re
c
o
g
n
it
io
n
M
o
d
e
ls,
”
I
S
RN
M
a
c
h
in
e
Visio
n
,
v
o
l.
2
0
1
2
,
2
0
1
1
.
[2
8
]
Bielz
a
,
Co
n
c
h
a
,
a
n
d
P
e
d
ro
L
a
rra
ñ
a
g
a
,
“
Disc
re
t
e
Ba
y
e
sia
n
Ne
tw
o
rk
Clas
sif
ier
s:
A
S
u
rv
e
y
,
”
ACM
Co
mp
u
ti
n
g
S
u
rv
e
y
s (
CS
UR)
,
v
o
l.
4
7
,
n
o
.
1
.
2
0
1
4
.
[2
9
]
F
ried
m
a
n
,
Nir,
Da
n
G
e
i
g
e
r,
a
n
d
M
o
ise
s
G
o
ld
sz
m
id
t,
“
Ba
y
e
sia
n
N
e
tw
o
rk
Cla
ss
i
f
iers
,
”
M
a
c
h
in
e
lea
rn
in
g
,
v
o
l.
2
9
,
n
o
.
2
-
3
,
p
p
.
1
3
1
-
1
6
3
,
1
9
9
7
.
[3
0
]
Jia
n
g
,
L
ian
g
x
i
a
o
,
Dia
n
h
o
n
g
W
a
n
g
,
Zh
ih
u
a
Ca
i,
a
n
d
X
u
e
so
n
g
Ya
n
,
“
S
u
rv
e
y
o
f
I
m
p
ro
v
in
g
Na
iv
e
Ba
y
e
s
f
o
r
Clas
sif
ic
a
ti
o
n
,
”
P
a
p
e
r
p
re
se
n
ted
a
t
th
e
I
n
ter
n
a
t
io
n
a
l
C
o
n
fer
e
n
c
e
o
n
Ad
v
a
n
c
e
d
Da
t
a
M
in
i
n
g
a
n
d
A
p
p
l
i
c
a
ti
o
n
s
,
2
0
0
7
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esia
n
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4752
A
n
a
lysi
s
o
f m
a
ch
in
e
lea
r
n
in
g
a
lg
o
r
ith
ms fo
r
ch
a
r
a
cter reco
g
n
itio
n
:
a
ca
s
e
…
(
Ow
a
is
Mu
jt
a
b
a
K
h
a
n
d
a
y
)
581
[3
1
]
Be
rn
a
rd
,
S
im
o
n
,
S
é
b
a
stien
A
d
a
m
,
a
n
d
L
a
u
re
n
t
He
u
tt
e
,
“
Us
in
g
Ra
n
d
o
m
F
o
re
sts
f
o
r
Ha
n
d
w
rit
ten
Dig
it
Re
c
o
g
n
it
io
n
,
”
Do
c
u
me
n
t
An
a
lys
is
a
n
d
Rec
o
g
n
it
io
n
(
ICDAR,
2
0
0
7
)
,
v
o
l.
2
,
p
p
.
1
0
4
3
-
1
0
4
7
,
2
0
0
7
.
[3
2
]
Die
tt
e
rich
,
T
h
o
m
a
s
G
.
,
“
A
n
Ex
p
e
rim
e
n
tal
Co
m
p
a
riso
n
o
f
T
h
re
e
M
e
th
o
d
s
f
o
r
Co
n
stru
c
ti
n
g
En
se
m
b
les
o
f
De
c
isio
n
T
re
e
s,
”
M
a
c
h
in
e
lea
rn
i
n
g
,
v
o
l.
1,
p
p
.
2
2
,
1
9
9
9
.
[3
3
]
Ba
ra
n
d
iara
n
,
Iñ
ig
o
,
“
T
h
e
Ra
n
d
o
m
S
u
b
sp
a
c
e
M
e
th
o
d
f
o
r
Co
n
str
u
c
ti
n
g
De
c
isio
n
F
o
re
sts
,
”
IEE
E
T
ra
n
s,
P
a
tt
e
r
n
An
a
l
,
M
a
c
h
.
In
tell
,
v
o
l.
2
0
,
n
o
.
8
,
p
p
.
1
-
2
2
,
1
9
9
8
.
[3
4
]
Ku
n
c
h
e
v
a
,
L
u
d
m
il
a
I
.
,
a
n
d
Co
m
b
in
i
n
g
P
a
tt
e
r
n
Clas
sif
iers
,
“
M
e
th
o
d
s
a
n
d
A
lg
o
rit
h
m
s,
”
Jo
n
h
W
il
e
y
&
S
o
n
s,
Ne
w
Yo
rk
,
NY
,
2
0
0
4
.
[3
5
]
Bre
ima
n
,
L
e
o
,
“
Ra
n
d
o
m
F
o
re
sts,
”
M
a
c
h
in
e
lea
rn
in
g
,
v
o
l
.
4
5
,
n
o
.
1
,
p
p
.
5
-
3
2
,
2
0
0
1
.
