T
E
L
K
O
M
N
I
K
A
T
elec
o
m
m
un
ica
t
io
n,
Co
m
pu
t
ing
,
E
lect
ro
nics
a
nd
Co
ntr
o
l
Vo
l.
19
,
No
.
3
,
J
u
n
e
2
0
2
1
,
p
p
.
90
2
~
91
0
I
SS
N:
1
6
9
3
-
6
9
3
0
,
ac
cr
ed
ited
First Gr
ad
e
b
y
Kem
en
r
is
tek
d
i
k
ti,
Dec
r
ee
No
: 2
1
/E/KPT
/2
0
1
8
DOI
:
1
0
.
1
2
9
2
8
/TE
L
KOM
NI
K
A.
v
1
9
i3
.
1
8
8
7
6
902
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//jo
u
r
n
a
l.u
a
d
.
a
c.
id
/in
d
ex
.
p
h
p
/TELK
OM
N
I
K
A
Unidirec
tiona
l
-
bi
directiona
l re
curr
ent
ne
tworks
for cardia
c diso
rd
ers clas
sifica
tion
Anni
s
a
Da
rm
a
wa
hy
un
i,
Siti
Nurm
a
ini
,
M
uh
a
m
m
a
d Na
ufa
l R
a
chm
a
t
ull
a
h,
F
irda
us
F
irda
u
s
,
B
a
m
ba
ng
T
utuk
o
In
telli
g
e
n
t
S
y
ste
m
Re
se
a
rc
h
G
ro
u
p
,
F
a
c
u
l
ty
o
f
Co
m
p
u
ter
S
c
ien
c
e
,
Un
iv
e
rsitas
S
riwij
a
y
a
,
P
a
lem
b
a
n
g
,
In
d
o
n
e
sia
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
u
l 1
,
2
0
2
0
R
ev
is
ed
No
v
7
,
2
0
2
0
Acc
ep
ted
No
v
2
5
,
2
0
2
0
Th
e
d
e
e
p
lea
rn
i
n
g
a
p
p
ro
a
c
h
o
f
su
p
e
rv
ise
d
re
c
u
rre
n
t
n
e
two
r
k
c
las
sif
iers
m
o
d
e
l,
i.
e
.
,
re
c
u
rre
n
t
n
e
u
ra
l
n
e
two
r
k
s
(R
NN
s),
lo
n
g
sh
o
rt
-
term
m
e
m
o
ry
(L
S
TM
),
a
n
d
g
a
ted
re
c
u
rre
n
t
u
n
it
s
(G
RUs
)
a
re
u
se
d
in
t
h
is
stu
d
y
.
Th
e
u
n
i
d
irec
ti
o
n
a
l
a
n
d
b
id
irec
ti
o
n
a
l
fo
r
e
a
c
h
c
a
rd
iac
d
iso
rd
e
r
(CDs
)
c
las
s
is
a
lso
c
o
m
p
a
re
d
.
Co
m
p
a
rin
g
b
o
t
h
p
h
a
se
s
is
n
e
e
d
e
d
to
f
ig
u
re
o
u
t
t
h
e
o
p
ti
m
u
m
p
h
a
se
a
n
d
th
e
b
e
st
m
o
d
e
l
p
e
rfo
rm
a
n
c
e
fo
r
ECG
u
si
n
g
t
h
e
P
h
y
sio
n
e
t
d
a
tas
e
t
to
c
las
sify
fiv
e
c
las
se
s
o
f
CDs
wit
h
1
5
lea
d
s
ECG
sig
n
a
ls.
Th
e
re
su
l
t
sh
o
ws
t
h
a
t
t
h
e
b
i
d
irec
ti
o
n
a
l
RNN
s
m
e
th
o
d
p
ro
d
u
c
e
s
b
e
tt
e
r
re
su
lt
s
th
a
n
th
e
u
n
id
irec
ti
o
n
a
l
m
e
th
o
d
.
In
c
o
n
tras
t
t
o
RNN
s,
th
e
u
n
id
irec
t
io
n
a
l
LS
T
M
a
n
d
G
RU
o
u
tp
e
rf
o
rm
e
d
t
h
e
b
id
irec
ti
o
n
a
l
p
h
a
se
.
T
h
e
b
e
st
re
c
u
rre
n
t
n
e
tw
o
rk
c
las
sifier
p
e
rf
o
rm
a
n
c
e
is
u
n
i
d
irec
ti
o
n
a
l
G
RU
with
a
v
e
ra
g
e
a
c
c
u
ra
c
y
,
se
n
siti
v
it
y
,
sp
e
c
ifi
c
it
y
,
p
re
c
isio
n
,
a
n
d
F
1
-
sc
o
re
o
f
9
8
.
5
0
%
,
9
5
.
5
4
%
,
9
8
.
4
2
%
,
8
9
.
9
3
%
,
9
2
.
3
1
%
,
re
sp
e
c
ti
v
e
ly
.
Ov
e
ra
ll
,
d
e
e
p
lea
rn
i
n
g
is
a
p
ro
m
isi
n
g
imp
r
o
v
e
d
m
e
th
o
d
fo
r
ECG
c
las
sifica
ti
o
n
.
K
ey
w
o
r
d
s
:
B
id
ir
ec
tio
n
al
G
ated
r
ec
u
r
r
e
n
t u
n
it
L
o
n
g
s
h
o
r
t
-
ter
m
m
e
m
o
r
y
R
ec
u
r
r
en
t n
eu
r
al
n
etwo
r
k
s
U
n
id
ir
ec
tio
n
al
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
:
Sit
i N
u
r
m
ain
i
I
n
tellig
en
t Sy
s
tem
R
esear
ch
Gr
o
u
p
Facu
lty
o
f
C
o
m
p
u
ter
Scien
ce
,
Un
iv
er
s
itas
Sriwijay
a
Palem
b
an
g
3
0
1
3
9
,
I
n
d
o
n
esia
E
m
ail:
s
iti_
n
u
r
m
ain
i@
u
n
s
r
i.a
c.
id
,
s
itin
u
r
m
ain
i@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
C
ar
d
iac
d
is
o
r
d
er
s
(
C
Ds)
ar
e
in
cr
ea
s
in
g
ly
r
ec
o
g
n
ize
d
as
th
e
wo
r
ld
’
s
lead
i
n
g
ca
u
s
e
o
f
d
ea
th
.
T
h
e
d
is
o
r
d
er
s
in
clu
d
e
th
e
ca
r
d
iac
m
u
s
cle
an
d
th
e
v
ascu
lar
s
y
s
tem
s
u
p
p
ly
in
g
th
e
b
r
ain
,
h
e
ar
t,
an
d
o
th
er
v
ital
o
r
g
an
s
[
1
,
2
]
.
I
d
en
tif
y
in
g
s
u
b
s
ets
o
f
th
e
C
D
b
y
u
s
in
g
a
n
ele
ctr
o
ca
r
d
io
g
r
am
(
E
C
G)
s
ig
n
al
h
as
b
ee
n
o
n
e
o
f
th
e
g
r
ea
t
ad
v
a
n
ce
s
o
f
m
o
d
er
n
m
ed
icin
e.
Ho
wev
er
,
th
e
e
x
p
er
t
d
o
e
s
n
o
t
alwa
y
s
r
ea
lize
th
e
im
p
o
r
t
an
ce
o
f
class
if
y
in
g
d
is
ea
s
es
b
ased
o
n
th
e
E
C
G
s
i
g
n
al.
Du
e
to
a
h
i
g
h
m
o
r
tality
r
ate
o
f
C
Ds,
ea
r
ly
d
etec
tio
n
o
f
th
e
n
o
r
m
al
an
d
ab
n
o
r
m
al
E
C
G
s
ig
n
al
is
ess
en
t
ial
f
o
r
th
e
p
atien
t’
s
tr
ea
tm
en
t.
B
y
m
an
u
ally
d
if
f
er
en
t
E
C
G
wa
v
ef
o
r
m
s
o
f
p
atien
ts
,
with
th
e
d
o
m
ain
wo
r
k
lo
ad
,
th
e
ex
p
er
ts
m
ay
h
av
e
m
is
u
n
d
er
s
to
o
d
an
d
ca
n
af
f
ec
t
th
e
p
r
ec
is
e
ju
d
g
m
en
t
f
o
r
C
Ds
d
iag
n
o
s
e.
