I
nte
rna
t
io
na
l J
o
urna
l o
f
E
lect
rica
l a
nd
Co
m
p
ute
r
E
ng
in
ee
ring
(
I
J
E
CE
)
Vo
l.
11
,
No
.
5
,
Octo
b
er
2
0
2
1
,
p
p
.
4
3
9
2
~
4
4
0
2
I
SS
N:
2
0
8
8
-
8708
,
DOI
: 1
0
.
1
1
5
9
1
/
i
j
ec
e
.
v
1
1
i
5
.
pp
4
3
9
2
-
4
4
0
2
4392
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
Artif
i
cia
l neural
netw
o
rk
t
echn
iqu
e f
o
r i
m
p
ro
v
ing
predict
io
n of
credit ca
rd de
faul
t:
A
stac
k
ed
spa
rs
e auto
enco
der ap
pro
a
ch
Sa
ra
h.
A.
E
bia
re
do
h
-
M
ieny
e
,
E
.
E
s
eno
g
ho
,
T
heo
G
.
Sw
a
r
t
Ce
n
tre f
o
r
T
e
lec
o
m
m
u
n
ica
ti
o
n
,
D
e
p
a
rtme
n
t
o
f
El
e
c
tri
c
a
l
a
n
d
El
e
c
tr
o
n
ic E
n
g
in
e
e
rin
g
S
c
ien
c
e
,
Un
iv
e
rsit
y
o
f
Jo
h
a
n
n
e
sb
u
rg
,
J
o
h
a
n
n
e
sb
u
rg
,
S
o
u
th
A
f
ri
c
a
Art
icle
I
nfo
AB
ST
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
u
n
21
,
2
0
20
R
ev
i
s
ed
Mar
16
,
2
0
21
A
cc
ep
ted
Mar
26
,
2
0
21
P
re
se
n
tl
y
,
th
e
u
se
o
f
a
c
re
d
it
c
a
rd
h
a
s
b
e
c
o
m
e
a
n
in
teg
ra
l
p
a
rt
o
f
c
o
n
tem
p
o
ra
r
y
b
a
n
k
in
g
a
n
d
f
in
a
n
c
ial
s
y
ste
m
.
P
re
d
ictin
g
p
o
te
n
ti
a
l
c
re
d
it
c
a
rd
d
e
f
a
u
lt
e
rs
o
r
d
e
b
t
o
rs
is
a
c
ru
c
ial
b
u
si
n
e
ss
o
p
p
o
r
tu
n
it
y
f
o
r
f
in
a
n
c
ial
in
stit
u
ti
o
n
s.
F
o
r
n
ow
,
so
m
e
m
a
c
h
in
e
lea
rn
in
g
m
e
th
o
d
s
h
a
v
e
b
e
e
n
a
p
p
li
e
d
to
a
c
h
iev
e
th
is
ta
sk
.
Ho
w
e
v
e
r,
w
it
h
th
e
d
y
n
a
m
ic
a
n
d
i
m
b
a
lan
c
e
d
n
a
tu
re
o
f
c
re
d
it
c
a
rd
d
e
f
a
u
lt
d
a
ta
,
it
is
c
h
a
ll
e
n
g
in
g
f
o
r
c
las
sic
a
l
m
a
c
h
in
e
lea
rn
in
g
a
lg
o
rit
h
m
s
to
p
ro
f
f
e
r
ro
b
u
st
m
o
d
e
ls
w
it
h
o
p
ti
m
a
l
p
e
rf
o
r
m
a
n
c
e
.
Re
se
a
rc
h
h
a
s
sh
o
w
n
th
a
t
th
e
p
e
rf
o
r
m
a
n
c
e
o
f
m
a
c
h
in
e
lea
rn
in
g
a
lg
o
rit
h
m
s
c
a
n
b
e
sig
n
if
ica
n
tl
y
i
m
p
ro
v
e
d
w
h
e
n
p
r
o
v
id
e
d
w
it
h
o
p
ti
m
a
l
f
e
a
tu
re
s.
In
t
h
is
p
a
p
e
r,
w
e
p
ro
p
o
se
a
n
u
n
s
u
p
e
rv
ise
d
fe
a
tu
re
lea
rn
in
g
m
e
th
o
d
to
imp
ro
v
e
th
e
p
e
rf
o
r
m
a
n
c
e
o
f
v
a
rio
u
s
c
las
si
f
i
e
rs
u
sin
g
a
sta
c
k
e
d
sp
a
rse
a
u
to
e
n
c
o
d
e
r
(S
S
A
E)
.
T
h
e
S
S
A
E
wa
s
o
p
ti
m
iz
e
d
to
a
c
h
iev
e
i
m
p
ro
v
e
d
p
e
r
f
o
r
m
a
n
c
e
.
T
h
e
p
ro
p
o
se
d
S
S
A
E
lea
rn
e
d
e
x
c
e
ll
e
n
t
f
e
a
tu
re
re
p
re
se
n
tatio
n
s
th
a
t
w
e
re
u
se
d
t
o
train
th
e
c
las
sif
iers
.
T
h
e
p
e
r
f
o
rm
a
n
c
e
o
f
th
e
p
ro
p
o
se
d
a
p
p
ro
a
c
h
is
c
o
m
p
a
re
d
w
it
h
a
n
in
sta
n
c
e
w
h
e
re
th
e
c
las
si
fiers
w
e
re
train
e
d
u
sin
g
th
e
ra
w
d
a
ta.
A
lso
,
a
c
o
m
p
a
riso
n
is
m
a
d
e
w
it
h
p
re
v
io
u
s
sc
h
o
larly
w
o
rk
s,
a
n
d
th
e
p
ro
p
o
se
d
a
p
p
ro
a
c
h
s
h
o
w
e
d
su
p
e
ri
o
r
p
e
rf
o
r
m
a
n
c
e
o
v
e
r
o
th
e
r
m
e
th
o
d
s.
K
ey
w
o
r
d
s
:
A
r
ti
f
icial
n
e
u
r
al
n
et
w
o
r
k
C
r
ed
it c
ar
d
d
ef
au
lt
Dee
p
lear
n
in
g
Featu
r
e
lear
n
i
n
g
Ma
ch
i
n
e
lear
n
i
n
g
Sp
ar
s
e
au
to
en
co
d
er
Un
s
u
p
er
v
i
s
ed
lear
n
i
n
g
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
:
E
b
en
ez
er
E
s
en
eg
h
o
C
en
tr
e
f
o
r
T
elec
o
m
m
u
n
icatio
n
Dep
ar
t
m
en
t o
f
E
lectr
ical
an
d
E
lectr
o
n
ic
E
n
g
in
ee
r
i
n
g
Scie
n
c
e
Un
i
v
er
s
it
y
o
f
J
o
h
an
n
esb
u
r
g
,
K
in
g
s
w
a
y
A
v
e
n
u
e
an
d
U
n
iv
er
s
i
t
y
R
o
ad
,
Au
c
k
la
n
d
P
ar
k
J
o
h
an
n
e
s
b
u
r
g
2
0
0
6
,
So
u
th
Af
r
ica
E
m
ail: e
b
en
ez
er
e
@
u
j
.
ac
.
za
1.
I
NT
RO
D
UCT
I
O
N
I
n
ar
ti
f
icial
in
telli
g
en
ce
an
d
m
ac
h
in
e
lear
n
i
n
g
,
ta
s
k
s
s
u
c
h
as
clas
s
i
f
icatio
n
a
n
d
cl
u
s
ter
i
n
g
,
t
h
e
i
n
p
u
t
d
ata
ten
d
s
to
in
f
l
u
e
n
ce
th
e
p
er
f
o
r
m
a
n
ce
o
f
th
e
al
g
o
r
ith
m
s
.
O
p
ti
m
al
p
er
f
o
r
m
a
n
ce
is
o
b
tain
e
d
w
h
e
n
alg
o
r
it
h
m
s
ar
e
g
iv
e
n
s
u
itab
le
d
ata.
T
o
th
is
en
d
,
s
o
m
e
m
ac
h
in
e
lear
n
i
n
g
m
et
h
o
d
s
f
o
cu
s
o
n
p
r
o
ce
s
s
in
g
h
ig
h
d
i
m
en
s
io
n
al
d
ata,
in
clu
d
i
n
g
li
n
ea
r
d
i
m
e
n
s
io
n
alit
y
r
ed
u
ctio
n
m
et
h
o
d
s
s
u
c
h
as
li
n
ea
r
d
is
cr
i
m
i
n
a
n
t
an
al
y
s
is
,
p
r
in
c
ip
al
co
m
p
o
n
e
n
t
a
n
al
y
s
is
,
a
n
d
m
u
l
t
ip
le
d
im
e
n
s
io
n
al
s
ca
li
n
g
an
d
n
o
n
li
n
ea
r
d
i
m
e
n
s
io
n
alit
y
r
ed
u
ctio
n
m
et
h
o
d
s
s
u
c
h
as
i
s
o
m
etr
ic
f
ea
t
u
r
e
m
ap
p
i
n
g
,
lo
ca
ll
y
li
n
ea
r
e
m
b
ed
d
in
g
,
an
d
L
ap
lacia
n
E
ig
e
n
m
ap
.
Me
an
w
h
ile,
f
ea
t
u
r
e
en
g
i
n
ee
r
i
n
g
a
n
d
r
ep
r
esen
tatio
n
lear
n
i
n
g
ar
e
t
h
e
t
w
o
m
ain
m
et
h
o
d
s
u
s
ed
to
ac
h
ie
v
e
r
ep
r
esen
tat
io
n
f
r
o
m
r
a
w
d
ata.
R
ec
en
t
r
esear
ch
h
a
s
f
o
c
u
s
ed
o
n
t
h
e
la
t
ter
s
i
n
ce
f
ea
t
u
r
e
en
g
i
n
ee
r
in
g
m
et
h
o
d
s
ar
e
u
s
u
all
y
d
ep
en
d
en
t
o
n
d
o
m
ai
n
k
n
o
w
led
g
e,
ar
e
lab
o
u
r
-
i
n
te
n
s
i
v
e,
a
n
d
ti
m
e
-
co
n
s
u
m
i
n
g
[
1
]
.
Fu
r
th
er
m
o
r
e,
r
ep
r
esen
tat
io
n
lear
n
i
n
g
m
et
h
o
d
s
te
n
d
to
lear
n
a
r
ep
r
esen
tatio
n
f
r
o
m
d
ata
a
u
to
m
atica
ll
y
,
w
h
ic
h
ca
n
t
h
e
n
b
e
u
s
ed
f
o
r
class
i
f
icatio
n
.
A
n
au
to
en
co
d
er
(
A
E
)
is
a
t
y
p
e
o
f
u
n
s
u
p
er
v
is
ed
r
ep
r
esen
tatio
n
le
ar
n
in
g
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2
0
8
8
-
8708
A
r
tifi
cia
l n
eu
r
a
l n
etw
o
r
k
tech
n
iq
u
e
fo
r
imp
r
o
vin
g
p
r
ed
ictio
n
o
f c
r
ed
it c
a
r
d
…
(
S
a
r
a
h
.
A
.
E
b
ia
r
ed
o
h
-
Mien
ye
)
4393
Au
to
en
co
d
er
s
ar
e
u
n
s
u
p
er
v
is
e
d
n
eu
r
al
n
et
w
o
r
k
s
w
it
h
t
h
r
ee
la
y
er
s
,
in
cl
u
d
i
n
g
a
n
in
p
u
t
la
y
er
,
a
h
id
d
en
la
y
er
,
an
d
an
o
u
tp
u
t
la
y
er
.
T
h
e
o
u
tp
u
t la
y
er
ca
n
b
e
co
n
s
id
er
ed
as th
e
r
ec
o
n
s
tr
u
ctio
n
la
y
er
[
2
]
.
T
h
e
s
tr
u
ct
u
r
e
o
f
a
b
asic
au
to
en
co
d
er
is
s
h
o
w
n
in
Fig
u
r
e
1
.
A
u
to
en
co
d
er
s
t
en
d
to
lear
n
a
r
e
p
r
esen
tatio
n
o
f
th
e
in
p
u
t
d
ata,
u
s
u
all
y
f
o
r
d
i
m
e
n
s
io
n
alit
y
r
e
d
u
ctio
n
,
b
y
tr
ai
n
i
n
g
th
e
n
et
wo
r
k
to
i
g
n
o
r
e
n
o
is
e.
T
o
g
eth
er
w
it
h
t
h
e
r
ed
u
c
tio
n
s
id
e,
a
r
ec
o
n
s
tr
u
cti
n
g
s
id
e
is
lear
n
ed
,
w
h
er
e
t
h
e
A
E
atte
m
p
t
s
to
cr
ea
te
a
r
e
p
r
esen
tatio
n
of
th
e
o
r
ig
in
al
i
n
p
u
t
[
3
]
,
[
4
]
.
T
h
er
e
ar
e
d
if
f
er
e
n
t
t
y
p
es
o
f
au
to
e
n
co
d
er
s
,
in
clu
d
i
n
g
s
p
ar
s
e,
d
en
o
is
i
n
g
,
co
n
tr
ac
ti
v
e,
v
ar
iatio
n
al,
an
d
co
n
v
o
lu
tio
n
al
a
u
to
en
co
d
er
s
[
5
]
.
Fig
u
r
e
1
.
Stru
ct
u
r
e
o
f
an
a
u
to
en
co
d
er
C
r
ed
it
ca
r
d
d
ef
au
lt
/f
r
a
u
d
d
etec
tio
n
is
a
cr
u
cial
p
r
o
b
lem
t
h
at
h
as
g
o
tte
n
t
h
e
atte
n
tio
n
o
f
m
ac
h
i
n
e
lear
n
in
g
r
esear
c
h
er
s
,
an
d
a
s
ig
n
if
ican
t
n
u
m
b
er
o
f
ap
p
r
o
ac
h
es
h
av
e
b
ee
n
p
r
o
p
o
s
ed
[
6
]
-
[
9
]
.
Ho
w
e
v
er
,
th
e
p
r
o
b
lem
is
s
ti
ll
c
h
allen
g
i
n
g
s
in
ce
m
o
s
t
cr
ed
it
ca
r
d
d
ata
s
e
e
m
to
s
u
f
f
er
f
r
o
m
clas
s
i
m
b
a
lan
ce
as
n
o
n
-
f
r
au
d
tr
an
s
ac
tio
n
s
o
v
er
w
h
el
m
in
g
l
y
s
u
p
er
s
ed
e
f
r
au
d
tr
an
s
ac
tio
n
s
,
m
a
k
i
n
g
it
d
if
f
icu
l
t
f
o
r
m
a
n
y
m
ac
h
i
n
e
lear
n
i
n
g
alg
o
r
ith
m
s
to
ac
h
ie
v
e
g
o
o
d
p
e
r
f
o
r
m
an
ce
.