[3
6
]
M
a
rk
o
Ro
b
n
ik
,
“
Im
p
ro
v
in
g
Ra
n
d
o
m
F
o
re
sts,
”
ECM
L
’0
4
:
Pro
c
e
e
d
in
g
s
o
f
t
h
e
1
5
t
h
E
u
ro
p
e
a
n
Co
n
fer
e
n
c
e
o
n
M
a
c
h
in
e
L
e
a
r
n
in
g
,
p
p
.
3
5
9
-
3
7
0
,
2
0
0
4
.
[3
7
]
Ne
v
e
s,
Re
n
a
ta F
.
P
.
,
Cleb
e
r
Zan
c
h
e
tt
in
,
a
n
d
A
lb
e
rto
N
.
G
.
L
o
p
e
s F
il
h
o
,
“
A
n
Eff
icie
n
t
W
a
y
o
f
Co
m
b
in
in
g
S
VMs
f
o
r
Ha
n
d
w
rit
ten
Dig
it
Re
c
o
g
n
it
io
n
,
”
P
a
p
e
r
p
re
se
n
ted
a
t
th
e
In
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
Arti
fi
c
i
a
l
Ne
u
ra
l
Ne
two
rk
s
,
p
p
.
2
2
9
-
2
3
7
,
2
0
1
2
.
[3
8
]
Ow
a
i
s
M
.
Kh
a
n
d
a
y
,
S
a
m
a
d
Da
d
v
a
n
d
ip
o
u
r,
“
Co
n
v
o
lu
ti
o
n
a
l
Ne
u
ra
l
Ne
tw
o
rk
s
a
n
d
Im
p
a
c
t
o
f
F
il
ter
S
ize
s
o
n
Im
a
g
e
Clas
sif
ic
a
ti
o
n
,
”
M
u
lt
i
d
iszc
ip
li
n
á
ri
s T
u
d
o
má
n
y
o
k
,
v
o
l.
1
0
,
n
o
.
1
,
p
p
.
5
5
-
6
0
,
2
0
2
0
.
[3
9
]
M
.
Ow
a
is
a
n
d
.
S
.
Da
d
a
n
d
i
p
o
u
r,
“
T
h
e
in
f
lu
e
n
c
e
o
f
th
e
f
il
ter
a
g
g
re
g
a
tes
o
n
th
e
im
a
g
e
c
las
sif
ica
ti
o
n
u
sin
g
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
tw
o
rk
s: a ca
se
stu
d
y
o
f
h
a
n
d
w
rit
ten
d
ig
it
re
c
o
g
n
it
io
n
,
”
i
n
T
a
v
a
sz
i
S
zé
l
K
o
n
fe
re
n
c
ia
2
0
1
9
,
p
p
.
4
3
4
,
2
0
1
9
.
[4
0
]
Ha
ss
a
n
,
A
li
,
a
n
d
Ro
b
e
rt
I
.
Da
m
p
e
r,
“
Clas
sif
ica
ti
o
n
o
f
Em
o
ti
o
n
a
l
S
p
e
e
c
h
Us
in
g
3
d
e
c
Hie
ra
rc
h
ica
l
C
las
sif
ier,
”
S
p
e
e
c
h
Co
mm
u
n
ica
ti
o
n
,
v
o
l
.
5
4
,
n
o
.
7
,
p
p
.
9
0
3
-
9
1
6
,
2
0
1
2
.
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
O
w
a
is
M
u
jta
b
a
K
h
a
n
d
a
y
re
c
e
iv
e
d
h
is
B.
S
c
.
(I.
T
)
f
ro
m
th
e
Un
iv
e
rsity
o
f
Ka
sh
m
ir
(S
.
P
Co
ll
e
g
e
S
rin
a
g
a
r)
a
n
d
M
.
S
c
.
(C
o
m
p
u
ter
S
c
ien
c
e
)
f
ro
m
th
e
Un
iv
e
rsit
y
o
f
P
o
n
d
ic
h
e
rry
.
Cu
rre
n
tl
y
,
h
e
is
a
P
h
.
D.
S
t
u
d
e
n
t
a
t
th
e
Un
iv
e
rsity
o
f
M
isk
o
lc,
Hu
n
g
a
r
y
,
u
n
d
e
r
th
e
S
ti
p
e
n
d
i
u
m
Hu
n
g
a
ricu
m
S
c
h
o
lars
h
i
p
p
ro
g
ra
m
.
a
it
o
w
a
is@u
n
i
-
m
is
k
o
lc.h
u
+
3
6
7
0
4
2
0
2
8
6
5
Dr
.
S
a
m
a
d
Da
d
v
a
n
d
ip
o
u
r
,
A
s
so
c
iate
P
ro
f
e
ss
o
r,
In
sti
tu
te
o
f
In
f
o
rm
a
ti
o
n
S
c
ien
c
e
s,
U
n
iv
e
rsity
o
f
M
isk
o
lc,
3
5
1
5
Hu
n
g
a
r
y
d
r.
sa
m
a
d
@u
n
i
-
m
is
k
o
lc.h
u
Evaluation Warning : The document was created with Spire.PDF for Python.