T
h
e
s
ig
n
al
m
o
r
p
h
o
l
o
g
y
ca
n
g
et
ch
an
g
ed
b
y
an
y
ir
r
eg
u
lar
ity
in
th
e
ca
r
d
iac
r
h
y
t
h
m
o
r
ca
r
d
iac
m
u
s
cle
d
am
ag
e.
Ne
v
er
th
eless
,
th
e
n
o
r
m
al
E
C
G
ca
n
d
if
f
e
r
f
o
r
ea
ch
p
er
s
o
n
,
a
n
d
two
d
is
tin
ct
d
is
ea
s
es
ca
n
h
av
e
a
b
o
u
t
th
e
s
am
e
ef
f
ec
ts
o
n
n
o
r
m
al
E
C
G
s
ig
n
als.
T
h
er
ef
o
r
e
,
th
e
au
t
o
m
atic
class
i
f
icatio
n
s
ch
em
e
i
s
n
ee
d
ed
to
a
d
d
r
ess
th
e
m
is
in
ter
p
r
etatio
n
o
f
th
e
E
C
G
s
ig
n
al
v
ar
iab
ilit
y
.
T
h
e
class
if
icatio
n
p
r
o
ce
s
s
o
f
E
C
G
p
lay
s
th
e
m
o
s
t
cr
u
cial
r
o
le
in
th
e
clin
ical
d
iag
n
o
s
is
o
f
C
Ds
[
3
]
.
Af
ter
id
en
tify
in
g
th
e
a
b
n
o
r
m
ality
,
C
Ds
ca
n
b
e
d
etec
ted
,
an
d
th
e
p
a
tien
ts
g
et
b
etter
tr
ea
tm
en
t.
C
u
r
r
en
tly
,
co
m
p
u
ter
-
aid
ed
d
iag
n
o
s
is
(
C
AD)
wo
u
ld
b
e
ab
le
to
p
r
o
v
id
e
th
e
C
D
class
if
icatio
n
s
.
I
t
ca
n
d
ev
elo
p
an
E
C
G
s
ig
n
al
class
if
icatio
n
alg
o
r
ith
m
with
v
ar
io
u
s
s
ig
n
al
p
r
o
ce
s
s
in
g
tech
n
i
q
u
es
to
im
p
r
o
v
e
E
C
G
class
if
ic
atio
n
p
e
r
f
o
r
m
an
c
e
.
T
h
e
class
if
icatio
n
p
r
o
ce
s
s
ca
n
p
r
o
v
id
e
s
u
b
s
tan
tial
in
p
u
t
to
ex
p
er
ts
to
co
n
f
ir
m
th
e
d
iag
n
o
s
is
.
Un
f
o
r
tu
n
ately
,
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
Un
id
ir
ec
tio
n
a
l
-
b
id
ir
ec
tio
n
a
l r
ec
u
r
r
en
t n
etw
o
r
ks fo
r
ca
r
d
ia
c
d
is
o
r
d
ers
cla
s
s
ifica
tio
n
(
A
n
n
is
a
Da
r
ma
w
a
h
yu
n
i
)
903
th
er
e
ar
e
s
ev
er
al
s
ig
n
if
ican
t p
r
o
b
lem
s
in
E
C
G
s
ig
n
al
class
if
i
ca
tio
n
[
4
,
5
]
,
e.
g
.
,
lack
o
f
s
tan
d
ar
d
izatio
n
o
f
E
C
G
f
ea
tu
r
es,
th
e
v
ar
ia
b
ilit
y
o
f
p
a
tien
ts
E
C
G
wav
ef
o
r
m
s
,
n
o
n
-
ex
is
ten
ce
o
f
o
p
tim
al
class
if
icatio
n
r
u
les,
an
d
t
h
e
co
n
f
id
en
tial
in
f
o
r
m
atio
n
an
d
d
if
f
er
en
t
k
in
d
s
o
f
n
o
is
e
in
E
C
G
s
ig
n
als,
s
u
ch
as
b
aselin
e
d
r
if
t
an
d
p
o
wer
lin
e
in
ter
f
er
en
ce
.
B
esid
es,
th
e
av
a
ilab
le
lab
eled
d
ata
ar
e
n
ec
ess
ar
y
to
en
h
a
n
ce
th
e
p
r
ec
is
io
n
o
f
th
e
class
if
icatio
n
p
r
o
ce
s
s
.
I
f
o
n
ly
a
lim
ited
n
u
m
b
er
o
f
lab
eled
ex
a
m
p
les
is
av
ailab
le,
th
e
class
if
icatio
n
p
er
f
o
r
m
an
ce
m
ay
b
e
q
u
ite
u
n
s
atis
f
ac
to
r
y
[
6
]
.
Hen
ce
,
ac
c
o
r
d
in
g
to
s
u
ch
co
n
d
itio
n
s
,
th
e
E
C
G
s
ig
n
al
class
if
icatio
n
o
n
C
D
is
d
esira
b
le
to
in
v
esti
g
ate
to
p
r
o
d
u
ce
ac
c
u
r
at
e
au
to
m
atic
d
iag
n
o
s
tic.
Sev
er
al
m
eth
o
d
s
h
av
e
b
ee
n
p
r
o
p
o
s
ed
to
o
v
er
c
o
m
e
th
e
E
C
G
s
ig
n
al
class
if
icatio
n
p
r
o
b
lem
with
g
o
o
d
r
esu
lts
.
Ma
ch
in
e
lear
n
in
g
te
ch
n
iq
u
es
class
if
y
th
e
h
ig
h
n
u
m
b
er
o
f
E
C
G
s
ig
n
al
class
e
s
in
an
au
to
m
ated
ma
n
n
er
[7
-
1
1
]
.
Sti
ll,
th
e
f
ea
tu
r
es
ar
e
ty
p
ically
h
an
d
-
cr
af
ted
o
r
ex
tr
ac
ted
h
e
u
r
is
tically
.
Hen
ce
,
m
ac
h
in
e
lear
n
i
n
g
r
eq
u
ir
es m
o
r
e
ef
f
o
r
t,
is
tim
e
-
c
o
n
s
u
m
in
g
,
an
d
s
o
m
etim
es,
f
ea
tu
r
e
r
e
p
r
esen
tatio
n
s
ar
e
o
f
ten
u
n
r
eliab
le.
I
n
r
ec
en
t
y
ea
r
s
,
d
ee
p
lear
n
i
n
g
(
DL
)
h
as
b
ee
n
u
s
ed
f
o
r
im
p
r
o
v
in
g
f
ea
tu
r
e
r
ep
r
esen
tatio
n
p
r
o
b
le
m
s
in
co
n
v
en
tio
n
al
m
ac
h
in
e
lear
n
in
g
.
DL
h
as
ap
p
ea
r
ed
as
t
h
e
lead
i
n
g
tech
n
iq
u
e
th
at
u
s
es
s
u
p
er
v
is
ed
o
r
u
n
s
u
p
er
v
is
ed
ap
p
r
o
ac
h
es
to
lear
n
f
ea
tu
r
es
au
to
m
atica
ll
y
.
DL
i
s
clo
s
ely
r
elate
d
to
a
c
lass
o
f
b
r
ain
d
ev
elo
p
m
e
n
t
th
e
o
r
ies
d
is
co
v
er
ed
as
f
ea
tu
r
e
in
ter
ac
tio
n
th
at
ca
n
b
e
s
im
u
ltan
eo
u
s
ly
m
ain
tai
n
ed
with
in
th
e
m
o
r
e
in
-
d
e
p
th
n
eu
r
al
n
etwo
r
k
ar
ch
itectu
r
e
[
1
2
]
.
DL
h
as
b
ee
n
im
p
lem
en
ted
in
b
io
m
ed
ical
e
n
g
in
ee
r
in
g
ap
p
licatio
n
s
,
i
.
e.
,
cla
s
s
if
icatio
n
o
f
C
Ds.