Me
a
n
w
h
ile,
a
g
o
o
d
f
ea
tu
r
e
r
ep
r
esen
tatio
n
ca
n
b
e
o
b
tain
ed
f
r
o
m
th
e
d
ataset,
w
h
ic
h
ca
n
e
n
h
an
ce
t
h
e
clas
s
if
icatio
n
p
er
f
o
r
m
an
ce
o
f
th
e
al
g
o
r
ith
m
s
.
R
ep
r
esen
t
atio
n
lear
n
i
n
g
is
a
p
o
s
s
ib
le
s
o
lu
tio
n
to
th
e
ch
alle
n
g
e
o
f
cr
ed
it
ca
r
d
d
ef
au
lt
an
d
f
r
au
d
p
r
ed
ictio
n
b
ec
au
s
e
o
f
it
s
r
em
ar
k
ab
le
f
ea
t
u
r
e
lear
n
in
g
ab
ilit
y
i
n
lar
g
e
an
d
u
n
b
alan
ce
d
d
ataset
s
.
W
h
ile
b
as
ic
au
to
en
co
d
er
s
ai
m
at
lear
n
i
n
g
a
r
ep
r
ese
n
tatio
n
o
r
en
co
d
e
d
ata
b
y
tr
ai
n
i
n
g
th
e
n
et
w
o
r
k
to
ig
n
o
r
e
n
o
is
e
an
d
r
ec
o
n
s
tr
u
ct
in
g
t
h
e
d
ata
as
clo
s
e
as
p
o
s
s
ib
le
to
t
h
e
in
p
u
t
d
ata,
h
o
w
e
v
er
,
tr
ain
in
g
th
e
au
to
en
co
d
er
n
et
w
o
r
k
i
n
s
u
ch
a
w
a
y
t
h
at
e
n
co
u
r
ag
e
s
s
p
ar
s
it
y
ca
n
r
es
u
lt
i
n
o
p
tim
a
l
f
ea
t
u
r
e
lear
n
in
g
.
S
p
ar
s
it
y
i
n
d
u
ce
d
n
e
u
r
al
n
e
t
wo
r
k
s
h
a
v
e
b
ee
n
e
x
te
n
s
i
v
el
y
ap
p
lied
in
i
m
a
g
e
r
ec
o
g
n
itio
n
an
d
s
e
v
er
al
o
th
er
ap
p
licatio
n
s
r
esu
lti
n
g
i
n
s
tate
-
of
-
t
h
e
-
ar
t p
er
f
o
r
m
a
n
ce
[
1
0
]
-
[
1
2
]
.
I
n
th
i
s
p
ap
er
,
an
ap
p
r
o
ac
h
is
p
r
o
p
o
s
ed
to
im
p
r
o
v
e
t
h
e
cla
s
s
if
ica
tio
n
p
er
f
o
r
m
an
ce
o
f
clas
s
if
ier
s
b
y
u
s
i
n
g
t
h
e
u
n
s
u
p
er
v
is
ed
f
ea
t
u
r
e
lear
n
i
n
g
ca
p
ab
ilit
y
o
f
a
u
to
e
n
co
d
er
s
.
Du
r
i
n
g
th
e
tr
ain
in
g
o
f
t
h
e
au
to
en
co
d
er
,
s
p
ar
s
it
y
is
e
n
co
u
r
a
g
ed
,
an
d
th
e
m
o
d
el
is
o
p
ti
m
ized
u
s
i
n
g
t
h
e
A
d
aM
a
x
al
g
o
r
ith
m
[
1
3
]
in
s
tead
o
f
th
e
co
n
v
e
n
tio
n
al
s
to
c
h
asti
c
g
r
ad
i
en
t
d
esce
n
t.
T
o
en
s
u
r
e
ac
c
u
r
ate
f
ea
t
u
r
e
r
ep
r
esen
tatio
n
,
we
s
tack
t
w
o
s
p
ar
s
e
au
to
en
co
d
er
s
to
g
et
th
e
f
i
n
al
m
o
d
el.
A
ls
o
,
to
f
u
r
t
h
er
p
r
ev
en
t
o
v
er
f
itti
n
g
a
n
d
en
h
a
n
ce
th
e
p
er
f
o
r
m
a
n
ce
,
s
p
ee
d
,
an
d
s
tab
ilit
y
o
f
t
h
e
n
et
w
o
r
k
,
w
e
in
tr
o
d
u
ce
d
t
h
e
b
atc
h
n
o
r
m
aliza
tio
n
tec
h
n
iq
u
e
[
1
4
]
to
th
e
n
et
w
o
r
k
.
T
h
e
lo
w
-
d
i
m
en
s
io
n
al
f
ea
tu
r
e
s
ar
e
t
h
en
u
s
ed
to
tr
ain
v
ar
io
u
s
c
lass
i
f
ier
s
,
in
cl
u
d
i
n
g
lo
g
is
tic
r
eg
r
ess
i
o
n
(
L
R
)
,
class
if
i
ca
t
io
n
an
d
r
eg
r
ess
i
o
n
t
r
e
e
(
C
A
R
T
)
,
k
-
n
e
ar
est
n
eig
h
b
o
r
(
KNN
)
,
s
u
p
p
o
r
t
v
ec
t
o
r
m
a
ch
in
e
(
SVM
)
,
an
d
lin
ea
r
d
is
cr
im
in
an
t
an
aly
s
is
(
L
DA
)
.
T
h
e
p
er
f
o
r
m
a
n
ce
o
f
t
h
e
p
r
o
p
o
s
ed
m
e
th
o
d
is
co
m
p
ar
ed
w
it
h
a
n
in
s
ta
n
ce
w
h
er
e
t
h
e
c
lass
if
ier
s
w
er
e
tr
ai
n
ed
w
it
h
th
e
r
a
w
d
ata.
Fu
r
t
h
er
co
m
p
ar
is
o
n
is
m
ad
e
w
it
h
o
t
h
er
s
ch
o
lar
l
y
w
o
r
k
s
,
an
d
o
u
r
p
r
o
p
o
s
ed
m
et
h
o
d
s
h
o
w
s
b
etter
p
er
f
o
r
m
a
n
ce
.
T
h
e
m
ain
co
n
tr
ib
u
t
io
n
s
o
f
th
i
s
s
t
u
d
y
ca
n
b
e
s
u
m
m
ar
ized
is
b
ein
g
as
:
T
o
c
o
n
s
tr
u
ct
a
n
ef
f
e
cti
v
e
ar
ti
f
icial
n
eu
r
al
n
et
w
o
r
k
f
o
r
f
ea
t
u
r
e
lear
n
i
n
g
u
s
in
g
m
u
ltip
le
l
a
y
er
s
o
f
s
p
ar
s
e
au
to
en
co
d
er
.
T
o
im
p
r
o
v
e
th
e
cla
s
s
i
f
icat
io
n
p
er
f
o
r
m
a
n
ce
o
f
v
ar
io
u
s
cl
ass
i
f
ier
s
u
s
i
n
g
t
h
e
p
r
o
p
o
s
ed
s
tack
ed
s
p
ar
s
e
au
to
en
co
d
er
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
n
t J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
11
,
No
.
5
,
Octo
b
e
r
2
0
2
1
:
4
3
9
2
-
4402
4394
T
o
d
em
o
n
s
tr
ate
th
e
e
f
f
ec
ti
v
en
ess
o
f
t
h
e
p
r
o
p
o
s
ed
m
et
h
o
d
b
y
ap
p
ly
i
n
g
i
t to
a
p
o
p
u
lar
cr
ed
it
ca
r
d
d
ataset.
T
h
e
r
est
o
f
th
e
p
ap
er
is
o
r
g
an
i
ze
d
is
b
ein
g
a
s
.
I
n
s
ec
tio
n
2
,
w
e
b
r
ief
l
y
r
e
v
ie
w
p
r
ev
io
u
s
r
el
ated
w
o
r
k
s
th
at
u
tili
ze
d
d
if
f
er
en
t
t
y
p
es
o
f
au
to
en
co
d
er
s
.
Sectio
n
3
p
r
ese
n
ts
t
h
e
p
r
o
p
o
s
ed
m
et
h
o
d
an
d
s
ec
tio
n
4
p
r
o
v
id
es
a
b
r
ief
ca
s
e
s
t
u
d
y
o
f
cr
ed
it
ca
r
d
d
ef
au
lt
in
g
p
r
ed
ictio
n
m
o
d
el
s
.
T
h
e
o
b
tain
ed
r
esu
lts
ar
e
p
r
ese
n
ted
an
d
d
is
c
u
s
s
ed
in
s
ec
tio
n
5
.
L
ast
l
y
,
s
ec
t
io
n
6
co
n
clu
d
es t
h
e
p
ap
er
an
d
h
ig
h
li
g
h
t
s
s
o
m
e
f
u
t
u
r
e
r
esear
ch
d
ir
ec
tio
n
s
.
2.
RE
L
AT
E
D
WO
RK
S
R
ec
en
t
l
y
,
a
u
to
en
co
d
er
s
h
a
v
e
b
ee
n
ap
p
lied
to
s
ev
er
al
ta
s
k
s
,
an
d
t
h
e
y
ac
h
ie
v
ed
s
tat
e
-
of
-
t
h
e
-
ar
t
p
er
f
o
r
m
a
n
ce
.
I
n
th
is
s
ec
tio
n
,
w
e
d
is
c
u
s
s
s
o
m
e
p
r
ev
io
u
s
w
o
r
k
s
th
at
u
tili
ze
d
v
ar
io
u
s
au
to
en
co
d
er
s
an
d
la
y
th
e
f
o
u
n
d
atio
n
f
o
r
th
e
p
r
o
p
o
s
ed
s
t
ac
k
ed
s
p
ar
s
e
au
to
en
co
d
er
n
et
w
o
r
k
.
S
u
n
et
a
l.
[
1
5
]
p
r
o
p
o
s
e
d
a
m
et
h
o
d
f
o
r
f
au
l
t
d
iag
n
o
s
i
s
b
y
ap
p
l
y
i
n
g
a
s
p
ar
s
e
s
tack
ed
d
e
n
o
is
i
n
g
au
to
e
n
co
d
er
f
o
r
f
ea
t
u
r
e
ex
tr
ac
tio
n
d
u
e
t
o
its
r
o
b
u
s
t
n
es
s
a
n
d
d
ata
r
ec
o
n
s
tr
u
ctio
n
ca
p
ab
ilit
y
,
w
h
ich
i
m
p
r
o
v
ed
th
e
d
iag
n
o
s
t
ic
ac
cu
r
ac
y
.
T
h
e
au
to
en
co
d
er
w
a
s
u
s
ed
to
g
et
h
er
w
it
h
an
o
p
ti
m
ized
tr
an
s
f
er
le
ar
n
in
g
alg
o
r
it
h
m
.
Si
m
ilar
l
y
,
Z
h
u
et
a
l.
[
1
6
]
p
r
o
p
o
s
ed
a
n
o
v
el
s
tac
k
ed
p
r
u
n
in
g
s
p
ar
s
e
d
en
o
is
i
n
g
a
u
to
en
co
d
er
f
o
r
in
tell
ig
e
n
t
f
au
lt
d
iag
n
o
s
i
s
o
f
r
o
llin
g
b
ea
r
in
g
s
.
T
h
e
m
et
h
o
d
co
m
p
r
is
ed
o
f
a
f
u
ll
y
co
n
n
ec
ted
a
u
to
en
co
d
er
n
et
w
o
r
k
,
co
n
n
ec
tin
g
th
e
o
p
tim
al
f
ea
t
u
r
es
ex
tr
ac
ted
f
r
o
m
p
r
ev
io
u
s
la
y
er
s
to
s
u
b
s
eq
u
en
t
la
y
er
s
.
T
o
ef
f
ec
ti
v
el
y
tr
ai
n
t
h
e
a
u
to
en
co
d
er
,
a
p
r
u
n
i
n
g
o
p
er
atio
n
w
a
s
ad
d
e
d
to
th
e
m
o
d
el
to
r
estrict
n
o
n
-
s
u
p
er
io
r
u
n
it
s
f
r
o
m
p
ar
ticip
atin
g
i
n
all
s
u
b
s
e
q
u
en
t
la
y
er
s
.
W
h
en
co
m
p
ar
e
d
w
it
h
o
t
h
er
f
a
u
l
t
d
iag
n
o
s
t
ic
m
o
d
els,
t
h
eir
ap
p
r
o
ac
h
s
h
o
w
ed
s
u
p
er
io
r
p
er
f
o
r
m
a
n
ce
.
Fu
r
t
h
er
m
o
r
e,
San
k
ar
an
et
a
l.
[
1
7
]
p
r
o
p
o
s
ed
a
f
ea
tu
r
e
ex
tr
ac
tio
n
m
et
h
o
d
u
s
i
n
g
an
a
u
to
en
co
d
er
n
et
w
o
r
k
,
an
d
ℓ
2,
1
-
n
o
r
m
b
ased
r
eg
u
lar
izatio
n
w
as
u
s
ed
to
ac
h
iev
e
s
p
ar
s
it
y
.
T
h
e
au
t
h
o
r
s
id
en
ti
f
ied
th
at
d
u
e
to
th
e
p
r
esen
ce
o
f
m
a
n
y
tr
ai
n
i
n
g
p
ar
am
eter
s
,
s
e
v
er
al
f
ea
t
u
r
e
lear
n
in
g
m
o
d
els
ar
e
s
u
s
ce
p
tib
le
to
o
v
er
f
itti
n
g
,
an
d
d
if
f
er
e
n
t
r
e
g
u
lar
izat
io
n
ap
p
r
o
ac
h
es
h
a
v
e
b
ee
n
s
t
u
d
ied
in
l
i
ter
atu
r
e
to
m
iti
g
ate
o
v
er
f
itti
n
g
i
n
d
ee
p
lear
n
i
n
g
m
o
d
el
s
.
T
h
e
p
er
f
o
r
m
an
ce
o
f
t
h
eir
m
o
d
el
w
a
s
s
t
u
d
ied
o
n
p
u
b
licl
y
av
ailab
le
late
n
t
f
i
n
g
er
p
r
in
t
d
atasets
,
an
d
it
g
av
e
an
i
m
p
r
o
v
ed
p
er
f
o
r
m
an
ce
.
C
h
e
n
e
t
a
l.