Sev
er
al
s
tu
d
ies
p
r
o
p
o
s
ed
DL
in
v
ar
ian
t
ar
c
h
itectu
r
es,
e.
g
.
c
o
n
v
o
lu
ti
o
n
al
n
e
u
r
al
n
etwo
r
k
s
(
C
NNs)
[
1
3
,
1
4
]
,
r
ec
u
r
r
en
t
n
eu
r
al
n
etwo
r
k
s
(
R
NNs)
[
1
5
,
1
6
]
,
d
ee
p
b
elief
n
etw
o
r
k
s
(
DB
Ns)
[
1
7
]
,
a
n
d
Au
t
o
en
co
d
er
[
1
8
]
.
Dee
p
lear
n
in
g
o
u
tp
er
f
o
r
m
ed
t
h
e
co
n
v
en
tio
n
al
class
if
ier
alg
o
r
ith
m
s
f
o
r
class
if
icatio
n
task
s
[
1
9
-
2
1
]
.
No
n
eth
eless
,
n
o
n
e
o
f
all
th
e
af
o
r
em
en
tio
n
ed
ar
c
h
itectu
r
es
o
f
DL
ca
n
a
p
p
r
o
p
r
iate
f
o
r
th
e
clin
ical
p
r
o
b
lem
s
.
Acc
o
r
d
in
g
to
th
e
liter
atu
r
e,
th
e
ex
is
tin
g
p
u
b
lis
h
ed
a
r
ticles
ar
e
lim
ite
d
f
o
r
E
C
G
-
r
h
y
th
m
-
b
ased
class
if
icatio
n
b
ec
au
s
e
th
e
r
ig
h
t
d
eter
m
in
atio
n
o
f
tim
e
-
win
d
o
ws
in
E
C
G
-
r
h
y
th
m
class
if
icatio
n
is
n
o
t
s
tr
ai
g
h
tf
o
r
war
d
[
2
2
]
.
T
h
e
o
p
tim
u
m
win
d
o
w
s
ize
d
ep
en
d
s
o
n
th
e
task
;
if
it
is
to
o
s
m
all
,
th
e
n
etwo
r
k
will
ig
n
o
r
e
im
p
o
r
tan
t
in
f
o
r
m
atio
n
;
o
th
er
wis
e,
it
will
o
v
er
f
it
th
e
tr
ain
in
g
d
ata
[
2
3
,
2
4
]
.
Fo
r
E
C
G
class
if
icatio
n
,
th
e
two
m
o
s
t
ex
citin
g
f
ield
s
o
f
DL
ar
e
R
NNs
an
d
C
NNs,
b
u
t
C
NNs,
wh
en
ap
p
lied
to
E
C
G,
cu
t
th
e
win
d
o
w
s
ize
o
f
a
f
ix
ed
len
g
th
th
at
ev
en
t
u
ally
r
ed
u
ce
s
th
e
class
if
icatio
n
p
er
f
o
r
m
an
ce
[
2
5
]
.
R
NNs
ca
n
b
e
im
p
r
o
v
ed
in
asp
ec
t,
as
th
e
p
er
f
o
r
m
a
n
ce
ca
n
b
e
o
p
tim
ized
b
y
p
r
o
v
i
d
in
g
th
e
cl
ass
if
ier
with
cr
af
ted
f
ea
t
u
r
es
[
2
5
]
.
R
NNs
u
s
e
in
ter
n
al
m
em
o
r
y
to
p
r
o
ce
s
s
an
d
id
en
tify
a
r
b
itra
r
y
in
p
u
t
s
eq
u
en
ce
s
,
an
d
th
ese
r
elatio
n
s
b
etwe
en
th
e
u
n
its
f
o
r
m
a
d
i
r
e
cted
cy
cle
[
2
5
]
.
L
u
i
et
a
l.
[
1
5
]
wer
e
f
o
u
n
d
th
at
th
e
ad
d
itio
n
o
f
a
r
ec
u
r
r
e
n
t
lay
er
im
p
r
o
v
e
d
th
e
C
Ds
clas
s
if
icatio
n
s
en
s
itiv
ity
u
s
in
g
E
C
G
b
y
2
8
%
co
m
p
ar
ed
to
th
e
C
NNs
alo
n
e.
T
h
e
liter
atu
r
e
h
as
s
h
o
wn
th
e
p
er
ce
n
tag
e
o
f
s
en
s
itiv
ity
ca
r
r
ies
o
u
t
th
e
ef
f
ec
tiv
en
ess
o
f
C
Ds
class
if
icatio
n
u
s
in
g
R
NN
s
.
R
NNs
ca
n
b
e
im
p
lem
e
n
ted
f
o
r
s
e
q
u
en
tial
p
r
ed
ictio
n
to
m
o
d
el
th
e
f
lo
w
o
f
tim
e
d
ir
ec
tly
.
R
NNs
an
d
its
v
ar
ian
ts
(
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
L
STM
)
an
d
g
ated
r
ec
u
r
r
e
n
t
u
n
it
(
GR
U)
)
ca
n
b
e
im
p
lem
en
t
ed
in
a
u
n
id
ir
ec
tio
n
al
an
d
b
id
ir
ec
tio
n
al
p
h
ase.
A
s
tan
d
a
r
d
R
NNs,
u
n
id
ir
ec
tio
n
al,
in
wh
ich
th
e
in
p
u
t
is
in
ter
p
r
ete
d
f
r
o
m
lef
t
to
r
ig
h
t
(
f
u
tu
r
e
in
p
u
ts
)
,
i.e
.
,
th
e
in
f
o
r
m
atio
n
f
lo
w
is
a
f
o
r
war
d
d
ir
ec
tio
n
o
n
ly
.
Sch
u
s
ter
s
u
g
g
ested
th
e
b
id
ir
ec
tio
n
al
R
NNs
p
h
ase
to
u
s
e
b
o
th
p
ast
an
d
f
u
tu
r
e
in
p
u
ts
f
o
r
p
r
e
d
ictio
n
[
2
6
]
.
A
u
n
id
ir
ec
t
io
n
al
p
h
ase
also
h
as
lim
itatio
n
s
b
ec
au
s
e
it
is
d
if
f
icu
lt
to
attain
f
u
tu
r
e
in
p
u
t
in
f
o
r
m
atio
n
f
r
o
m
th
e
cu
r
r
en
t
s
tate.
On
th
e
co
n
tr
a
r
y
,
b
id
ir
ec
tio
n
al
d
o
es
n
o
t
r
eq
u
ir
e
f
ix
in
g
o
f
its
in
p
u
t d
ata.
B
esid
es,
f
u
tu
r
e
in
p
u
t
d
ata
is
ac
ce
s
s
ib
le
f
r
o
m
th
e
cu
r
r
en
t state
[2
7]
.
Yild
ir
im
h
as d
esig
n
ed
b
o
t
h
u
n
id
ir
ec
tio
n
al
an
d
b
id
ir
ec
tio
n
al
p
h
ase
f
o
r
E
C
G
b
ea
t
m
u
lticlas
s
clas
s
if
ic
atio
n
[
2
8
]
.
T
h
e
p
er
f
o
r
m
a
n
ce
r
esu
lt
s
h
o
ws
th
e
u
n
id
ir
ec
tio
n
al
s
h
o
wed
a
7
3
.
1
0
%
s
u
cc
ess
r
ate,
an
d
th
e
b
id
ir
ec
tio
n
al
p
h
ase
p
r
o
v
id
ed
a
m
u
ch
b
etter
p
er
f
o
r
m
an
ce
with
a
7
9
.
5
3
%
s
u
cc
ess
r
ate.
Ho
wev
er
,
in
s
o
m
e
ca
s
es,
th
e
u
n
id
ir
ec
tio
n
al
r
ec
u
r
r
en
t
n
etw
o
r
k
s
ar
e
s
till
o
u
tp
er
f
o
r
m
ed
th
e
b
id
ir
ec
tio
n
al,
b
y
win
d
o
win
g
,
l
o
o
k
in
g
ah
ea
d
,
o
r
d
elay
in
g
t
h
e
o
u
t
p
u
t,
it
ca
n
s
till
ac
ce
s
s
f
u
tu
r
e
in
p
u
ts
with
a
la
r
g
e
in
cr
ea
s
e
in
t
h
e
n
u
m
b
er
o
f
p
ar
am
eter
s
[
2
9
]
.