[
1
8
]
p
r
o
p
o
s
ed
a
m
et
h
o
d
to
ad
d
r
ess
t
h
e
c
h
alle
n
g
e
o
f
lear
n
i
n
g
ef
f
icien
c
y
an
d
co
m
p
u
ta
tio
n
al
co
m
p
lex
it
y
i
n
d
ee
p
n
eu
r
al
n
et
w
o
r
k
s
.
T
h
e
tech
n
iq
u
e
u
s
ed
a
d
ee
p
s
p
ar
s
e
au
to
en
co
d
er
n
et
w
o
r
k
to
lear
n
f
ac
ial
f
ea
t
u
r
es
a
n
d
s
o
f
t
m
a
x
r
eg
r
ess
io
n
ap
p
lied
to
class
i
f
y
ex
p
r
ess
io
n
f
ea
t
u
r
es.
T
h
e
s
o
f
t
m
a
x
r
e
g
r
ess
io
n
ai
m
ed
at
h
a
n
d
lin
g
e
x
te
n
s
i
v
e
d
ata
i
n
th
e
o
u
tp
u
t
o
f
th
e
a
u
to
e
n
co
d
er
n
et
w
o
r
k
.
A
l
s
o
,
to
o
v
er
co
m
e
lo
ca
l
e
x
tr
e
m
a
a
n
d
th
e
c
h
alle
n
g
e
o
f
g
r
ad
ien
t
d
if
f
u
s
io
n
d
u
r
i
n
g
tr
ai
n
in
g
,
t
h
e
n
et
w
o
r
k
w
ei
g
h
t
s
w
er
e
f
i
n
etu
n
ed
,
an
d
th
is
i
m
p
r
o
v
ed
th
e
p
er
f
o
r
m
a
n
ce
o
f
th
e
ar
c
h
itec
tu
r
e.
Mo
s
t
ap
p
r
o
ac
h
es
u
s
ed
to
i
m
p
l
e
m
en
t
a
u
to
en
co
d
er
s
d
ep
en
d
o
n
t
h
e
s
i
n
g
le
a
u
to
en
co
d
er
m
o
d
el,
an
d
t
h
i
s
p
r
esen
ts
a
p
r
o
b
lem
w
h
e
n
lear
n
in
g
d
if
f
er
en
t
ch
ar
ac
ter
is
tic
s
o
f
d
ata.
Yan
g
et
a
l.
[
1
9
]
p
r
o
p
o
s
ed
a
m
eth
o
d
to
s
o
lv
e
t
h
e
p
r
o
b
le
m
b
y
i
m
p
le
m
en
ti
n
g
a
f
ea
t
u
r
e
lear
n
in
g
f
r
a
m
e
w
o
r
k
u
s
in
g
s
er
ia
l
au
to
e
n
co
d
er
s
.
T
h
e
tech
n
iq
u
e
ac
h
iev
ed
s
u
p
er
io
r
r
ep
r
esen
tatio
n
lear
n
i
n
g
b
y
s
er
iall
y
co
n
n
ec
tin
g
t
w
o
d
i
f
f
er
en
t
t
y
p
es
o
f
au
to
en
co
d
er
s
.
T
h
e
ap
p
r
o
ac
h
in
co
r
p
o
r
ated
tw
o
e
n
co
d
in
g
s
tag
e
s
u
s
i
n
g
a
m
ar
g
i
n
a
lized
d
en
o
is
i
n
g
a
u
to
e
n
co
d
er
an
d
a
s
tac
k
ed
r
o
b
u
s
t
au
to
en
co
d
er
v
ia
g
r
ap
h
r
eg
u
lar
izatio
n
.
W
h
en
co
m
p
ar
ed
to
b
aselin
e
m
e
t
h
o
d
s
,
th
e
p
r
o
p
o
s
e
d
ap
p
r
o
ac
h
s
h
o
w
ed
s
ig
n
i
f
ica
n
t
i
m
p
r
o
v
e
m
e
n
t.
Me
an
w
h
ile,
A
l
-
H
m
o
u
z
et
a
l.
[
2
0
]
,
in
tr
o
d
u
ce
d
a
lo
g
ic
-
d
r
iv
e
n
au
t
o
en
co
d
er
,
w
h
er
eb
y
th
e
n
et
w
o
r
k
s
tr
u
ct
u
r
e
w
as
ac
h
iev
ed
u
s
i
n
g
s
o
m
e
f
u
zz
y
lo
g
ic
o
p
er
atio
n
s
.
T
h
e
au
to
en
co
d
er
w
a
s
also
o
p
ti
m
ized
u
s
i
n
g
g
r
ad
ie
n
t
-
b
ased
lear
n
i
n
g
.
L
as
tl
y
,
s
p
ar
s
e
a
u
to
en
co
d
er
n
e
t
w
o
r
k
s
h
av
e
ac
h
ie
v
ed
r
e
m
ar
k
a
b
le
p
er
f
o
r
m
a
n
ce
i
n
r
ep
r
esen
tatio
n
lear
n
i
n
g
[
2
1
]
,
[
2
2
]
.
Ho
w
e
v
er
,
b
etter
r
ep
r
e
s
en
tat
io
n
lear
n
i
n
g
ca
n
b
e
g
o
tten
w
h
e
n
m
u
l
tip
le
s
p
ar
s
e
au
to
en
co
d
er
s
ar
e
s
tack
e
d
an
d
o
p
tim
ized
e
f
f
ec
t
iv
e
l
y
,
wh
ich
i
s
t
h
e
f
o
cu
s
o
f
t
h
is
r
esear
ch
.
3.
P
RO
P
O
SE
D
M
E
T
H
O
D
T
h
is
s
ec
tio
n
co
n
s
id
er
s
t
h
e
m
et
h
o
d
ap
p
lied
to
d
ev
elo
p
in
g
th
e
p
r
o
p
o
s
ed
au
to
en
co
d
er
.
A
n
a
u
to
en
co
d
er
co
n
s
is
ts
o
f
t
w
o
f
u
n
ctio
n
s
,
i.e
.
,
an
en
co
d
er
an
d
d
ec
o
d
er
,
th
e
f
o
r
m
er
m
ap
s
t
h
e
d
-
d
i
m
e
n
s
io
n
al
in
p
u
t
d
ata
to
g
et
a
h
id
d
en
r
ep
r
esen
tatio
n
,
an
d
th
e
latter
m
ap
s
t
h
e
h
id
d
en
r
ep
r
es
en
tatio
n
b
ac
k
to
a
d
-
d
i
m
e
n
s
io
n
al
v
ec
to
r
th
at
i
s
as
clo
s
e
as p
o
s
s
ib
le
to
th
e
e
n
co
d
er
in
p
u
t
[
2
3
]
.
A
s
s
u
m
i
n
g
th
e
o
r
ig
in
a
l i
n
p
u
t i
s
,
th
e
a
u
to
en
co
d
e
r
en
co
d
es it i
n
to
a
h
id
d
en
la
y
er
ℎ
to
r
ed
u
ce
th
e
i
n
p
u
t
d
i
m
e
n
s
io
n
,
w
h
ich
is
th
e
n
d
ec
o
d
ed
at
th
e
o
u
tp
u
t.
T
h
e
in
p
u
t
v
ec
to
r
is
en
co
d
ed
ac
co
r
d
in
g
to
:
ℎ
=
(
+
)
(
1
)
w
h
er
e
r
ep
r
esen
ts
t
h
e
ac
ti
v
ati
o
n
f
u
n
c
tio
n
;
i
n
t
h
is
ca
s
e,
th
e
s
ig
m
o
id
ac
tiv
a
tio
n
f
u
n
ctio
n
,
is
th
e
w
e
ig
h
t
m
atr
i
x
,
an
d
is
a
b
ias
v
ec
to
r
.
T
h
e
h
id
d
en
r
ep
r
esen
tatio
n
i
s
d
ec
o
d
ed
to
g
et
th
e
d
ata
as
clo
s
e
as
p
o
s
s
ib
le
to
th
e
in
p
u
t
u
s
i
n
g
:
̂
=
(
′
ℎ
+
′
)
(
2
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2
0
8
8
-
8708
A
r
tifi
cia
l n
eu
r
a
l n
etw
o
r
k
tech
n
iq
u
e
fo
r
imp
r
o
vin
g
p
r
ed
ictio
n
o
f c
r
ed
it c
a
r
d
…
(
S
a
r
a
h
.
A
.
E
b
ia
r
ed
o
h
-
Mien
ye
)
4395
w
h
er
e
′
is
w
eig
h
t m
atr
ix
an
d
′
r
ep
r
es
en
ts
th
e
b
ias v
e
ct
o
r
[
2
4
]
.
T
h
e
s
ig
m
o
id
ac
tiv
a
tio
n
f
u
n
ct
i
o
n
is
d
escr
ib
ed
as:
=
1
1
+
−
(
3)
T
h
e
d
is
p
ar
ity
b
et
w
ee
n
th
e
o
r
ig
in
a
l
i
n
p
u
t
an
d
t
h
e
r
ec
o
n
s
tr
u
cted
in
p
u
t
̂
is
ca
ll
e
d
th
e
r
ec
o
n
s
tr
u
ct
io
n
er
r
o
r
.
T
o
o
p
tim
ize
th
e
p
ar
am
ete
r
s
W
,
′
,
,
′
,
th
e
m
ea
n
s
q
u
a
r
e
d
er
r
o
r
(
MS
E
)
f
u
n
cti
o
n
is
u
s
ed
as
th
e
r
e
co
n
s
t
r
u
ct
io
n
e
r
r
o
r
f
u
n
ct
io
n
:
=
1
∑
‖
̂
−
‖
2
=
1
(
4
)
T
h
e
av
er
ag
e
ac
tiv
at
io
n
o
f
n
eu
r
o
n
s
in
t
h
e
h
id
d
en
la
y
er
i
s
r
ep
r
esen
ted
as:
̂
=
1
∑
ℎ
(
)
=
1
(
5
)
T
o
in
d
u
ce
s
p
ar
s
i
ty
in
th
e
au
to
en
co
d
er
,
w
e
lim
it
̂
=
,
w
h
er
e
is
t
h
e
s
p
ar
s
ity
p
r
o
p
o
r
ti
o
n
,
an
d
it
is
u
s
u
ally
a
s
m
all
p
o
s
itiv
e
n
u
m
b
er
n
ea
r
0
.
T
h
e
r
ef
o
r
e,
w
e
tr
y
to
m
in
im
ize
th
e
k
u
llb
ac
k
-
leib
l
er
(
KL
)
d
iv
e
r
g
en
c
e
b
etw
ee
n
̂
an
d
ac
c
o
r
d
in
g
to
:
(
|
|
̂
)
=
l
og
(
̂
)
+
(
1
−
)
l
og
(
1
−
1
−
̂
)
(
6
)
A
ls
o
,
t
o
en
s
u
r
e
b
et
te
r
f
e
atu
r
e
r
e
p
r
esen
t
ati
o
n
an
d
,
b
y
ex
te
n
s
io
n
,
en
h
an
ce
th
e
p
er
f
o
r
m
an
ce
o
f
th
e
class
if
i
er
s
,
m
u
ltip
l
e
s
p
a
r
s
e
au
t
o
en
c
o
d
e
r
s
ar
e
s
t
ac
k
e
d
.
A
s
ta
c
k
ed
s
p
a
r
s
e
au
t
o
en
co
d
e
r
(
SS
A
E
)
ca
n
c
o
m
p
r
is
e
o
f
n
u
m
er
o
u
s
s
p
ar
s
e
au
t
o
en
co
d
e
r
s
w
h
er
eb
y
th
e
o
u
t
p
u
ts
o
f
e
ac
h
l
ay
er
ar
e
c
o
n
n
ec
t
e
d
t
o
th
e
in
p
u
ts
o
f
th
e
n
ex
t
l
ay
er
[
2
5
]
.
T
h
e
SS
A
E
is
b
ase
d
o
n
r
es
ea
r
ch
co
n
d
u
ct
ed
b
y
Hin
to
n
an
d
Sal
ak
h
u
td
in
o
v
[
2
6
]
,
w
h
e
r
e
th
ey
p
r
o
p
o
s
e
d
a
d
e
ep
n
eu
r
al
n
etw
o
r
k
w
ith
lay
er
b
y
lay
er
in
iti
ali
za
t
io
n
.
T
h
e
e
r
r
o
r
f
u
n
cti
o
n
o
f
th
e
SS
A
E
is
ex
p
r
ess
e
d
as:
(
,
)
=
1
∑
[
1
2
‖
(
(
)
+
)
−
(
)
‖
2
]
+
2
∑
∑
∑
[
(
)
]
2
=
1
=
1
−
1
=
1
=
1
(
7
)
w
h
er
e
an
d
r
e
p
r
esen
ts
th
e
n
u
m
b
er
o
f
s
am
p
les
an
d
th
e
n
u
m
b
e
r
o
f
l
ay
er
s
,
r
es
p
ec
tiv
ely
,
th
e
o
r
ig
in
al
in
p
u
t
is
,
an
d
d
en
o
tes
th
e
c
o
r
r
es
p
o
n
d
i
n
g
la
b
el
.
T
h
e
r
eg
u
la
r
i
za
t
io
n
c
o
ef
f
icien
t
is
r
e
p
r
esen
t
ed
b
y
.
an
d
d
en
o
t
es
th
e
r
o
w
s
an
d
c
o
lu
m
n
s
o
f
th
e
m
atr
ix
(
)
[
2
7
]
.
B
y
a
d
d
in
g
th
e
s
p
ar
s
i
ty
ter
m
t
o
(
7
)
,
th
e
o
v
er
all
co
s
t
f
u
n
cti
o
n
o
f
th
e
SS
A
E
b
ec
o
m
es:
(
,
)
=
(
,
)
+
∑
(
|
|
̂
)
=
1
(
8
)
w
h
er
e
S
r
e
p
r
es
en
ts
th
e
t
o
t
al
n
u
m
b
er
o
f
n
eu
r
o
n
s
in
a
lay
er
an
d
is
th
e
s
p
a
r
s
i
ty
r
eg
u
l
ar
i
za
ti
o
n
p
a
r
am
ete
r
,
an
d
i
t
s
ets
th
e
s
p
ar
s
i
ty
p
en
a
lty
te
r
m
.
W
e
n
o
w
h
av
e
th
r
e
e
o
p
tim
izat
i
o
n
p
a
r
am
ete
r
s
,
in
c
lu
d
in
g
,
,
an
d
,
an
d
w
e
s
et
th
eir
v
a
lu
es
as
3
,
0
.
0
0
0
1
,
an
d
0
.
0
5
,
r
es
p
ec
tiv
ely
.
I
n
th
e
s
p
a
r
s
e
au
t
o
en
c
o
d
er
n
etw
o
r
k
,
a
n
eu
r
o
n
is
s
ai
d
t
o
b
e
ac
tiv
e
if
its
o
u
t
p
u
t
is
a
v
a
lu
e
c
lo
s
e
to
1
,
w
h
ile
it
is
in
ac
tiv
e
if
its
o
u
t
p
u
t
is
a
v
alu
e
cl
o
s
e
r
t
o
0
[
8
]
.
A
lg
o
r
ith
m
1
s
h
o
w
s
t
h
e
p
r
o
p
o
s
ed
s
p
ar
s
e
a
u
to
en
co
d
er
p
r
o
ce
d
u
r
e.
Fig
u
r
e
2
s
h
o
w
s
th
e
s
tr
u
ctu
r
e
o
f
th
e
p
r
o
p
o
s
e
d
s
ta
ck
e
d
s
p
a
r
s
e
au
t
o
en
co
d
er
(
SS
A
E
)
.
Fo
r
s
im
p
li
city
,
th
e
d
e
c
o
d
er
p
a
r
ts
o
f
th
e
SA
E
ar
e
n
o
t
s
h
o
w
n
.
T
h
e
o
u
tp
u
t
o
f
th
e
SS
A
E
is
th
en
u
s
e
d
t
o
t
r
a
in
th
e
v
ar
i
o
u
s
class
if
ie
r
s
.
A
l
g
o
r
ith
m
1
.
P
r
o
p
o
s
ed
m
et
h
o
d
o
f
th
e
SS
A
E
Input:
train
set x
Process:
Start
Initialize
,
,
′
,
,
′
Ob
t
ai
n
t
he
co
st
f
u
nc
ti
o
n
ac
c
or
d
in
g
t
o
(4
)
Apply weight penalty to the cost function according to (7)
Add the sparsity regularizer to the cost function according to (8)
Train network until convergence
End
Output:
Reconstructed representation of the input
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
n
t J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
11
,
No
.
5
,
Octo
b
e
r
2
0
2
1
:
4
3
9
2
-
4402
4396
T
h
e
g
r
e
ed
y
lay
er
-
w
is
e
tr
ain
in
g
s
tr
at
eg
y
p
r
o
p
o
s
e
d
b
y
B
en
g
i
o
et
a
l
.
[
2
8
]
is
em
p
l
o
y
ed
to
s
u
cc
ess
iv
e
ly
tr
a
in
ev
e
r
y
lay
er
o
f
th
e
SS
A
E
in
o
r
d
e
r
t
o
o
b
tain
ac
ce
s
s
t
o
t
h
e
w
eig
h
ts
an
d
b
ias
p
a
r
am
ete
r
s
o
f
th
e
n
etw
o
r
k
.
A
ls
o
,
th
e
n
e
tw
o
r
k
is
f
in
etu
n
e
d
u
s
in
g
th
e
b
a
ck
p
r
o
p
ag
at
io
n
al
g
o
r
i
th
m
to
o
b
tain
th
e
b
est
p
a
r
am
eter
s
ettin
g
s
.
T
h
e
A
d
aM
ax
alg
o
r
i
th
m
[
1
3
]
,
a
v
a
r
ian
t
o
f
th
e
a
d
a
p
t
iv
e
m
o
m
en
t
e
s
tim
atio
n
(
A
d
am
)
alg
o
r
ith
m
th
at
u
s
es
th
e
in
f
in
ity
n
o
r
m
,
w
as
ap
p
li
e
d
to
o
p
t
im
ize
th
e
au
to
en
c
o
d
e
r
n
etw
o
r
k
.
L
astl
y
,
w
e
in
tr
o
d
u
ce
d
th
e
b
at
ch
n
o
r
m
aliz
ati
o
n
tech
n
i
q
u
e
[
1
4
]
t
o
p
r
ev
en
t
o
v
e
r
f
itti
n
g
an
d
en
h
an
ce
th
e
p
e
r
f
o
r
m
an
ce
,
s
p
ee
d
,
an
d
s
ta
b
i
lity
o
f
t
h
e
n
etw
o
r
k
.
Fig
u
r
e
2
.
Stru
ct
u
r
e
o
f
th
e
p
r
o
p
o
s
ed
SS
A
E
m
o
d
el
4.
CASE
S
T
UD
Y
O
F
CRED
I
T
CARD
DE
F
AU
L
T
I
N
G
P
R
E
DIC
T
I
O
N
M
O
DE
L
S
C
r
ed
it
r
is
k
p
la
y
s
a
cr
u
c
ial
r
o
le
in
th
e
f
in
a
n
cial
i
n
d
u
s
tr
y
.
Mo
s
t
f
i
n
a
n
cial
i
n
s
tit
u
tio
n
s
g
r
an
t
lo
a
n
s
,
m
o
r
t
g
a
g
e,
an
d
cr
ed
it
ca
r
d
s
,
a
m
o
n
g
m
a
n
y
o
t
h
er
s
er
v
ice
s
.
Du
e
to
th
e
r
is
in
g
n
u
m
b
er
o
f
cr
ed
it
ca
r
d
clien
ts
,
th
ese
in
s
t
itu
tio
n
s
h
a
v
e
f
ac
ed
an
i
n
c
r
ea
s
in
g
d
ef
a
u
lt
r
ate.
T
h
e
y
ar
e
th
er
eb
y
r
eso
r
ti
n
g
to
th
e
u
s
e
o
f
m
ac
h
i
n
e
lear
n
in
g
m
et
h
o
d
s
to
au
to
m
ate
th
e
ap
p
licatio
n
p
r
o
ce
s
s
an
d
p
r
ed
ict
th
e
p
r
o
b
a
b
ilit
y
o
f
a
clien
t
’
s
f
u
t
u
r
e
d
ef
au
lt.
Ho
w
e
v
er
,
s
ev
er
al
m
ac
h
in
e
lear
n
i
n
g
m
et
h
o
d
s
h
a
v
e
b
ee
n
d
ev
elo
p
ed
in
v
ar
io
u
s
l
iter
atu
r
e
w
it
h
v
ar
y
i
n
g
p
er
f
o
r
m
an
ce
.
A
m
aj
o
r
lim
ita
tio
n
to
ac
h
ie
v
in
g
o
p
tim
a
l
p
er
f
o
r
m
an
ce
i
n
th
e
cr
ed
it
ca
r
d
d
ef
au
lt
p
r
ed
ictio
n
is
th
at
th
e
d
ataset
s
ar
e
h
ig
h
l
y
i
m
b
a
lan
ce
d
,
i.e
.
,
th
e
i
n
s
tan
ce
s
w
h
er
e
clien
ts
d
o
n
o
t d
ef
au
lt a
r
e
m
o
r
e
th
a
n
th
e
d
e
f
au
l
t
in
g
ca
s
es.
C
er
tain
s
tu
d
ie
s
h
a
v
e
u
s
ed
t
h
e
d
ef
au
lt
o
f
c
r
e
d
it
c
ar
d
cl
ien
ts
d
a
tase
t
[
2
9
]
an
d
ac
h
ie
v
ed
g
o
o
d
p
e
r
f
o
r
m
an
ce
.
F
o
r
ex
am
p
le,
Pr
u
s
ti
an
d
R
ath
[
3
0
]
u
s
ed
v
a
r
i
o
u
s
alg
o
r
ith
m
s
s
u
ch
as
d
ec
is
i
o
n
tr
ee
,
KNN
,
SVM
,
an
d
m
u
ltil
ay
er
p
e
r
ce
p
t
r
o
n
to
m
ak
e
p
r
ed
ict
io
n
s
o
n
th
e
d
at
a
s
et.
A
d
d
iti
o
n
a
lly
,
th
ey
p
r
o
p
o
s
ed
a
m
eth
o
d
th
a
t
h
y
b
r
i
d
i
ze
d
d
e
cisi
o
n
t
r
ee
,
S
V
M,
an
d
K
NN,
w
h
ich
g
av
e
i
m
p
r
o
v
e
d
p
e
r
f
o
r
m
an
ce
c
o
m
p
a
r
e
d
t
o
th
e
s
tan
d
-
al
o
n
e
alg
o
r
ith
m
s
.
Say
j
a
d
ah
et
a
l
.
[
3
1
]
c
o
n
d
u
c
te
d
a
p
e
r
f
o
r
m
an
ce
e
v
alu
ati
o
n
o
f
cr
ed
it
c
a
r
d
d
ef
au
lt
p
r
e
d
i
cti
o
n
u
s
in
g
lo
g
is
t
ic
r
eg
r
ess
i
o
n
,
r
an
d
o
m
f
o
r
est
,
an
d
d
e
cisi
o
n
t
r
ee
.
T
h
e
ex
p
er
im
en
tal
r
esu
lts
s
h
o
w
ed
th
at
r
an
d
o
m
f
o
r
est
ac
h
iev
e
d
s
u
p
er
io
r
p
er
f
o
r
m
an
ce
w
ith
an
ac
cu
r
ac
y
o
f
ab
o
u
t
8
2
%.
Fu
r
th
e
r
m
o
r
e
,
b
ec
au
s
e
th
e
d
a
t
aset
is
im
b
alan
c
ed
,
a
m
eth
o
d
is
p
r
o
p
o
s
e
d
t
o
t
ac
k
l
e
th
e
p
r
o
b
lem
u
s
in
g
s
y
n
th
etic
m
in
o
r
ity
o
v
er
-
s
am
p
l
in
g
tech
n
iq
u
e
(
SMO
T
E
)
[
3
2
]
.
Usi
n
g
th
e
SMO
T
E
m
eth
o
d
t
o
g
eth
e
r
w
ith
s
ev
en
o
th
e
r
a
lg
o
r
ith
m
s
,
th
e
r
an
d
o
m
f
o
r
est
alg
o
r
i
th
m
ac
h
iev
ed
th
e
b
est
p
e
r
f
o
r
m
an
ce
w
ith
an
ac
cu
r
ac
y
o
f
8
9
.
0
1
%
an
d
F1
-
s
c
o
r
e
o
f
8
9
%
.
L
astl
y
,
Hsu
et
a
l
.
[
3
3
]
an
d
C
h
is
h
ti
an
d
Aw
a
n
[
3
4
]
als
o
p
r
o
p
o
s
e
d
m
o
d
els
t
o
p
r
ed
ict
th
e
d
ef
au
l
tin
g
o
f
c
r
ed
it c
ar
d
cli
en
t
s
an
d
ac
h
iev
ed
c
o
m
p
ar
ab
le
p
e
r
f
o
r
m
an
ce
.
Ho
w
ev
er
,
w
e
ar
e
ai
m
in
g
to
im
p
r
o
v
e
o
n
w
h
at
h
as
b
e
en
d
o
n
e
b
y
ap
p
ly
in
g
o
u
r
p
r
o
p
o
s
e
d
m
eth
o
d
o
n
th
e
s
am
e
d
at
ase
t.
5.
RE
SUL
T
S AN
D
DIS
C
USSI
O
N
I
n
th
is
w
o
r
k
,
th
e
d
ef
au
ltin
g
o
f
th
e
cr
e
d
it
ca
r
d
cli
en
t
d
atas
et
[
2
9
]
is
u
s
ed
.
T
h
e
d
a
tase
t
w
a
s
o
b
tain
e
d
f
r
o
m
th
e
Un
iv
e
r
s
ity
o
f
C
alif
o
r
n
ia
I
r
v
in
e
(
UC
I
)
m
ac
h
in
e
l
ea
r
n
in
g
r
e
p
o
s
it
o
r
y
,
an
d
it
co
n
t
ain
s
3
0
,
0
0
0
in
s
t
an
ce
s
an
d
2
5
att
r
i
b
u
tes
,
in
clu
d
in
g
d
em
o
g
r
a
p
h
i
c
an
d
f
in
an
cia
l
r
ec
o
r
d
s
.
T
h
e
d
a
tase
t
w
as
est
ab
l
is
h
e
d
t
o
p
r
e
d
i
ct
cu
s
to
m
er
s
w
h
o
ar
e
lik
e
ly
to
d
ef
au
l
t
o
n
p
ay
m
en
ts
in
T
aiw
an
.
Ou
t
o
f
th
e
3
0
,
0
0
0
in
s
t
an
c
es
2
3
,
3
6
4
a
r
e
n
o
n
-
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2
0
8
8
-
8708
A
r
tifi
cia
l n
eu
r
a
l n
etw
o
r
k
tech
n
iq
u
e
fo
r
imp
r
o
vin
g
p
r
ed
ictio
n
o
f c
r
ed
it c
a
r
d
…
(
S
a
r
a
h
.
A
.
E
b
ia
r
ed
o
h
-
Mien
ye
)
4397
d
ef
au
l
t
an
d
6
,
6
3
6
a
r
e
d
ef
au
lt
ca
s
es
.
T
h
e
r
at
io
n
ale
b
eh
in
d
th
e
d
a
tas
et
is
f
o
r
f
in
an
cia
l
in
s
tit
u
ti
o
n
s
t
o
b
e
a
b
le
t
o
id
en
t
if
y
p
o
s
s
i
b
le
cu
s
t
o
m
er
s
w
h
o
w
ill
d
ef
au
l
t
o
n
th
ei
r
cr
ed
it
c
ar
d
p
ay
m
en
ts
,
th
er
e
b
y
d
ec
l
in
in
g
s
u
ch
ap
p
li
ca
ti
o
n
s
.
W
e
u
s
e
th
e
7
0
-
3
0
%
t
r
a
in
-
test
s
p
l
it
.
T
h
e
SS
A
E
is
t
r
ain
e
d
w
ith
th
e
tr
ain
in
g
s
et
in
an
u
n
s
u
p
er
v
is
e
d
f
ash
io
n
,
w
h
ile
th
e
test
s
et
is
in
p
u
t
w
ith
th
e
lear
n
ed
SS
A
E
m
o
d
el
t
o
o
b
t
ain
th
e
lo
w
-
d
im
en
s
io
n
al
d
a
ta
.
T
h
e
class
if
i
er
s
a
r
e
th
en
tr
ain
e
d
u
s
in
g
th
e
lo
w
-
d
im
en
s
io
n
al
tr
ain
s
et,
an
d
th
e
p
e
r
f
o
r
m
an
ce
tes
ted
u
s
in
g
th
e
l
o
w
-
d
im
en
s
io
n
al
t
est
s
et
.
T
h
e
n
u
m
b
er
o
f
n
eu
r
o
n
s
in
th
e
f
ir
s
t
an
d
s
ec
o
n
d
h
id
d
en
lay
er
s
w
as
s
et
at
1
0
0
an
d
8
5
,
r
es
p
e
ctiv
e
ly
.