I
n
ad
d
itio
n
,
in
te
r
m
s
o
f
tim
e
ef
f
ic
ien
cy
,
b
id
ir
ec
tio
n
al
p
h
ase
r
e
q
u
ir
e
s
m
o
r
e
tim
e
th
an
u
n
id
ir
ec
tio
n
al
p
h
ase
in
th
e
lea
r
n
in
g
p
r
o
ce
s
s
[
2
8
]
.
T
h
is
p
ap
er
aim
s
to
ex
p
l
o
r
e
th
e
DL
tech
n
iq
u
e
with
r
ec
u
r
r
en
t
n
etwo
r
k
class
if
ier
s
f
o
r
m
u
lticlas
s
E
C
G
-
r
h
y
th
m
-
b
ased
class
if
icatio
n
.
T
h
e
ex
p
e
r
im
en
ts
co
n
ce
n
tr
ate
o
n
th
e
co
m
p
ar
is
o
n
o
f
u
n
id
ir
ec
tio
n
al
an
d
b
id
ir
ec
tio
n
al
r
ec
u
r
r
e
n
t
n
etwo
r
k
p
er
f
o
r
m
a
n
ce
.
T
h
e
co
m
p
a
r
is
o
n
is
n
ee
d
ed
t
o
e
v
alu
ate
an
d
f
ig
u
r
e
o
u
t
th
e
o
p
tim
u
m
p
h
ase
f
o
r
E
C
G
m
u
lticlas
s
p
er
f
o
r
m
an
ce
(
ac
cu
r
ac
y
,
s
en
s
itiv
ity
,
s
p
ec
if
icity
,
p
r
ec
is
io
n
,
an
d
F1
-
s
co
r
e)
u
s
in
g
th
e
av
ailab
le
p
u
b
lic
d
ataset
f
r
o
m
Ph
y
s
io
n
et.
T
h
e
p
r
o
ce
s
s
co
n
s
is
ts
o
f
s
o
m
e
s
tep
s
;
First,
th
e
tim
e
-
win
d
o
w
was
d
eter
m
in
ed
f
o
r
lear
n
in
g
in
ea
ch
ce
ll
s
tate
in
th
e
r
ec
u
r
r
e
n
t
n
etwo
r
k
.
Seco
n
d
,
i
n
s
tead
o
f
p
er
f
o
r
m
in
g
b
in
ar
y
class
if
icatio
n
,
a
m
u
lticlas
s
cla
s
s
if
ier
was
tr
ain
ed
with
h
ea
lth
y
co
n
tr
o
l,
m
y
o
ca
r
d
ial
in
f
ar
ctio
n
,
ca
r
d
io
m
y
o
p
at
h
y
,
b
u
n
d
le
b
r
an
c
h
b
lo
c
k
,
an
d
d
y
s
r
h
y
th
m
ia
class
es.
2.
RE
S
E
ARCH
M
E
T
H
O
D
T
h
is
p
ap
er
p
r
esen
ts
th
e
p
r
o
ce
s
s
o
f
m
u
lticlas
s
E
C
G
f
o
r
h
ea
lth
y
co
n
tr
o
l,
m
y
o
ca
r
d
ial
in
f
ar
ctio
n
,
ca
r
d
io
m
y
o
p
ath
y
,
b
u
n
d
le
b
r
a
n
ch
b
lo
ck
,
a
n
d
d
y
s
r
h
y
th
m
ia
class
if
icat
io
n
u
s
in
g
a
p
u
b
li
c
d
ataset,
th
e
PT
B
d
iag
n
o
s
tic
E
C
G
d
atab
ase.
I
t
co
n
s
is
ts
o
f
th
e
f
o
llo
win
g
m
ain
s
tep
s
:
1
)
d
eter
m
in
in
g
th
e
tim
e
-
win
d
o
w,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6
9
3
0
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
,
Vo
l.
19
,
No
.
3
,
J
u
n
e
2
0
2
1
:
90
2
-
91
0
904
2
)
co
m
p
ar
in
g
u
n
id
ir
ec
tio
n
al
a
n
d
b
id
ir
ec
tio
n
al
-
b
ased
lear
n
in
g
alg
o
r
ith
m
s
,
3
)
p
r
o
p
o
s
in
g
t
h
e
b
est
m
o
d
el
f
o
r
t
h
e
ap
p
licatio
n
.
All th
e
p
h
ases
o
f
t
h
e
p
r
o
p
o
s
ed
m
et
h
o
d
a
r
e
p
r
ese
n
ted
in
Fig
u
r
e
1
.
Fig
u
r
e
1
.
E
C
G
m
u
lticlas
s
clas
s
if
icatio
n
wo
r
k
f
lo
w
2
.
1
.
E
CG
ra
w
da
t
a
T
h
e
E
C
G
d
ata
u
s
ed
i
n
th
is
s
tu
d
y
was
co
llected
f
r
o
m
Ph
y
s
io
Net:
th
e
PTB
d
iag
n
o
s
tic
E
C
G
d
atab
ase
[
3
0
]
.
T
h
is
s
tu
d
y
'
s
alg
o
r
ith
m
test
s
th
e
d
atab
ase
b
ec
a
u
s
e
it
p
r
o
v
id
es
a
t
o
tal
o
f
5
4
9
r
ec
o
r
d
s
:
8
0
r
ec
o
r
d
s
with
h
ea
lth
y
c
o
n
tr
o
l
an
d
th
e
r
e
m
ain
in
g
r
ec
o
r
d
s
o
f
n
in
e
ca
r
d
ia
c
d
is
o
r
d
er
s
.
T
h
e
d
is
o
r
d
er
s
ca
n
b
e
s
ee
n
in
T
ab
le
1
.
E
ac
h
r
ec
o
r
d
in
cl
u
d
es
th
e
co
n
v
en
tio
n
al
1
2
lead
s
,
an
d
th
e
3
Fra
n
k
lead
E
C
Gs,
with
1
5
lea
d
s
,
ar
e
u
s
ed
in
th
is
s
tu
d
y
.
T
h
is
s
tu
d
y
aim
s
to
cl
ass
if
y
h
ea
lth
y
co
n
tr
o
l
p
atien
t
s
an
d
f
o
u
r
C
Ds,
i.e
.
,
1
)
m
y
o
ca
r
d
ial
i
n
f
ar
ctio
n
,
2
)
ca
r
d
io
m
y
o
p
ath
y
,
3
)
b
u
n
d
le
b
r
an
ch
b
lo
ck
,
an
d
4
)
d
y
s
r
h
y
th
m
ia.
T
h
e
o
th
er
ca
r
d
iac
d
is
o
r
d
er
s
in
th
is
d
atab
ase
wer
e
d
is
ca
r
d
ed
b
ec
au
s
e
th
ey
b
elo
n
g
ed
to
class
es
n
o
t
c
o
n
s
id
e
r
ed
in
th
e
s
tu
d
y
.
B
ased
o
n
o
u
r
p
r
ev
i
o
u
s
wo
r
k
f
o
r
b
in
ar
y
class
if
icatio
n
[
2
0
]
,
a
f
ix
ed
win
d
o
w
s
ize
o
f
4
s
ec
o
n
d
s
was
d
eter
m
in
ed
f
o
r
ea
ch
s
eq
u
en
ce
f
o
r
E
C
G
p
r
e
-
p
r
o
ce
s
s
in
g
.
T
h
e
t
o
tal
s
eq
u
en
ce
d
ata
f
o
r
f
iv
e
class
es wa
s
1
3
.
6
1
0
s
eq
u
e
n
ce
s
.
T
ab
le
1
.