T
o
ef
f
ici
en
tly
ev
alu
at
e
th
e
p
er
f
o
r
m
an
ce
o
f
o
u
r
ap
p
r
o
ac
h
,
w
e
u
tili
ze
p
e
r
f
o
r
m
an
ce
m
etr
i
cs
s
u
ch
as
ac
cu
r
a
cy
,
s
en
s
itiv
ity
,
p
r
e
cisi
o
n
,
an
d
F
1
s
co
r
e
.
A
cc
u
r
ac
y
is
th
e
r
at
io
o
f
th
e
n
u
m
b
er
o
f
c
o
r
r
ec
t
p
r
ed
ict
io
n
s
t
o
th
e
to
t
al
n
u
m
b
er
o
f
p
r
e
d
ic
ti
o
n
s
m
ad
e,
s
en
s
itiv
ity
is
th
e
r
ati
o
o
f
th
e
n
u
m
b
er
o
f
c
o
r
r
e
ct
p
o
s
itiv
e
p
r
e
d
ict
io
n
s
t
o
th
e
to
t
al
ac
tu
al
p
o
s
itiv
es,
p
r
e
cisi
o
n
is
th
e
n
u
m
b
er
o
f
c
o
r
r
e
ct
p
o
s
i
tiv
e
p
r
e
d
ic
ti
o
n
s
d
iv
id
e
d
b
y
th
e
n
u
m
b
er
o
f
p
o
s
it
iv
e
r
esu
lts
p
r
e
d
i
ct
ed
,
an
d
F1
s
co
r
e
is
th
e
h
a
r
m
o
n
ic
m
ea
n
b
etw
ee
n
p
r
e
cisi
o
n
an
d
s
en
s
i
tiv
ity
.
Ma
th
em
atic
ally
,
th
e
p
e
r
f
o
r
m
an
ce
m
etr
i
cs
ca
n
b
e
r
e
p
r
esen
t
ed
as
:
=
+
+
+
+
(
9
)
See
ap
p
en
d
ix
f
o
r
th
e
co
m
p
le
te
d
er
iv
atio
n
o
f
F1
s
co
r
e
:
=
+
(
1
0
)
=
+
(
1
1
)
1
=
2
×
×
+
(
1
2
)
T
h
er
ef
o
r
e,
it c
an
b
e
r
ep
r
esen
te
d
as
(
1
3
)
:
1
=
2
[
(
+
)
(
+
)
(
+
)
+
(
+
)
]
(
1
3
)
Fo
r
s
i
m
p
licit
y
,
F1
s
co
r
e
in
(
1
3
)
ca
n
b
e
d
er
iv
ed
is
b
ein
g
as:
2
[
(
+
)
(
+
)
(
+
)
+
(
+
)
]
=
2
(
)
2
(
+
)
(
+
)
(
+
)
+
(
+
)
[
2
(
)
2
(
+
)
(
+
)
]
÷
[
(
+
)
+
(
+
)
]
Fin
d
i
n
g
t
h
e
lo
w
est co
m
m
o
n
f
a
cto
r
[
2
(
)
2
(
+
)
(
+
)
]
÷
[
(
+
)
+
(
+
)
(
+
)
(
+
)
]
Op
en
th
e
b
r
ac
k
et
[
2
(
)
2
(
+
)
(
+
)
]
÷
[
(
)
2
+
+
(
)
2
+
(
+
)
(
+
)
]
B
r
in
g
li
k
e
ter
m
s
to
g
et
h
er
an
d
f
ac
to
r
ize.
2
(
)
2
(
+
)
(
+
)
÷
2
(
)
2
+
(
+
)
(
+
)
(
+
)
I
n
v
er
t a
n
d
ca
n
ce
l o
u
t li
k
e
ter
m
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
n
t J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
11
,
No
.
5
,
Octo
b
e
r
2
0
2
1
:
4
3
9
2
-
4402
4398
2
(
)
2
(
+
)
(
+
)
×
(
+
)
(
+
)
2
(
)
2
+
(
+
)
2
(
)
2
2
(
)
2
+
(
+
)
Facto
r
ized
TP
o
u
t o
f
th
e
d
en
o
m
i
n
ato
r
2
(
)
2
[
(
2
)
+
(
+
)
]
Fo
r
s
i
m
p
licit
y
w
e
h
a
v
e,
1
=
[
2
2
+
(
+
)
]
(
1
4
)
w
h
er
e
,
,
,
s
tan
d
s
f
o
r
th
e
n
u
m
b
er
o
f
t
r
u
e
p
o
s
itiv
es
,
th
e
n
u
m
b
er
o
f
t
r
u
e
n
eg
a
tiv
es
,
th
e
n
u
m
b
er
o
f
f
alse
p
o
s
it
iv
es,
an
d
th
e
n
u
m
b
er
o
f
f
alse
n
eg
at
iv
es,
r
es
p
e
ctiv
ely
.
Me
an
w
h
ile,
all
th
e
ex
p
er
i
m
en
ts
w
er
e
ca
r
r
i
e
d
o
u
t
u
s
in
g
a
c
o
m
p
u
ter
w
ith
th
e
f
o
ll
o
w
in
g
s
p
ec
if
ica
ti
o
n
s
:
I
n
tel
C
o
r
e
i5
-
6
3
0
0
U,
2
.
4
0
GHz
,
w
it
h
1
6
GB
R
A
M.
An
d
P
y
th
o
n
p
r
o
g
r
a
m
m
in
g
l
an
g
u
a
g
e
w
as
u
s
ed
f
o
r
th
e
c
o
m
p
u
tatio
n
s
.
T
o
s
h
o
w
th
e
e
f
f
ec
tiv
en
ess
o
f
th
e
p
r
o
p
o
s
e
d
a
p
p
r
o
a
ch
,
w
e
co
n
d
u
ct
a
c
o
m
p
ar
ativ
e
s
tu
d
y
w
ith
f
iv
e
b
as
e
cl
ass
if
ie
r
s
.
T
h
er
ef
o
r
e
,
w
e
f
ir
s
t
s
h
o
w
th
e
p
e
r
f
o
r
m
an
ce
o
f
th
ese
cl
ass
if
ie
r
s
o
n
th
e
r
aw
d
atas
et.
T
h
e
cl
ass
if
ier
s
in
clu
d
e
C
A
R
T
,
L
R
,
KN
N,
SV
M,
an
d
L
DA
,
an
d
th
e
r
esu
lts
a
r
e
s
h
o
w
n
in
T
ab
l
e
1
.
T
ab
le
1
.
P
er
f
o
r
m
a
n
ce
o
f
t
h
e
b
ase
class
i
f
ier
s
o
n
th
e
d
atase
t
A
l
g
o
r
i
t
h
m
A
c
c
u
r
a
c
y
(
%)
P
r
e
c
i
si
o
n
(
%)
S
e
n
si
t
i
v
i
t
y
(
%)
F
1
sco
r
e
(
%)
LR
78
62
78
69
C
A
R
T
73
74
73
73
K
N
N
75
71
75
72
S
V
M
36
74
36
35
L
D
A
81
79
81
78
T
a
b
le
2
s
u
m
m
ar
izes
th
e
r
esu
l
ts
o
b
tain
e
d
w
h
en
th
e
class
if
ie
r
s
ar
e
tr
ain
e
d
u
s
in
g
th
e
f
ea
tu
r
es
lea
r
n
e
d
f
r
o
m
th
e
s
t
ac
k
e
d
s
p
ar
s
e
au
to
en
co
d
er
.
I
t
ca
n
b
e
s
ee
n
th
at
th
e
lea
r
n
e
d
f
ea
tu
r
es
s
ig
n
if
ic
an
tly
im
p
r
o
v
e
th
e
p
e
r
f
o
r
m
an
ce
o
f
th
e
class
if
ie
r
s
.
Fu
r
th
e
r
m
o
r
e
,
th
e
r
es
u
lts
s
h
o
w
th
e
a
b
ili
ty
o
f
th
e
p
r
o
p
o
s
e
d
SS
A
E
to
le
a
r
n
a
g
o
o
d
r
e
p
r
esen
t
ati
o
n
o
f
th
e
d
ata
.
T
o
f
u
r
th
er
s
h
o
w
th
e
ef
f
ec
t
iv
en
e
s
s
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
,
t
h
e
b
est
p
e
r
f
o
r
m
in
g
m
o
d
el
f
r
o
m
o
u
r
ex
p
er
im
en
ts
,
w
h
ich
is
th
e
L
DA
,
is
u
s
ed
to
c
o
m
p
ar
e
w
ith
o
th
er
w
ell
-
p
er
f
o
r
m
in
g
m
eth
o
d
s
p
r
o
p
o
s
e
d
in
r
ec
en
t
s
tu
d
ies
th
a
t
h
av
e
b
ee
n
d
is
cu
s
s
ed
in
s
ec
t
i
o
n
4
.
T
o
g
iv
e
a
f
ai
r
co
m
p
a
r
is
o
n
,
w
e
f
o
cu
s
ed
o
n
s
tu
d
ies
th
at
u
s
e
d
s
im
ilar
d
atas
ets.
T
h
is
c
o
m
p
ar
is
o
n
is
s
h
o
w
n
in
T
ab
le
3
,
an
d
it
c
an
b
e
s
e
en
th
at
o
u
r
m
eth
o
d
o
u
t
p
e
r
f
o
r
m
s
th
o
s
e
in
th
e
s
ta
te
d
li
te
r
a
tu
r
e
.
A
ls
o
,
th
e
r
ec
eiv
e
r
o
p
e
r
at
in
g
ch
a
r
ac
te
r
is
ti
c
(
R
OC
)
cu
r
v
e
is
em
p
l
o
y
ed
to
s
h
o
w
th
e
i
m
p
r
o
v
e
d
p
er
f
o
r
m
an
ce
o
f
th
e
SS
A
E
b
ased
L
D
A
co
m
p
ar
e
d
to
th
e
L
DA
th
at
w
as
tr
ain
e
d
w
ith
th
e
r
aw
d
atas
et
.
T
h
e
R
OC
cu
r
v
e
i
s
a
g
r
a
p
h
ica
l
p
l
o
t
w
h
ich
s
h
o
w
s
th
e
p
r
e
d
ict
i
o
n
p
er
f
o
r
m
an
ce
o
f
b
in
a
r
y
class
if
ie
r
s
.
Fro
m
th
e
R
OC
cu
r
v
e
s
h
o
w
n
i
n
Fig
u
r
e
3
,
it
ca
n
b
e
s
e
en
th
a
t
th
e
p
r
o
p
o
s
ed
m
eth
o
d
p
e
r
f
o
r
m
ed
b
e
tte
r
th
an
th
e
co
n
v
en
ti
o
n
a
l L
DA
.
T
ab
le
2
.
I
m
p
ac
t o
f
th
e
f
ea
t
u
r
es
lear
n
ed
b
y
t
h
e
SS
AE
o
n
th
e
b
ase
class
i
f
ier
s
A
l
g
o
r
i
t
h
m
A
c
c
u
r
a
c
y
(
%)
P
r
e
c
i
si
o
n
(
%)
S
e
n
si
t
i
v
i
t
y
(
%)
F
1
sco
r
e
(
%)
LR
90
87
91
89
C
A
R
T
86
84
84
84
K
N
N
89
87
90
88
S
V
M
88
86
88
87
L
D
A
90
91
90
90
Fro
m
th
e
a
b
o
v
e
r
esu
lts
,
w
e
ca
n
s
ee
th
at
o
u
r
p
r
o
p
o
s
ed
a
p
p
r
o
a
ch
a
ch
iev
e
d
b
ett
er
p
er
f
o
r
m
an
ce
co
m
p
a
r
e
d
t
o
th
e
o
th
er
m
eth
o
d
s
.
T
h
e
im
p
r
o
v
ed
p
er
f
o
r
m
an
ce
ca
n
b
e
a
tt
r
i
b
u
te
d
t
o
th
e
p
r
o
p
o
s
ed
SS
A
E
th
at
w
as
ab
l
e
t
o
l
ea
r
n
a
g
o
o
d
r
e
p
r
es
en
t
ati
o
n
o
f
th
e
o
r
ig
in
a
l
in
p
u
t
d
at
a
.
A
ls
o
,
th
e
r
esu
lts
h
av
e
s
h
o
w
n
th
e
ca
p
a
b
i
lity
o
f
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2
0
8
8
-
8708
A
r
tifi
cia
l n
eu
r
a
l n
etw
o
r
k
tech
n
iq
u
e
fo
r
imp
r
o
vin
g
p
r
ed
ictio
n
o
f c
r
ed
it c
a
r
d
…
(
S
a
r
a
h
.
A
.
E
b
ia
r
ed
o
h
-
Mien
ye
)
4399
d
e
ep
le
ar
n
in
g
in
ac
h
iev
in
g
ex
c
ep
t
io
n
al
p
e
r
f
o
r
m
an
ce
in
d
if
f
e
r
en
t
task
s
,
in
clu
d
in
g
f
ea
tu
r
e
r
e
p
r
es
en
tati
o
n
.
L
astl
y
,
th
is
s
tu
d
y
h
as d
em
o
n
s
tr
at
e
d
th
e
im
p
o
r
tan
ce
o
f
t
r
ain
in
g
m
ac
h
i
n
e
le
ar
n
in
g
a
lg
o
r
ith
m
s
w
ith
s
u
ita
b
le
d
ata
,
an
d
th
at
im
p
r
o
v
e
d
p
e
r
f
o
r
m
an
ce
c
an
b
e
o
b
ta
in
e
d
n
o
t
o
n
ly
b
y
h
y
p
er
-
p
ar
am
eter
tu
n
in
g
b
u
t
als
o
an
d
m
o
r
e
ef
f
ici
en
tly
b
y
ef
f
ec
tiv
e
f
ea
tu
r
e
l
ea
r
n
in
g
.
T
ab
le
3
.
C
o
m
p
ar
is
o
n
w
i
th
o
t
h
er
m
e
th
o
d
s
L
i
t
e
r
a
t
u
r
e
A
c
c
u
r
a
c
y
(
%)
P
r
e
c
i
si
o
n
(
%)
S
e
n
si
t
i
v
i
t
y
(
%)
F
1
sco
r
e
(
%)
P
r
u
st
i
a
n
d
R
a
t
h
[
2
6
]
8
2
.
5
8
9
6
.
8
3
8
3
.
5
7
8
9
.
7
1
S
a
y
j
a
d
a
h
e
t
a
l
.
[
2
7
]
8
1
.
8
1
-
-
-
S
u
b
a
si
a
n
d
C
a
n
k
u
r
t
[
3
2
]
8
9
.
0
1
-
-
89
H
su
e
t
a
l
.
[
3
3
]
8
0
.
2
-
-
-
C
h
i
sh
t
i
a
n
d
A
w
a
n
[
3
4
]
82
84
96
89
P
r
o
p
o
se
d
S
S
A
E+
L
D
A
90
91
90
90
Fig
u
r
e
3
.
R
OC
C
u
r
v
e
s
h
o
w
in
g
i
m
p
r
o
v
ed
p
er
f
o
r
m
a
n
ce
6.
CO
NCLU
SI
O
N
C
o
n
v
en
t
io
n
al
m
ac
h
i
n
e
lear
n
i
n
g
al
g
o
r
ith
m
s
ar
e
o
f
te
n
i
n
e
f
f
ec
t
iv
e
i
n
p
er
f
o
r
m
i
n
g
cla
s
s
i
f
icatio
n
o
n
lar
g
e
d
atasets
s
u
c
h
as
m
o
s
t
cr
ed
it
c
ar
d
d
atasets
.
Hen
ce
,
i
n
t
h
is
p
a
p
er
,
a
s
tack
ed
s
p
ar
s
e
au
to
e
n
c
o
d
er
is
p
r
o
p
o
s
ed
to
ac
h
iev
e
o
p
t
i
m
al
f
ea
tu
r
e
lear
n
i
n
g
.
I
n
t
h
e
p
r
o
p
o
s
ed
au
to
en
co
d
er
n
et
w
o
r
k
,
w
e
i
n
tr
o
d
u
ce
d
a
b
atch
n
o
r
m
aliza
t
io
n
tech
n
iq
u
e
to
en
h
a
n
ce
th
e
p
er
f
o
r
m
a
n
ce
,
s
p
ee
d
,
an
d
s
tab
ilit
y
o
f
th
e
m
o
d
el
an
d
to
p
r
ev
en
t
o
v
er
f
itti
n
g
f
u
r
th
er
.
A
l
s
o
,
th
e
m
o
d
el
w
as
o
p
ti
m
ized
u
s
in
g
t
h
e
A
d
aM
a
x
alg
o
r
it
h
m
.
T
h
e
lear
n
ed
d
ata
w
as
t
h
e
n
u
s
ed
to
tr
ain
f
iv
e
s
h
allo
w
m
ac
h
i
n
e
lear
n
in
g
al
g
o
r
ith
m
s
,
an
d
t
h
e
p
er
f
o
r
m
a
n
ce
test
ed
.
W
h
e
n
co
m
p
ar
ed
w
it
h
a
ca
s
e
w
h
er
e
th
e
alg
o
r
ith
m
s
w
er
e
tr
ai
n
ed
w
it
h
t
h
e
r
a
w
d
ata,
o
u
r
p
r
o
p
o
s
ed
m
et
h
o
d
s
h
o
w
ed
s
u
p
er
io
r
p
er
f
o
r
m
an
ce
.
F
u
r
th
er
m
o
r
e,
th
e
r
esu
l
ts
w
er
e
co
m
p
ar
ed
w
ith
m
e
th
o
d
s
in
s
o
m
e
liter
at
u
r
e
th
at
u
s
ed
a
s
i
m
ilar
d
ataset,
an
d
th
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
also
s
h
o
w
ed
s
i
g
n
i
f
ic
an
t
i
m
p
r
o
v
e
m
en
t.
F
u
t
u
r
e
r
ese
ar
ch
w
ill
f
o
cu
s
o
n
s
t
u
d
y
i
n
g
t
h
e
ef
f
ec
t
o
f
d
i
f
f
er
e
n
t
o
p
tim
izer
s
an
d
s
tac
k
i
n
g
d
i
v
e
r
s
e
au
to
en
co
d
e
r
s
an
d
o
b
s
er
v
i
n
g
t
h
e
r
esu
ltan
t
i
m
p
ac
t
o
n
t
h
e
f
ea
t
u
r
e
lear
n
i
n
g
p
r
o
ce
s
s
.
A
ls
o
,
f
u
t
u
r
e
r
esear
ch
w
ill
co
n
s
id
er
co
m
p
ar
in
g
th
e
f
ea
tu
r
e
lea
n
i
n
g
ca
p
ab
ilit
y
o
f
th
e
s
tac
k
ed
s
p
ar
s
e
au
to
en
co
d
er
w
i
th
o
t
h
er
f
ea
t
u
r
e
lear
n
in
g
an
d
f
ea
tu
r
e
e
n
g
in
ee
r
i
n
g
m
et
h
o
d
s
.
ACK
NO
WL
E
D
G
E
M
E
NT
S
T
h
is
w
o
r
k
is
s
u
p
p
o
r
ted
p
ar
tiall
y
b
y
t
h
e
C
en
ter
o
f
T
elec
o
m
m
u
n
ica
tio
n
s
,
Un
i
v
er
s
i
t
y
o
f
J
o
h
an
n
e
s
b
u
r
g
,
So
u
t
h
A
f
r
ica.
T
h
is
r
e
s
ea
r
ch
r
ec
eiv
ed
n
o
e
x
ter
n
al
f
u
n
d
i
n
g
.
T
h
e
A
P
C
w
ill
b
e
p
aid
f
r
o
m
o
u
r
r
esear
ch
ce
n
ter
f
u
n
d
s
.
RE
F
E
R
E
NC
E
S
[1
]
A
.
S
.
S
h
a
m
sa
b
a
d
i,
M
.
Ba
b
a
ie
-
Z
a
d
e
h
,
S
.
Z.
S
e
y
y
e
d
sa
l
e
h
i,
H.
R
.
Ra
b
iee
,
a
n
d
C.
Ju
tt
e
n
,
“
A
n
e
w
a
lg
o
rit
h
m
f
o
r
train
in
g
sp
a
rse
a
u
to
e
n
c
o
d
e
rs,”
in
2
0
1
7
2
5
t
h
Eu
r
o
p
e
a
n
S
i
g
n
a
l
Pro
c
e
ss
in
g
Co
n
fer
e
n
c
e
(
EUS
IP
CO
)
,
A
u
g
.
2
0
1
7
,
p
p
.
2
1
4
1
-
2
1
4
5
,
d
o
i:
1
0
.
2
3
9
1
9
/
EUS
I
P
CO.2
0
1
7
.
8
0
8
1
5
8
8
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
n
t J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
11
,
No
.
5
,
Octo
b
e
r
2
0
2
1
:
4
3
9
2
-
4402
4400
[2
]
B.
Ya
n
a
n
d
G
.
Ha
n
,
“
Eff
e
c
ti
v
e
F
e
a
tu
re
Ex
trac
ti
o
n
v
ia
S
tac
k
e
d
S
p
a
rse
A
u
to
e
n
c
o
d
e
r
t
o
Im
p
ro
v
e
In
tru
sio
n
De
tec
ti
o
n
S
y
st
e
m
,
”
IEE
E
Acc
e
ss
,
v
o
l.
6
,
p
p
.
4
1
2
3
8
-
4
1
2
4
8
,
2
0
1
8
,
d
o
i
:
1
0
.
1
1
0
9
/A
CCES
S
.
2
0
1
8
.
2
8
5
8
2
7
7
.
[3
]
G
.
E.
Hin
t
o
n
,
A
.
Kriz
h
e
v
sk
y
,
a
n
d
S
.
D.
W
a
n
g
,
“
T
ra
n
s
f
o
r
m
in
g
a
u
to
-
e
n
c
o
d
e
rs,”
in
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
2
1
th
in
ter
n
a
t
io
n
a
l
c
o
n
fer
e
n
c
e
o
n
Arti
fi
c
ia
l
n
e
u
r
a
l
n
e
two
rk
s
,
Esp
o
o
,
F
in
lan
d
,
J
u
n
.
2
0
1
1
,
p
p
.
4
4
-
5
1
,
d
o
i:
d
o
i.
o
rg
/
1
0
.
1
0
0
7
/9
7
8
-
3
-
6
4
2
-
2
1
7
3
5
-
7
_
6
.
[4
]
S
.
Na
re
jo
,
E
.
P
a
se
ro
,
a
n
d
F
.
K
u
lso
o
m
,
“
EE
G
Ba
se
d
Ey
e
S
tate
Clas
sif
ica
ti
o
n
u
sin
g
De
e
p
Be
li
e
f
Ne
t
w
o
rk
a
n
d
S
tac
k
e
d
A
u
to
En
c
o
d
e
r,
”
I
n
ter
n
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
El
e
c
trica
l
a
n
d
Co
mp
u
ter
E
n
g
in
e
e
rin
g
(
IJ
ECE
)
,
v
o
l.
6
,
n
o
.
6
,
p
p
.
3
1
3
1
-
3
1
4
1
,
De
c
e
m
b
e
r
2
0
1
6
,
d
o
i:
1
0
.
1
1
5
9
1
/
ij
e
c
e
.
v
6
i6
.
p
p
3
1
3
1
-
3
1
4
1
.
[5
]
G
.
Do
n
g
,
G
.
L
i
a
o
,
H.
L
iu
,
a
n
d
G
.
Ku
a
n
g
,
“
A
Re
v
ie
w
o
f
th
e
A
u
to
e
n
c
o
d
e
r
a
n
d
Its
V
a
rian
ts:
A
Co
m
p
a
ra
ti
v
e
P
e
rsp
e
c
ti
v
e
f
ro
m
T
a
rg
e
t
Re
c
o
g
n
it
io
n
in
S
y
n
th
e
ti
c
-
A
p
e
rtu
re
Ra
d
a
r
Im
a
g
e
s,”
IEE
E
Ge
o
sc
ien
c
e
a
n
d
Rem
o
te
S
e
n
si
n
g
M
a
g
a
zin
e
,
v
o
l.
6
,
n
o
.
3
,
p
p
.
4
4
-
6
8
,
S
e
p
.
2
0
1
8
,
d
o
i:
1
0
.
1
1
0
9
/M
G
RS
.
2
0
1
8
.
2
8
5
3
5
5
5
.
[6
]
“
Cre
d
it
Ca
rd
F
ra
u
d
De
tec
ti
o
n
:
A
Re
a
li
stic M
o
d
e
li
n
g
a
n
d
a
No
v
e
l
Lea
rn
in
g
S
trate
g
y
,
”
IEE
E
T
ra
n
sa
c
ti
o
n
s o
n
Ne
u
ra
l
Ne
two
rk
s a
n
d
L
e
a
rn
in
g
S
y
ste
ms
,
v
o
l.
2
9
,
n
o
.
8
,
p
p
.
3
7
8
4
-
3
7
9
7
,
A
u
g
.
2
0
1
8
,
d
o
i:
1
0
.
1
1
0
9
/T
NN
L
S
.
2
0
1
7
.
2
7
3
6
6
4
3
.
[7
]
C.
Jia
n
g
,
J.
S
o
n
g
,
G
.
L
iu
,
L
.
Zh
e
n
g
,
a
n
d
W
.
L
u
a
n
,
“
Cre
d
it
Ca
rd
F
ra
u
d
De
tec
ti
o
n
:
A
No
v
e
l
Ap
p
r
o
a
c
h
Us
in
g
Ag
g
re
g
a
ti
o
n
S
trate
g
y
a
n
d
F
e
e
d
b
a
c
k
M
e
c
h
a
n
is
m
,
”
IEE
E
In
ter
n
e
t
o
f
T
h
in
g
s
J
o
u
rn
a
l
,
v
o
l.
5
,
n
o
.
5
,
p
p
.
3
6
3
7
-
3
6
4
7
,
Oc
t.
2
0
1
8
,
d
o
i:
1
0
.
1
1
0
9
/JIO
T
.
2
0
1
8
.
2
8
1
6
0
0
7
.
[8
]
A
.
A
.
T
a
h
a
a
n
d
S
.
J.
M
a
leb
a
ry
,
"
A
n
In
telli
g
e
n
t
A
p
p
ro
a
c
h
to
Cre
d
it
Ca
rd
F
ra
u
d
De
tec
ti
o
n
Us
i
n
g
a
n
Op
ti
m
ize
d
L
ig
h
t
G
ra
d
ien
t
Bo
o
stin
g
M
a
c
h
in
e
,
"
in
IEE
E
Acc
e
ss
,
v
o
l.
8
,
p
p
.
2
5
5
7
9
-
2
5
5
8
7
,
2
0
2
0
,
d
o
i:
1
0
.
1
1
0
9
/A
CCES
S
.
2
0
2
0
.
2
9
7
1
3
5
4
.
[9
]
S
.
Ka
m
le
y
,
S
.
Ja
lo
re
e
,
a
n
d
R.
S
.
T
h
a
k
u
r,
“
P
e
rf
o
rm
a
n
c
e
F
o
re
c
a
stin
g
o
f
S
h
a
re
M
a
rk
e
t
u
sin
g
M
a
c
h
in
e
L
e
a
rn
in
g
T
e
c
h
n
iq
u
e
s:
A
Re
v
ie
w
,
”
In
ter
n
a
ti
o
n
a
l
J
o
u
rn
a
l
o
f
El
e
c
trica
l
a
n
d
Co
mp
u
ter
En
g
i
n
e
e
rin
g
(
IJ
ECE
)
,
v
o
l.
6
,
n
o
.
6
,
p
p
.
3
1
9
6
-
3
2
0
4
,
De
c
.
2
0
1
6
,
d
o
i
:
1
0
.
1
1
5
9
1
/
ij
e
c
e
.
v
6
i6
.
p
p
3
1
9
6
-
32
0
4
.
[1
0
]
D.
L
iu
,
Z.
Wan
g
,
B.
W
e
n
,
J.
Y
a
n
g
,
W
.
H
a
n
,
a
n
d
T
.
S
.
Hu
a
n
g
,
“
Ro
b
u
st
S
i
n
g
le
Im
a
g
e
S
u
p
e
r
-
Re
so
lu
ti
o
n
v
ia
De
e
p
Ne
tw
o
rk
s
W
it
h
S
p
a
rse
P
r
io
r,
”
IE
EE
T
ra
n
s
a
c
ti
o
n
s
o
n
Im
a
g
e
Pro
c
e
ss
in
g
,
v
o
l.
2
5
,
n
o
.
7
,
p
p
.
3
1
9
4
-
3
2
0
7
,
Ju
l.
2
0
1
6
,
d
o
i:
1
0
.
1
1
0
9
/T
IP
.
2
0
1
6
.
2
5
6
4
6
4
3
.
[1
1
]
M
.
G
o
n
g
,
J.
L
iu
,
H.
L
i,
Q.
Ca
i,
a
n
d
L
.