T
h
e
PTB Diag
n
o
s
tic
E
C
G
d
atab
ase
d
escr
ip
tio
n
C
a
r
d
i
a
c
D
i
s
o
r
d
e
r
s (CDs
)
R
e
c
o
r
d
s
H
e
a
l
t
h
y
C
o
n
t
r
o
l
80
M
y
o
c
a
r
d
i
a
l
I
n
f
a
r
c
t
i
o
n
3
6
8
C
a
r
d
i
o
my
o
p
a
t
h
y
17
B
u
n
d
l
e
B
r
a
n
c
h
B
l
o
c
k
17
D
y
sr
h
y
t
h
mi
a
16
To
t
a
l
4
9
8
2
.
2
.
Rec
urre
nt
net
wo
r
k
cla
s
s
if
iers
T
h
e
s
eq
u
en
ce
m
o
d
el
co
n
s
is
ts
o
f
s
eq
u
en
ce
s
o
f
o
r
d
er
ed
elem
e
n
ts
,
r
ec
o
r
d
ed
with
o
r
with
o
u
t
a
co
n
cr
ete
n
o
tio
n
o
f
tim
e.
T
h
e
r
ec
u
r
r
en
t
p
r
o
ce
s
s
in
th
e
n
eu
r
al
n
etwo
r
k
o
p
er
ates
o
n
s
eq
u
e
n
ce
s
o
f
d
ata.
T
h
e
r
ec
u
r
r
en
t
n
etwo
r
k
tak
es
ea
ch
elem
en
t
o
f
a
s
eq
u
en
ce
,
m
u
ltip
lies
th
e
elem
en
t
b
y
a
m
atr
ix
,
an
d
th
e
p
r
e
v
io
u
s
o
u
tp
u
t
is
s
u
m
m
ed
f
r
o
m
th
e
n
etwo
r
k
.
T
h
er
e
ar
e
two
d
ir
ec
tio
n
s
f
o
r
th
e
lear
n
in
g
p
h
ase
in
n
eu
r
al
n
etw
o
r
k
s
:
u
n
id
ir
ec
tio
n
al
an
d
b
id
ir
ec
tio
n
al
[
3
1
]
.
T
h
e
u
n
id
ir
ec
tio
n
al
p
r
eser
v
es
t
h
e
in
f
o
r
m
atio
n
o
f
th
e
p
ast
an
d
r
u
n
s
th
e
i
n
p
u
ts
o
n
ly
in
f
o
r
war
d
(
lef
t
-
to
-
r
ig
h
t)
p
ass
es.
T
h
e
b
id
ir
ec
tio
n
al
p
h
ase
r
u
n
s
th
e
in
p
u
ts
in
th
e
f
o
r
war
d
(
lef
t
-
to
-
r
ig
h
t
)
an
d
b
ac
k
war
d
(
r
ig
h
t
-
to
-
lef
t)
p
ass
es
an
d
p
r
eser
v
es
th
e
i
n
f
o
r
m
atio
n
f
r
o
m
b
o
th
p
ast
an
d
f
u
t
u
r
e,
as
p
r
esen
ted
in
Fig
u
r
e
2
.
R
ec
u
r
r
en
t
n
eu
r
al
n
etwo
r
k
s
(
R
NNs)
ca
p
tu
r
e
r
elatio
n
s
h
ip
s
am
o
n
g
s
eq
u
e
n
tial
d
ata
ty
p
es.
Feed
b
ac
k
lo
o
p
s
at
h
id
d
en
lay
er
s
o
f
R
NNs
ar
e
u
n
id
ir
ec
tio
n
al.
Un
id
ir
ec
tio
n
al
m
ea
n
s
th
e
p
r
o
ce
s
s
f
r
o
m
lef
t
-
to
-
r
ig
h
t,
i
n
w
h
ich
th
e
f
lo
w
o
f
th
e
in
f
o
r
m
atio
n
is
o
n
ly
in
th
e
f
o
r
war
d
d
ir
ec
tio
n
[
2
9
]
.
A
u
n
id
ir
ec
tio
n
al
m
o
d
el
ca
n
s
till
ac
ce
s
s
in
p
u
ts
b
y
win
d
o
win
g
,
lo
o
k
in
g
-
ah
ea
d
,
o
r
d
elay
in
g
o
u
tp
u
t
with
a
r
ea
s
o
n
ab
le
in
cr
ea
s
e
in
th
e
n
u
m
b
er
o
f
p
a
r
am
eter
s
.
B
en
g
io
et
a
l.
[
3
2
]
s
h
o
wed
t
h
a
t
ca
p
tu
r
in
g
lo
n
g
-
ter
m
d
e
p
en
d
e
n
cies
u
s
in
g
a
s
im
p
le
R
NNs
ar
e
d
if
f
icu
lt
b
ec
au
s
e
g
r
ad
ien
ts
ten
d
to
eit
h
er
v
a
n
is
h
o
r
ex
p
l
o
d
e
with
lo
n
g
s
eq
u
en
c
es.
T
wo
tech
n
iq
u
es
h
a
v
e
b
ee
n
p
r
o
p
o
s
ed
to
s
o
lv
e
th
e
g
r
ad
ien
t
p
r
o
b
lem
s
: lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
L
STM
)
an
d
g
ated
r
ec
u
r
r
en
t
u
n
it (
GR
U)
.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
Un
id
ir
ec
tio
n
a
l
-
b
id
ir
ec
tio
n
a
l r
ec
u
r
r
en
t n
etw
o
r
ks fo
r
ca
r
d
ia
c
d
is
o
r
d
ers
cla
s
s
ifica
tio
n
(
A
n
n
is
a
Da
r
ma
w
a
h
yu
n
i
)
905
Fig
u
r
e
2
.
T
h
e
f
o
r
war
d
a
n
d
b
ac
k
war
d
p
ass
es o
f
u
n
id
ir
ec
tio
n
al
an
d
b
i
d
ir
ec
tio
n
al
Sch
u
s
ter
p
r
esen
ted
n
ew
co
n
ce
p
ts
o
f
s
eq
u
en
ce
lear
n
in
g
in
wh
ich
th
e
in
f
o
r
m
atio
n
f
lo
w
is
in
f
o
r
war
d
an
d
b
ac
k
war
d
f
ee
d
b
ac
k
[
2
6
]
.
T
h
e
co
n
n
ec
tio
n
s
in
th
e
f
o
r
war
d
d
i
r
ec
tio
n
h
el
p
u
s
le
ar
n
f
r
o
m
p
r
e
v
io
u
s
r
ep
r
esen
tatio
n
s
,
an
d
th
o
s
e
g
o
i
n
g
b
ac
k
war
d
h
elp
u
s
to
lear
n
f
r
o
m
f
u
tu
r
e
r
e
p
r
esen
tatio
n
s
.
B
o
th
co
n
n
ec
tio
n
s
a
r
e
ca
lled
“b
id
ir
ec
tio
n
al
R
NN
(
B
iR
NNs)”
,
wh
ich
en
ab
les th
e
n
e
two
r
k
to
p
r
ed
ict
o
u
t
p
u
ts
u
s
in
g
in
p
u
ts
o
f
th
e
e
n
tire
s
eq
u
en
ce
.
B
iR
NNs
ca
n
b
e
le
ar
n
ed
u
s
in
g
all
av
ailab
le
in
p
u
t
d
ata
f
o
r
a
s
p
ec
if
ic
tim
ef
r
a
m
e
in
th
e
p
ast
an
d
f
u
tu
r
e
[
3
3
]
.
B
iR
NNs
ar
e
tr
ain
ed
with
th
e
s
am
e
alg
o
r
ith
m
as
a
r
eg
u
lar
u
n
id
ir
ec
tio
n
al
R
NN
s
,
b
ec
au
s
e
th
er
e
ar
e
n
o
in
ter
ac
tio
n
s
b
etwe
en
th
e
tw
o
ty
p
es o
f
s
tate
n
e
u
r
o
n
s
.
Gr
av
es
et
a
l.
p
r
o
p
o
s
ed
an
R
NNs
th
at
u
s
es
L
STM
ce
ll
s
a
n
d
co
m
p
u
tes
b
o
th
f
o
r
war
d
an
d
b
ac
k
war
d
h
id
d
en
s
eq
u
en
ce
s
[
2
3
]
.
I
t
is
ca
lled
b
id
ir
ec
tio
n
al
L
STM
(
B
iLST
M)
.
B
iLST
M
i
s
th
e
L
STM
v
er
s
io
n
o
f
th
e
B
iR
NNs
ar
ch
itectu
r
e
an
d
ca
n
ex
p
an
d
L
STM
p
er
f
o
r
m
a
n
ce
in
class
if
icatio
n
p
r
o
ce
d
u
r
es
[
2
8
]
.
I
n
c
o
n
tr
ast
to
th
e
r
eg
u
lar
L
STM
s
tr
u
ctu
r
e,
two
d
is
s
im
ilar
L
STM
n
e
two
r
k
s
ar
e
tr
ain
ed
f
o
r
s
eq
u
e
n
tial
in
p
u
ts
in
th
e
B
iLST
M
ar
ch
itectu
r
e
[
2
8
]
.
T
h
e
n
e
u
r
o
n
in
a
f
o
r
war
d
s
tate
o
f
B
iLST
M
ac
ts
as
a
u
n
id
ir
ec
tio
n
al
L
STM
s
tr
u
ctu
r
e,
b
u
t
b
id
ir
ec
tio
n
al
n
etwo
r
k
s
a
r
e
s
till
m
u
ch
m
o
r
e
ef
f
ec
tiv
e
th
an
u
n
id
ir
ec
tio
n
al
n
etwo
r
k
s
[
2
3
]
.
T
h
e
c
u
r
r
e
n
t
h
id
d
e
n
s
tate
d
ep
en
d
s
o
n
two
h
id
d
en
s
tates:
th
e
f
o
r
war
d
an
d
th
e
b
ac
k
wa
r
d
p
ass
o
f
L
STM
.
T
h
e
B
iLST
M
eq
u
atio
n
s
in
th
e
f
o
r
war
d
a
n
d
b
ac
k
war
d
p
ass
es a
r
e
g
iv
en
b
elo
w
[
3
1
]
:
1
1
1
1
1
1
t
a
n
h
(
h
t
h
h
t
h
i
ft
b
LS
TM
W
x
W
M
LS
T
+
+
=
−
(
1
)
1
1
1
1
1
1
t
a
n
h
(
h
t
h
h
t
h
i
bt
b
LS
TM
W
x
W
M
LS
T
+
+
=
+
(
2
)
F
r
o
m
(
1
0
)
an
d
(
1
1
)
,
th
e
o
u
tp
u
t
o
f
B
iLST
M
lay
er
at
a
tim
e
t
:
1
0
1
1
1
1
1
t
a
n
h
(
b
M
L
S
T
W
M
L
S
T
W
y
t
o
h
t
o
h
t
+
+
=
(
3
)
wh
er
e
th
e
o
u
tp
u
t d
e
p
en
d
s
o
n
t
M
L
S
T
an
d
t
M
L
S
T
;
0
h
is
in
itialized
as a
ze
r
o
v
ec
to
r
.
C
h
o
et
a
l.
im
p
lem
en
ted
th
e
G
R
U
to
allo
w
ea
ch
r
ec
u
r
r
en
t
u
n
i
t
to
ad
ap
tiv
ely
ca
p
tu
r
e
th
e
d
ep
en
d
en
cy
o
f
d
if
f
er
en
t
tim
e
s
ca
les
[
3
4
]
.
Similar
ly
,
to
th
e
B
iLST
M,
GR
U
co
n
tain
s
a
f
o
r
war
d
U
GR
wh
ich
r
ea
d
s
th
e
s
ig
n
al
f
r
o
m
1
i
w
to
i
T
i
w
,
an
d
a
b
ac
k
war
d
U
GR
f
r
o
m
i
T
i
w
to
1
i
w
.
I
t is ca
lled
th
e
b
id
ir
ec
ti
o
n
al
GR
U
(
B
iGR
U)
:
]
,
1
[
),
(
'
i
it
it
T
t
w
U
GR
w
=
(
4
)
]
1
[
),
(
,
'
i
it
it
T
t
w
U
GR
w
=
(
5
)
Alth
o
u
g
h
t
h
e
GR
U
is
s
till
r
elat
iv
ely
n
ew
an
d
is
n
o
t
c
o
m
m
o
n
ly
u
s
ed
co
m
p
ar
e
t
o
L
STM
,
s
o
m
e
p
r
ev
io
u
s
liter
atu
r
e
o
n
GR
Us
h
as
b
ee
n
e
x
p
lo
r
e
d
f
o
r
th
e
E
C
G
class
if
icatio
n
task
[
3
5
]
.
T
h
e
p
h
ase
o
f
B
iGR
U
is
s
i
m
p
ler
th
an
B
iLST
M
d
u
e
to
th
e
r
ed
u
ctio
n
in
th
e
g
at
es
an
d
t
h
e
c
o
m
b
in
atio
n
o
f
t
h
e
f
o
r
g
et
a
n
d
in
p
u
t
g
ates
in
t
o
th
e
u
p
d
ate
g
ate.
T
h
e
U
n
i
d
i
r
e
c
t
i
o
n
a
l
P
h
a
s
e
(
l
e
f
t
-
t
o
-
r
i
g
h
t
)
B
i
d
i
r
e
c
t
i
o
n
a
l
P
h
a
s
e
(
l
e
f
t
-
to
-
r
i
g
h
t
a
n
d
r
i
g
h
t
-
to
-
l
e
f
t
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6
9
3
0
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
,
Vo
l.
19
,
No
.
3
,
J
u
n
e
2
0
2
1
:
90
2
-
91
0
906
u
p
d
ate
(
t
z
)
an
d
r
eset
(
t
r
)
g
ates
o
f
GR
U
in
th
e
f
o
r
war
d
p
ass
h
av
e
b
ee
n
m
e
n
tio
n
ed
in
(
8
)
an
d
(
9
)
.
T
h
en
,
th
e
h
id
d
en
s
tates c
an
b
e
s
ee
n
b
elo
w
;
h
t
t
h
t
h
t
b
r
s
W
x
U
h
+
•
+
=
−
)
(
t
a
n
h
(
1
(
6
)
1
)
1
(
−
•
+
•
−
=
t
t
t
t
t
s
z
h
z
s
(
7
)
wh
er
e
W
an
d
U
ar
e
th
e
a
d
d
itio
n
al
p
ar
am
eter
s
.
T
h
e
n
o
tatio
n
o
f
•
is
a
Had
am
ar
d
p
r
o
d
u
ct.
3.
RE
SU
L
T
S
AND
D
I
SCU
SS
I
O
N
T
h
is
s
tu
d
y
p
r
esen
ts
th
e
o
n
e
-
p
e
r
-
class
-
co
d
in
g
,
w
h
ich
is
an
ex
t
en
s
ib
le
alg
o
r
ith
m
f
o
r
th
e
b
in
ar
y
ca
s
e.
Fo
r
th
is
s
tu
d
y
,
th
e
t
o
tal
n
u
m
b
er
o
f
class
es
in
th
is
s
tu
d
y
was
f
iv
e
,
wh
ich
r
e
p
r
esen
ts
HC
,
MI
,
C
,
B
B
B
,
an
d
D.
T
h
e
ex
ten
s
ib
le
alg
o
r
ith
m
s
f
r
o
m
b
i
n
ar
y
ca
n
u
s
e
a
one
-
p
e
r
-
class
-
co
d
in
g
co
n
ce
p
t.
Fo
r
in
s
tan
ce
,
if
a
f
iv
e
-
class
ca
s
e
is
co
n
s
id
er
ed
,
th
e
o
u
tp
u
t
co
d
es
f
o
r
HC
,
MI
,
C
,
B
B
B
,
an
d
D
wer
e
1
0
0
0
0
,
0
1
0
0
0
,
0
0
1
0
0
,
0
0
0
1
0
,
an
d
0
0
0
0
1
,
r
esp
ec
tiv
ely
.
T
h
is
tech
n
iq
u
e
f
ix
es
a
p
r
o
b
lem
th
at
is
en
co
u
n
ter
ed
with
lab
el
en
co
d
in
g
wh
en
wo
r
k
in
g
with
ca
teg
o
r
ical
d
ata
.