S
u
,
“
A
M
u
lt
io
b
jec
ti
v
e
S
p
a
rse
F
e
a
tu
re
L
e
a
rn
in
g
M
o
d
e
l
f
o
r
De
e
p
Ne
u
ra
l
Ne
tw
o
rk
s,”
IEE
E
T
ra
n
sa
c
t
io
n
s
o
n
Ne
u
ra
l
Ne
tw
o
rk
s
a
n
d
L
e
a
rn
in
g
S
y
ste
ms
,
v
o
l.
2
6
,
n
o
.
1
2
,
p
p
.
3
2
6
3
-
3
2
7
7
,
De
c
.
2
0
1
5
,
d
o
i:
1
0
.
1
1
0
9
/T
NN
L
S
.
2
0
1
5
.
2
4
6
9
6
7
3
.
[1
2
]
J.
A
.
El
li
s
a
n
d
S
.
Ra
jam
a
n
ick
a
m
,
“
S
c
a
lab
le
In
f
e
re
n
c
e
f
o
r
S
p
a
rse
De
e
p
Ne
u
ra
l
Ne
t
w
o
rk
s
u
sin
g
Ko
k
k
o
s
Ke
rn
e
ls,”
in
2
0
1
9
IEE
E
Hi
g
h
Per
fo
rm
a
n
c
e
Extre
me
Co
mp
u
ti
n
g
Co
n
fer
e
n
c
e
(
HPEC)
,
S
e
p
.
2
0
1
9
,
p
p
.
1
-
7
,
d
o
i:
1
0
.
1
1
0
9
/H
P
EC.
2
0
1
9
.
8
9
1
6
3
7
8
.
[1
3
]
D.
P
.
Ki
n
g
m
a
a
n
d
J.
Ba
,
“
A
d
a
m
:
A
M
e
th
o
d
f
o
r
S
t
o
c
h
a
stic Op
ti
m
iza
ti
o
n
,
”
a
rXiv:1
4
1
2
.
6
9
8
0
[
c
s
]
,
Ja
n
.
2
0
1
7
.
[1
4
]
S
.
Io
f
f
e
a
n
d
C.
S
z
e
g
e
d
y
,
“
Ba
tc
h
No
rm
a
li
z
a
ti
o
n
:
A
c
c
e
lera
ti
n
g
De
e
p
Ne
t
w
o
rk
T
ra
in
in
g
b
y
Re
d
u
c
in
g
In
ter
n
a
l
Co
v
a
riate
S
h
if
t,
”
a
rXiv:1
5
0
2
.
0
3
1
6
7
,
M
a
r.
2
0
1
5
.
[1
5
]
M
.
S
u
n
,
H
.
W
a
n
g
,
P
.
L
iu
,
S
.
Hu
a
n
g
,
a
n
d
P
.
F
a
n
,
“
A
sp
a
rse
sta
c
k
e
d
d
e
n
o
isi
n
g
a
u
to
e
n
c
o
d
e
r
w
it
h
o
p
t
im
iz
e
d
tran
sf
e
r
lea
rn
in
g
a
p
p
li
e
d
t
o
th
e
f
a
u
lt
d
ia
g
n
o
sis
o
f
ro
ll
in
g
b
e
a
rin
g
s,”
M
e
a
su
re
me
n
t
,
v
o
l.
1
4
6
,
p
p
.
3
0
5
-
3
1
4
,
No
v
.
2
0
1
9
,
d
o
i:
1
0
.
1
0
1
6
/j
.
m
e
a
su
re
m
e
n
t.
2
0
1
9
.
0
6
.
0
2
9
.
[1
6
]
H.
Zh
u
,
J.
Ch
e
n
g
,
C.
Zh
a
n
g
,
J.
W
u
,
a
n
d
X
.
S
h
a
o
,
“
S
tac
k
e
d
p
ru
n
i
n
g
sp
a
rse
d
e
n
o
isin
g
a
u
to
e
n
c
o
d
e
r
b
a
se
d
in
telli
g
e
n
t
f
a
u
lt
d
iag
n
o
sis
o
f
ro
ll
in
g
b
e
a
rin
g
s,”
Ap
p
li
e
d
S
o
ft
Co
mp
u
ti
n
g
,
v
o
l.
8
8
,
M
a
r.
2
0
2
0
,
A
rt.
No
.
1
0
6
0
6
0
,
d
o
i:
1
0
.
1
0
1
6
/j
.
a
so
c
.
2
0
1
9
.
1
0
6
0
6
0
.
[1
7
]
A
.
S
a
n
k
a
ra
n
,
M
.
V
a
tsa
,
R.
S
i
n
g
h
,
a
n
d
A
.
M
a
ju
m
d
a
r,
“
G
ro
u
p
sp
a
rse
a
u
to
e
n
c
o
d
e
r,
”
Ima
g
e
a
n
d
V
isio
n
C
o
mp
u
t
in
g
,
v
o
l.
6
0
,
p
p
.
6
4
-
7
4
,
A
p
r.
2
0
1
7
,
d
o
i
:
1
0
.
1
0
1
6
/j
.
im
a
v
is.2
0
1
7
.
0
1
.
0
0
5
.
[1
8
]
L
.
Ch
e
n
,
M
.
Zh
o
u
,
W
.
S
u
,
M
.
W
u
,
J.
S
h
e
,
a
n
d
K.
Hiro
ta,
“
S
o
ftm
a
x
r
e
g
re
ss
io
n
b
a
se
d
d
e
e
p
sp
a
rse
a
u
to
e
n
c
o
d
e
r
n
e
tw
o
rk
f
o
r
f
a
c
ial
e
m
o
ti
o
n
re
c
o
g
n
it
io
n
i
n
h
u
m
a
n
-
ro
b
o
t
in
tera
c
ti
o
n
,
”
In
fo
rm
a
ti
o
n
S
c
ien
c
e
s
,
v
o
l.
4
2
8
,
p
p
.
4
9
-
6
1
,
F
e
b
.
2
0
1
8
,
d
o
i:
1
0
.
1
0
1
6
/
j.
i
n
s.2
0
1
7
.
1
0
.
0
4
4
.
[1
9
]
S
.
Ya
n
g
,
Y.
Zh
a
n
g
,
Y.
Zh
u
,
P
.
L
i,
a
n
d
X
.
Hu
,
“
Re
p
re
se
n
tatio
n
lea
rn
in
g
v
ia
s
e
rial
a
u
to
e
n
c
o
d
e
rs
f
o
r
d
o
m
a
in
a
d
a
p
tatio
n
,
”
Ne
u
ro
c
o
m
p
u
t
in
g
,
v
o
l
.
3
5
1
,
p
p
.
1
-
9
,
J
u
l.
2
0
1
9
,
d
o
i:
1
0
.
1
0
1
6
/j
.
n
e
u
c
o
m
.
2
0
1
9
.
0
3
.
0
5
6
.
[2
0
]
R.
A
l
-
H
m
o
u
z
,
W
.
P
e
d
ry
c
z
,
A
.
Ba
lam
a
sh
,
a
n
d
A
.
M
o
rf
e
q
,
“
L
o
g
ic
-
d
riv
e
n
a
u
to
e
n
c
o
d
e
rs,”
Kn
o
wle
d
g
e
-
Ba
se
d
S
y
ste
ms
,
v
o
l.
1
8
3
,
No
v
.
2
0
1
9
,
A
rt.
No
.
1
0
4
8
7
4
,
d
o
i:
1
0
.
1
0
1
6
/j
.
k
n
o
sy
s.2
0
1
9
.
1
0
4
8
7
4
.
[2
1
]
S
.
F
e
n
g
,
H.
Yu
,
a
n
d
M
.
F
.
Du
a
rte,
“
A
u
to
e
n
c
o
d
e
r
b
a
se
d
sa
m
p
le
se
lec
ti
o
n
f
o
r
se
l
f
-
tau
g
h
t
lea
rn
in
g
,
”
Kn
o
wled
g
e
-
Ba
se
d
S
y
ste
ms
,
v
o
l.
1
9
2
,
M
a
r.
2
0
2
0
,
A
rt.
No
.
1
0
5
3
4
3
,
d
o
i
:
1
0
.
1
0
1
6
/j
.
k
n
o
sy
s.2
0
1
9
.
1
0
5
3
4
3
.
[2
2
]
B.
X
u
,
H.
L
in
,
Y.
L
in
,
a
n
d
K.
X
u
,
“
In
c
o
r
p
o
ra
ti
n
g
q
u
e
ry
c
o
n
stra
in
ts
f
o
r
a
u
to
e
n
c
o
d
e
r
e
n
h
a
n
c
e
d
ra
n
k
in
g
,
”
Ne
u
ro
c
o
mp
u
ti
n
g
,
v
o
l
.
3
5
6
,
p
p
.
1
4
2
-
1
5
0
,
S
e
p
.
2
0
1
9
,
d
o
i:
1
0
.
1
0
1
6
/j
.
n
e
u
c
o
m
.
2
0
1
9
.
0
3
.
0
6
8
.
[2
3
]
O.
İrso
y
a
n
d
E.
A
lp
a
y
d
ın
,
“
Un
su
p
e
rv
ise
d
f
e
a
tu
re
e
x
trac
ti
o
n
w
it
h
a
u
to
e
n
c
o
d
e
r
tree
s,”
Ne
u
r
o
c
o
mp
u
ti
n
g
,
v
o
l.
2
5
8
,
p
p
.
6
3
-
7
3
,
Oc
t.
2
0
1
7
,
d
o
i:
1
0
.
1
0
1
6
/j
.
n
e
u
c
o
m
.
2
0
1
7
.
0
2
.
0
7
5
.
[2
4
]
I.
D.
M
ien
y
e
,
Y.
S
u
n
,
a
n
d
Z.
W
a
n
g
,
“
Im
p
ro
v
e
d
sp
a
rse
a
u
to
e
n
c
o
d
e
r
b
a
se
d
a
rti
f
icia
l
n
e
u
ra
l
n
e
tw
o
rk
a
p
p
ro
a
c
h
f
o
r
p
re
d
ictio
n
o
f
h
e
a
rt
d
ise
a
se
,
”
In
fo
rm
a
t
ics
in
M
e
d
icin
e
Un
l
o
c
k
e
d
,
v
o
l.
1
8
,
Ja
n
.
2
0
2
0
,
A
rt.
No
.
1
0
5
3
4
3
,
d
o
i:
1
0
.
1
0
1
6
/j
.
im
u
.
2
0
2
0
.
1
0
0
3
0
7
.
[2
5
]
J.
X
u
e
t
a
l
.
,
"
S
tac
k
e
d
S
p
a
rse
A
u
to
e
n
c
o
d
e
r
(S
S
A
E)
f
o
r
Nu
c
lei
De
t
e
c
ti
o
n
o
n
Bre
a
st
Ca
n
c
e
r
Hi
sto
p
a
th
o
lo
g
y
I
m
a
g
e
s,"
in
IEE
E
T
ra
n
sa
c
ti
o
n
s
o
n
M
e
d
ic
a
l
Ima
g
i
n
g
,
v
o
l.
3
5
,
n
o
.
1
,
p
p
.
1
1
9
-
1
3
0
,
Ja
n
.
2
0
1
6
,
d
o
i:
1
0
.
1
1
0
9
/T
M
I.
2
0
1
5
.
2
4
5
8
7
0
2
.
[2
6
]
G
.
E.
Hin
to
n
a
n
d
R.
R
.
S
a
lak
h
u
t
d
in
o
v
,
“
Re
d
u
c
i
n
g
th
e
Dim
e
n
sio
n
a
li
ty
o
f
Da
ta
w
it
h
Ne
u
ra
l
Ne
t
w
o
rk
s
,
”
S
c
ien
c
e
,
v
o
l.
3
1
3
,
n
o
.
5
7
8
6
,
p
p
.
5
0
4
-
5
0
7
,
Ju
l
.
2
0
0
6
,
d
o
i:
1
0
.
1
1
2
6
/sc
ien
c
e
.
1
1
2
7
6
4
7
.
[2
7
]
C.
S
h
i,
B.
L
u
o
,
S
.
He
,
K.
L
i,
H.
L
iu
,
a
n
d
B.
L
i,
“
T
o
o
l
W
e
a
r
P
re
d
ictio
n
v
ia
M
u
lt
id
im
e
n
sio
n
a
l
S
tac
k
e
d
S
p
a
rse
A
u
to
e
n
c
o
d
e
rs
W
it
h
F
e
a
tu
re
F
u
si
o
n
,
”
IEE
E
T
ra
n
s
a
c
ti
o
n
s
o
n
I
n
d
u
s
tria
l
In
f
o
rm
a
ti
c
s
,
v
o
l.
1
6
,
n
o
.
8
,
p
p
.
5
1
5
0
-
5
1
5
9
,
A
u
g
.
2
0
2
0
,
d
o
i:
1
0
.
1
1
0
9
/T
II.
2
0
1
9
.
2
9
4
9
3
5
5
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2
0
8
8
-
8708
A
r
tifi
cia
l n
eu
r
a
l n
etw
o
r
k
tech
n
iq
u
e
fo
r
imp
r
o
vin
g
p
r
ed
ictio
n
o
f c
r
ed
it c
a
r
d
…
(
S
a
r
a
h
.
A
.
E
b
ia
r
ed
o
h
-
Mien
ye
)
4401
[2
8
]
Y.
Be
n
g
io
,
P
.
L
a
m
b
li
n
,
D.
P
o
p
o
v
ic
i,
a
n
d
H.
L
a
ro
c
h
e
ll
e
,
“
G
re
e
d
y
L
a
y
e
r
-
W
is
e
T
ra
in
in
g
o
f
De
e
p
Ne
tw
o
rk
s,”
in
Ad
v
a
n
c
e
s i
n
Ne
u
ra
l
In
f
o
rm
a
ti
o
n
Pro
c
e
ss
in
g
S
y
ste
ms
,
v
o
l.
1
9
,
p
p
.
1
5
3
-
1
6
0
,
2
0
0
7
.
[2
9
]
“
UCI
M
a
c
h
in
e
L
e
a
rn
in
g
Re
p
o
sit
o
ry
:
d
e
fa
u
lt
o
f
c
re
d
it
c
a
rd
c
li
e
n
ts
Da
ta
S
e
t.
”
[
On
li
n
e
]
,
A
v
a
ib
le:
h
tt
p
s:/
/ar
c
h
iv
e
.
ics
.
u
c
i.
e
d
u
/m
l/
d
a
tas
e
ts
/d
e
f
a
u
lt
+
o
f+
c
re
d
it
+
c
a
rd
+
c
li
e
n
ts.
[3
0
]
D.
P
r
u
sti
a
n
d
S
.
K
.