E
ac
h
class
'
s
p
er
f
o
r
m
a
n
ce
is
b
ased
o
n
f
iv
e
m
etr
ics:
ac
cu
r
ac
y
,
s
en
s
itiv
ity
,
s
p
ec
if
icity
,
p
r
ec
is
io
n
,
an
d
F1
-
s
co
r
e.
Similar
to
th
e
p
r
ev
io
u
s
s
tu
d
y
[
2
0
]
,
th
e
u
n
id
ir
e
ctio
n
al
an
d
b
id
ir
ec
tio
n
al
m
u
lticlas
s
clas
s
if
icatio
n
p
r
o
ce
s
s
in
th
is
p
ap
er
was
d
i
v
id
ed
as
9
0
%
a
n
d
1
0
%
f
o
r
tr
ain
in
g
a
n
d
test
in
g
,
r
esp
ec
tiv
ely
.
Fo
r
co
m
p
u
tin
g
p
latf
o
r
m
s
,
th
is
s
tu
d
y
u
s
ed
a
GeFo
r
ce
R
T
X
2
0
8
0
g
r
ap
h
ics
p
r
o
ce
s
s
in
g
u
n
it
(
GPU)
.
T
h
e
o
p
er
atin
g
s
y
s
tem
was
W
in
d
o
ws
1
0
6
4
-
b
it.
T
h
e
h
y
p
e
r
p
ar
am
eter
s
wer
e
th
e
b
atch
s
ize
o
f
5
1
2
s
am
p
les,
s
o
f
tm
ax
f
u
n
ctio
n
in
th
e
o
u
tp
u
t
lay
er
,
an
d
th
e
to
tal
ep
o
ch
s
wer
e
1
0
0
.
Fo
r
th
e
u
n
id
ir
ec
tio
n
al
m
u
lticl
ass
class
if
icatio
n
p
r
o
ce
s
s
,
th
e
class
o
f
MI
p
r
o
d
u
ce
s
g
o
o
d
s
en
s
itiv
ity
,
p
r
ec
is
io
n
,
an
d
F1
-
s
co
r
e,
h
ig
h
er
th
an
th
o
s
e
o
f
th
e
o
th
er
class
es
in
u
n
id
ir
ec
tio
n
al
R
NNs.
T
h
e
r
esu
lts
ar
e
p
r
esen
ted
in
T
ab
le
2
.
T
h
is
m
ig
h
t
b
e
d
u
e
t
o
th
e
q
u
a
n
tity
o
f
HC
,
C
,
B
B
B
,
an
d
D
d
ata
a
v
ailab
le,
wh
ic
h
w
as
less
th
an
th
e
MI
o
f
th
e
to
tal
d
ata.
T
h
e
im
b
alan
c
ed
d
is
tr
ib
u
tio
n
ca
u
s
es th
e
m
aj
o
r
ity
class
to
ac
h
ie
v
e
h
ig
h
er
p
er
f
o
r
m
a
n
ce
.
Fro
m
a
to
tal
o
f
f
iv
e
class
es,
u
n
id
ir
ec
ti
o
n
al
R
NNs
h
av
e
an
av
er
ag
e
o
f
s
en
s
itiv
ity
,
p
r
ec
is
io
n
,
an
d
F1
-
s
co
r
e
o
f
9
0
.
2
5
%
,
8
5
%,
an
d
8
7
.
2
7
%,
r
esp
ec
tiv
el
y
.
T
o
p
r
o
v
e
th
e
g
r
a
d
ien
t
p
r
o
b
lem
s
in
s
tan
d
ar
d
R
NNs
ar
ch
ite
ctu
r
e,
th
e
f
in
e
-
tu
n
in
g
o
f
u
n
id
ir
ec
tio
n
al
L
STM
an
d
G
R
U
ar
ch
itectu
r
e
was
p
r
o
p
o
s
ed
.
I
n
L
STM
co
m
p
u
tatio
n
,
th
e
av
er
ag
e
o
f
s
en
s
itiv
ity
,
p
r
ec
is
io
n
,
an
d
F1
-
s
co
r
e
in
cr
ea
s
ed
to
9
4
.
1
7
%,
8
9
.
5
8
%,
an
d
9
1
.
7
2
%,
r
esp
ec
tiv
ely
.
I
n
an
o
th
e
r
GR
U
co
m
p
u
tatio
n
,
th
e
p
er
f
o
r
m
an
ce
s
h
o
wed
g
r
ea
ter
in
cr
ea
s
es,
with
an
av
er
a
g
e
o
f
9
5
.
5
4
%,
8
9
.
9
3
%,
a
n
d
9
2
.
3
1
%,
r
esp
ec
tiv
ely
.
T
ab
le
2
.
Un
id
i
r
ec
tio
n
al
R
NNs,
L
STM
,
an
d
GR
U
in
th
e
test
in
g
s
et
M
o
d
e
l
M
e
t
r
i
c
s
C
a
r
d
i
a
c
D
i
s
o
r
d
e
r
s
C
l
a
ss (%)
M
e
a
n
V
a
l
u
e
(
%)
HC
MI
C
B
B
B
D
A
c
c
u
r
a
c
y
9
6
.
7
7
9
5
.
4
8
9
9
.
1
4
9
8
.
9
2
9
8
.
7
8
9
7
.
8
1
S
e
n
s
i
t
i
v
i
t
y
8
9
.
3
4
9
6
.
3
0
9
4
.
2
3
8
5
.
7
1
8
5
.
7
1
9
0
.
2
5
R
N
N
s
S
p
e
s
i
f
i
c
i
t
y
9
7
.
9
9
9
2
.
9
4
9
9
.
3
3
9
9
.
5
5
9
9
.
0
5
9
7
.
7
7
P
r
e
c
i
s
i
o
n
8
8
.
0
0
9
7
.
6
9
8
4
.
4
8
9
0
.
0
0
6
4
.
8
6
8
5
.
0
0
F1
-
S
c
o
r
e
8
8
.
6
6
9
6
.
9
9
8
9
.
0
9
8
7
.
8
0
7
3
.
8
5
8
7
.
2
7
A
c
c
u
r
a
c
y
9
7
.
2
0
9
6
.
6
3
9
9
.
2
8
9
9
.
6
4
9
9
.
2
1
9
8
.
3
9
S
e
n
s
i
t
i
v
i
t
y
9
2
.
1
5
9
6
.
8
8
9
6
.
1
5
9
5
.
0
8
9
0
.
6
2
9
4
.
1
7
LSTM
S
p
e
s
i
f
i
c
i
t
y
9
8
.
0
0
9
5
.
8
3
9
9
.
4
0
9
9
.
8
5
9
9
.
4
1
9
8
.
4
9
P
r
e
c
i
s
i
o
n
8
8
.
0
0
9
8
.
6
5
8
6
.
2
1
9
6
.
6
7
7
8
.
3
8
8
9
.
5
8
F1
-
S
c
o
r
e
9
0
.
0
3
9
7
.
7
6
9
0
.
9
1
9
5
.
8
7
8
4
.
0
6
9
1
.
7
2
A
c
c
u
r
a
c
y
9
7
.
4
9
9
6
.
7
7
9
9
.
5
0
9
9
.
5
0
9
9
.
2
8
9
8
.
5
0
S
e
n
s
i
t
i
v
i
t
y
9
0
.
6
4
9
7
.
4
2
9
4
.
7
4
9
2
.
9
2
1
0
0
.
0
9
5
.
5
4
G
R
U
S
p
e
s
i
f
i
c
i
t
y
9
8
.
6
6
9
4
.
8
0
9
9
.
7
0
9
9
.
7
0
9
9
.
2
7
9
8
.
4
2
P
r
e
c
i
s
i
o
n
9
2
.
0
0
9
8
.
2
7
9
3
.
1
0
9
3
.
3
3
7
2
.
9
7
8
9
.
9
3
F1
-
S
c
o
r
e
9
1
.
3
2
9
7
.
8
4
9
3
.
9
1
9
4
.
1
2
8
4
.
3
7
9
2
.