Ra
th
,
"
W
e
b
se
rv
ice
b
a
se
d
c
re
d
it
c
a
rd
f
ra
u
d
d
e
t
e
c
ti
o
n
b
y
a
p
p
ly
in
g
m
a
c
h
in
e
lea
rn
in
g
tec
h
n
iq
u
e
s,"
T
ENCON
2
0
1
9
-
2
0
1
9
IEE
E
Reg
io
n
1
0
C
o
n
fer
e
n
c
e
(
T
ENCON)
,
Ko
c
h
i,
In
d
ia,
2
0
1
9
,
p
p
.
4
9
2
-
4
9
7
,
d
o
i:
1
0
.
1
1
0
9
/T
ENCON
.
2
0
1
9
.
8
9
2
9
3
7
2
.
[3
1
]
Y.
S
a
y
j
a
d
a
h
,
I.
A
.
T
.
Ha
sh
e
m
,
F
.
A
lo
taib
i,
a
n
d
K.
A
.
Ka
s
m
iran
,
“
Cre
d
it
Ca
rd
De
f
a
u
lt
P
re
d
icti
o
n
u
sin
g
M
a
c
h
in
e
L
e
a
rn
in
g
Tec
h
n
iq
u
e
s,”
in
2
0
1
8
Fo
u
rt
h
In
ter
n
a
ti
o
n
a
l
C
o
n
fer
e
n
c
e
o
n
Ad
v
a
n
c
e
s
in
C
o
mp
u
t
in
g
,
Co
mm
u
n
ic
a
ti
o
n
Au
t
o
m
a
ti
o
n
(
ICACCA
)
,
Oc
t.
2
0
1
8
,
p
p
.
1
-
4
,
d
o
i:
1
0
.
1
1
0
9
/ICA
CCA
F
.
2
0
1
8
.
8
7
7
6
8
0
2
.
[3
2
]
A
.
S
u
b
a
si
a
n
d
S
.
Ca
n
k
u
rt,
“
P
re
d
i
c
ti
o
n
o
f
d
e
f
a
u
lt
p
a
y
m
e
n
t
o
f
c
r
e
d
it
c
a
rd
c
li
e
n
ts
u
sin
g
Da
ta
M
in
i
n
g
T
e
c
h
n
iq
u
e
s,”
in
2
0
1
9
I
n
ter
n
a
ti
o
n
a
l
En
g
in
e
e
rin
g
C
o
n
fer
e
n
c
e
(
IEC)
,
Ju
n
.
2
0
1
9
,
p
p
.
1
1
5
-
1
2
0
,
d
o
i:
1
0
.
1
1
0
9
/IE
C
4
7
8
4
4
.
2
0
1
9
.
8
9
5
0
5
9
7
.
[3
3
]
T
.
C.
Hs
u
,
S
.
T
.
L
io
u
,
Y.
-
P
.
W
a
n
g
,
Y.
-
S
.
Hu
a
n
g
,
a
n
d
C
h
e
-
L
in
,
“
En
h
a
n
c
e
d
Re
c
u
rre
n
t
Ne
u
r
a
l
Ne
tw
o
rk
f
o
r
Co
m
b
in
in
g
S
tatic
a
n
d
Dy
n
a
m
i
c
F
e
a
tu
re
s
f
o
r
Cre
d
it
Ca
rd
De
f
a
u
lt
P
re
d
icti
o
n
,
”
in
ICA
S
S
P
2
0
1
9
-
2
0
1
9
IEE
E
In
ter
n
a
t
io
n
a
l
Co
n
fer
e
n
c
e
o
n
Ac
o
u
stics
,
S
p
e
e
c
h
a
n
d
S
i
g
n
a
l
Pro
c
e
ss
in
g
(
ICAS
S
P)
,
M
a
y
2
0
1
9
,
p
p
.
1
5
7
2
-
1
5
7
6
,
d
o
i:
1
0
.
1
1
0
9
/ICA
S
S
P
.
2
0
1
9
.
8
6
8
2
2
1
2
.
[3
4
]
W
.
A
.
Ch
ish
ti
a
n
d
S
.
M
.
Aw
a
n
,
“
De
e
p
Ne
u
ra
l
N
e
t
w
o
rk
a
S
tep
b
y
S
tep
A
p
p
ro
a
c
h
to
Clas
sify
Cre
d
i
t
Ca
rd
De
f
a
u
lt
Cu
sto
m
e
r,
”
in
2
0
1
9
I
n
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
I
n
n
o
v
a
ti
v
e
Co
mp
u
ti
n
g
(
ICIC)
,
No
v
.
2
0
1
9
,
p
p
.
1
-
8
,
d
o
i:
1
0
.
1
1
0
9
/ICIC4
8
4
9
6
.
2
0
1
9
.
8
9
6
6
7
2
3
.
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
S
.
A.
Eb
ia
r
e
d
o
h
-
M
ien
y
e
re
c
e
iv
e
d
h
e
r
BS
d
e
g
re
e
in
S
o
f
twa
re
E
n
g
in
e
e
rin
g
in
2
0
1
2
f
ro
m
th
e
Un
iv
e
rsit
y
o
f
Eas
t
L
o
n
d
o
n
,
UK
.
Cu
rre
n
tl
y
,
sh
e
is
ro
u
n
d
in
g
u
p
h
e
r
M
a
ste
rs
in
E
lec
tri
c
a
l
a
n
d
El
e
c
tro
n
ic
E
n
g
in
e
e
rin
g
a
t
th
e
U
n
i
v
e
rsit
y
o
f
Jo
h
a
n
n
e
sb
u
rg
,
S
o
u
t
h
A
f
rica
.
S
h
e
is
a
re
se
a
rc
h
e
r
a
n
d
tea
c
h
e
r
w
it
h
o
v
e
r
5
y
e
a
rs
o
f
e
x
p
e
rien
c
e
.
He
r
r
e
se
a
r
c
h
in
tere
sts
in
c
lu
d
e
m
a
c
h
in
e
lea
rn
in
g
,
d
e
e
p
lea
rn
in
g
,
Io
T
,
f
in
a
n
c
ial
m
o
d
e
li
n
g
,
a
n
d
a
n
a
ly
ti
c
s.
Eb
e
n
e
z
e
r
Es
e
n
o
g
h
o
re
c
e
iv
e
d
th
e
Dip
lo
m
a
in
Co
m
p
u
ter
En
g
in
e
e
rin
g
(Up
p
e
r
Div
isio
n
)
in
2
0
0
3
,
B.
En
g
.
d
e
g
re
e
(
2
n
d
Clas
s
Ho
n
s
Up
p
e
r
Div
.
)
in
Co
m
p
u
ter
E
n
g
in
e
e
rin
g
in
2
0
0
8
,
t
h
e
M
.
E
n
g
.
d
e
g
re
e
(Up
p
e
r
Div
.
)
in
El
e
c
tro
n
ic/T
e
lec
o
m
m
u
n
ica
ti
o
n
En
g
in
e
e
rin
g
f
ro
m
th
e
Un
iv
e
rsit
y
o
f
Be
n
in
in
2
0
1
2
.
He
p
re
v
io
u
sly
lec
tu
re
d
a
t
th
e
Un
iv
e
rsit
y
o
f
Be
n
in
f
ro
m
M
a
rc
h
2
0
1
1
to
2
0
1
3
a
n
d
ro
se
t
o
ra
n
k
o
f
L
e
c
tu
re
r
I,
b
e
f
o
re
e
m
b
a
rk
in
g
o
n
h
is
P
h
.
D.
p
ro
g
ra
m
in
2
0
1
3
.
P
rio
r
to
n
o
w
,
h
e
w
a
s
in
v
o
lv
e
d
in
re
se
a
rc
h
a
n
d
tea
c
h
in
g
w
it
h
th
e
Ce
n
tre
f
o
r
Ra
d
io
Ac
c
e
ss
a
n
d
Ru
ra
l
T
e
c
h
n
o
lo
g
y
,
Un
iv
e
rsit
y
o
f
K
w
a
Zu
lu
-
Na
tal,
Ce
n
tre
o
f
Ex
c
e
ll
e
n
c
e
,
w
h
e
re
h
e
ro
u
n
d
e
d
-
u
p
h
is
P
h
.
D.
d
e
g
re
e
in
El
e
c
tro
n
ic
(5
G
Co
g
n
it
iv
e
N
e
t
w
o
rk
)
f
ro
m
2
0
1
3
-
2
0
1
7
f
u
n
d
e
d
b
y
A
l
c
a
tel,
Eri
c
c
sio
n
a
n
d
Hu
a
we
i.
Cu
rre
n
tl
y
,
h
e
is
h
ired
in
to
th
e
p
r
e
stig
io
u
s
G
lo
b
a
l
Ex
c
e
ll
e
n
c
e
a
n
d
S
tatu
re
(G
ES
)
P
o
st
-
Do
c
t
o
ra
l
Re
se
a
rc
h
F
e
ll
o
w
sh
ip
a
t
th
e
U
n
iv
e
rsit
y
o
f
Jo
h
a
n
n
e
sb
u
rg
,
A
u
c
k
lan
d
P
a
rk
u
n
d
e
r
t
h
e
In
st
it
u
te
f
o
r
In
telli
g
e
n
t
S
y
ste
m
(IIS
),
Ce
n
ter
f
o
r
T
e
lec
o
m
m
u
n
ica
ti
o
n
Re
se
a
rc
h
(Cf
T
)
to
p
u
rs
u
e
f
u
rth
e
r
re
se
a
rc
h
.
He
is
a
re
c
ip
ien
t
o
f
se
v
e
ra
l
g
ra
n
ts/sc
h
o
lars
h
ip
/aw
a
rd
/f
e
ll
o
w
sh
ip
s
in
c
lu
d
i
n
g
th
e
CEP
S
/E
sk
o
m
’s
HV
DC
2
0
1
3
,
CE
P
S
/E
sk
o
m
’s
HV
DC
2
0
1
4
,
J.
W
Ne
lso
n
2
0
1
5
a
n
d
G
ES
2
0
1
7
,
2
0
1
8
,
2
0
1
9
/2
0
.
He
is
p
a
ss
io
n
a
te
a
b
o
u
t
se
rv
ice
to
h
u
m
a
n
it
y
a
n
d
se
rv
e
d
in
se
v
e
ra
l
c
o
m
m
u
n
it
y
se
rv
ice
s
b
o
th
in
th
e
Re
p
u
b
li
c
o
f
S
o
u
t
h
Af
rica
a
n
d
h
is
h
o
m
e
c
o
u
n
try
.
His
le
a
d
e
rsh
ip
se
rv
ice
is
e
v
id
e
n
c
e
a
s
a
f
irst
p
o
stg
ra
d
u
a
te
re
p
re
se
n
tativ
e
f
o
r
S
tu
d
e
n
t
E
n
g
in
e
e
rin
g
Co
u
n
c
il
in
t
h
e
F
a
c
u
lt
y
m
a
n
a
g
e
m
e
n
t
b
o
a
rd
o
f
Un
iv
e
rsit
y
o
f
K
w
a
Zu
lu
-
Na
tal
d
u
ri
n
g
h
is
P
h
D
d
a
y
s,
2
0
1
6
-
2
0
1
7
.
P
re
se
n
tl
y
,
h
e
se
rv
in
g
a
s
th
e
f
irst
p
o
std
o
c
to
ra
l
re
se
a
rc
h
fe
ll
o
w
sh
ip
re
p
re
se
n
tativ
e
in
th
e
Un
iv
e
rsit
y
o
f
Jo
h
a
n
n
e
sb
u
rg
S
e
n
a
te/Co
u
n
c
il
,
2
0
1
8
-
d
a
te.
He
wa
s
a
UJ
/DS
T
/N
RF
re
se
a
rc
h
d
e
le
g
a
te
to
th
e
H2
0
2
0
-
ES
A
S
TA
P
EU
-
S
o
u
t
h
A
f
r
ica
S
T
I
Co
o
p
e
ra
ti
o
n
o
n
S
tren
g
th
i
n
g
T
e
c
h
n
o
lo
g
y
,
Re
s
e
a
rc
h
a
n
d
In
n
o
v
a
ti
o
n
i
n
V
ien
n
a
,
A
u
stria.
Dr.
Eb
e
n
e
z
e
r
h
a
s
a
u
t
h
o
re
d
/co
-
a
u
th
o
re
d
se
v
e
ra
l
p
e
e
r
-
re
v
ie
w
e
d
jo
u
r
n
a
ls
a
n
d
c
o
n
f
e
re
n
c
e
p
a
p
e
rs,
c
h
a
ired
se
ss
io
n
s
in
c
o
n
f
e
re
n
c
e
s
a
n
d
,
re
v
iew
e
d
f
o
r
so
m
e
n
o
tab
le
IS
I/S
c
o
p
u
s
j
o
u
r
n
a
ls
in
h
is
f
ield
.
He
is
v
e
ry
fa
m
il
iar
w
it
h
IS
0
-
9
0
0
1
/
1
8
0
0
1
/1
4
0
0
1
/
2
2
3
0
1
/
2
7
0
0
1
/
2
0
0
0
0
-
1
w
it
h
a
stro
n
g
a
ff
in
it
y
f
o
r
p
ro
jec
t
li
n
k
e
d
to
R&
D.
His
re
se
a
rc
h
in
te
re
sts
a
re
in
b
u
t
n
o
t
li
m
it
e
d
to
th
e
F
if
th
G
e
n
e
ra
ti
o
n
(5
G
)
W
irele
s
s
Ne
t
w
o
rk
s,
Co
g
n
it
iv
e
Ra
d
io
Ne
tw
o
rk
s,
Ne
t
w
o
rk
S
e
c
u
rit
y
/C
y
b
e
rs
e
c
u
rit
y
,
S
m
a
rt
G
rid
Ne
tw
o
rk
s,
Io
T
/Io
E,
S
DN
/S
DR,
W
irele
ss
S
e
n
so
r
Ne
t
w
o
rk
s,
A
rti
f
icia
l
In
telli
g
e
n
c
e
/M
a
c
h
in
e
L
e
a
rn
in
g
,
Big
Da
ta
&
M
o
b
i
le/Clo
u
d
Co
m
p
u
ti
n
g
a
n
d
V
isib
le
li
g
h
t
c
o
m
m
u
n
ica
ti
o
n
.
H
e
is
a
re
g
istere
d
En
g
in
e
e
r
in
Nig
e
ria
a
n
d
a
m
e
m
b
e
r
o
f
S
o
u
t
h
Af
rica
In
stit
u
te
o
f
El
e
c
tri
c
a
l
E
n
g
in
e
e
rs
(S
A
IEE
)
a
n
d
IEE
E
re
g
io
n
8
.
Evaluation Warning : The document was created with Spire.PDF for Python.