3
1
I
n
a
b
id
ir
ec
tio
n
al
ca
s
e,
th
e
L
ST
M
an
d
GR
U
s
till
o
u
tp
er
f
o
r
m
ed
R
NNs
an
d
ac
h
iev
ed
o
u
ts
tan
d
i
n
g
r
esu
lts
,
as
p
r
esen
ted
in
T
a
b
le
3
,
ev
en
th
o
u
g
h
th
e
d
if
f
e
r
en
t
o
u
tco
m
es
wer
e
n
o
t
s
ig
n
if
ican
t.
T
h
e
b
id
ir
ec
tio
n
al
L
STM
r
esu
lt,
p
r
esen
ted
in
T
ab
le
3
,
o
b
tain
ed
av
e
r
ag
es
f
o
r
s
en
s
itiv
ity
,
p
r
ec
is
io
n
,
an
d
F1
-
s
co
r
e
o
f
9
4
.
1
6
%,
8
7
.
2
1
%,
an
d
9
0
.
3
3
%,
r
esp
ec
tiv
ely
.
Ad
d
itio
n
ally
,
b
id
i
r
ec
tio
n
al
GR
U
s
h
o
w
ed
av
er
a
g
es
o
f
s
en
s
itiv
ity
,
p
r
ec
is
io
n
,
an
d
F1
-
s
co
r
e
as
9
3
.
7
1
%,
8
8
.
7
8
%,
a
n
d
9
0
.
9
6
%,
r
esp
ec
tiv
ely
.
First,
f
o
r
R
NNs
in
th
e
u
n
id
ir
ec
tio
n
al
s
eq
u
en
ce
m
o
d
el,
th
e
av
er
ag
e
v
alu
es
o
f
ac
c
u
r
ac
y
an
d
s
p
ec
if
icity
in
f
iv
e
class
es
o
f
C
Ds
wer
e
h
ig
h
e
r
th
a
n
th
o
s
e
f
o
r
th
e
b
id
ir
ec
tio
n
al
p
ass
,
as
p
r
esen
ted
in
T
a
b
le
s
2
a
n
d
3
.
T
h
e
av
er
a
g
es
o
f
ac
cu
r
ac
y
an
d
s
p
ec
if
icity
f
o
r
u
n
id
ir
e
ctio
n
al
a
n
d
b
id
ir
ec
tio
n
al
wer
e
9
7
.
8
1
%
an
d
9
7
.
7
7
%,
an
d
9
7
.
3
3
%
an
d
9
6
.
6
6
%,
r
esp
ec
tiv
ely
.
T
h
e
a
v
er
a
g
es
o
f
ac
cu
r
ac
y
a
n
d
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
Un
id
ir
ec
tio
n
a
l
-
b
id
ir
ec
tio
n
a
l r
ec
u
r
r
en
t n
etw
o
r
ks fo
r
ca
r
d
ia
c
d
is
o
r
d
ers
cla
s
s
ifica
tio
n
(
A
n
n
is
a
Da
r
ma
w
a
h
yu
n
i
)
907
s
p
ec
if
icity
,
r
esp
ec
tiv
ely
,
wer
e
9
7
.
8
1
%
an
d
9
7
.
7
7
%
f
o
r
th
e
u
n
id
ir
ec
tio
n
al
an
d
9
7
.
3
3
%
a
n
d
9
6
.
6
6
%
f
o
r
th
e
b
id
ir
ec
tio
n
al
R
NNs.
I
n
co
n
t
r
ast,
th
e
ac
cu
r
ac
y
a
n
d
s
p
ec
if
icity
f
o
r
th
e
av
er
a
g
e
o
f
s
en
s
itiv
ity
,
p
r
ec
is
io
n
,
a
n
d
F1
-
s
co
r
e,
th
e
b
id
ir
ec
tio
n
al
R
NNs
o
b
tain
ed
b
etter
p
er
f
o
r
m
an
c
e
th
an
th
e
u
n
id
ir
ec
tio
n
al
R
NNs.
T
h
e
av
er
ag
es
o
f
s
en
s
itiv
ity
,
p
r
ec
is
io
n
,
an
d
F1
-
s
co
r
e
f
o
r
b
id
ir
ec
tio
n
al
R
NNs
ar
e
9
0
.
2
6
%,
8
8
.
2
1
%,
an
d
8
9
.
0
7
%.
Seco
n
d
,
d
if
f
er
en
t
f
r
o
m
R
NNs,
th
e
u
n
id
ir
ec
tio
n
a
l
L
STM
ac
h
iev
ed
b
etter
o
v
er
all
p
er
f
o
r
m
an
ce
s
th
an
a
b
id
ir
e
ctio
n
al
m
o
d
el.
T
h
e
av
er
ag
e
a
cc
u
r
ac
y
,
s
en
s
itiv
ity
,
s
p
ec
if
icity
,
p
r
ec
is
io
n
,
a
n
d
F
1
-
s
co
r
e
ac
h
iev
ed
was
9
8
.
3
9
%,
9
4
.
1
7
%,
9
8
.
4
9
%,
8
9
.
5
8
%,
an
d
9
1
.
7
2
%,
r
esp
ec
ti
v
ely
.
Ho
w
a
b
o
u
t
GR
U
as
th
e
l
ast
m
o
d
el?
T
h
e
u
n
id
ir
ec
tio
n
al
GR
U
s
h
o
ws
b
etter
p
er
f
o
r
m
an
ce
th
a
n
th
e
b
i
d
ir
ec
tio
n
al
GR
U,
s
im
ilar
to
L
STM
.
T
ab
le
3
.
B
id
ir
ec
tio
n
al
R
NNs,
L
STM
,
an
d
GR
U
in
th
e
test
in
g
s
et
M
o
d
e
l
M
e
t
r
i
c
s
C
a
r
d
i
a
c
D
i
s
o
r
d
e
r
s
C
l
a
ss (%)
M
e
a
n
V
a
l
u
e
(
%)
HC
MI
C
B
B
B
D
A
c
c
u
r
a
c
y
9
4
.
8
4
9
4
.
1
9
99
9
9
.
2
8
9
9
.
3
5
9
7
.
3
3
S
e
n
s
i
t
i
v
i
t
y
7
8
.
3
2
9
6
.
8
7
8
9
.
2
9
9
3
.
1
0
9
3
.
7
5
9
0
.
2
6
R
N
N
s
S
p
e
s
i
f
i
c
i
t
y
9
8
.
0
3
8
6
.
8
3
9
9
.
4
0
9
9
.
5
5
9
9
.
4
9
9
6
.
6
6
P
r
e
c
i
s
i
o
n
8
8
.
5
0
9
5
.
2
8
8
6
.
2
1
9
0
.
0
0
8
1
.
0
8
8
8
.
2
1
F1
-
S
c
o
r
e
8
3
.
1
0
9
6
.
0
7
8
7
.
7
2
9
1
.
5
3
8
6
.
9
6
8
9
.
0
7
A
c
c
u
r
a
c
y
9
7
.
2
0
9
6
.
0
5
9
9
.
2
8
9
9
.
2
8
9
9
.
1
4
9
8
.
1
9
S
e
n
s
i
t
i
v
i
t
y
9
1
.
7
1
9
6
.
3
3
9
8
.
0
0
9
1
.
6
7
9
3
.
1
0
9
4
.
1
6
LSTM
S
p
e
s
i
f
i
c
i
t
y
9
8
.
0
8
9
5
.
1
8
9
9
.
3
3
9
9
.
6
3
9
9
.
2
7
9
8
.
2
9
P
r
e
c
i
s
i
o
n
8
8
.
5
0
9
8
.
4
6
8
4
.
4
8
9
1
.
6
7
7
2
.
9
7
8
7
.
2
1
F1
-
S
c
o
r
e
9
0
.
0
8
9
7
.
3
8
9
0
.
7
4
9
1
.
6
7
8
1
.
8
2
9
0
.
3
3
A
c
c
u
r
a
c
y
9
7
.
7
0
9
6
.
2
7
9
9
.
2
1
9
9
.
2
8
9
9
.
2
1
9
8
.
3
3
S
e
n
s
i
t
i
v
i
t
y
9
3
.
3
0
9
6
.
7
8
9
6
.
0
8
8
9
.
0
6
9
3
.
3
3
9
3
.
7
1
G
R
U
S
p
e
s
i
f
i
c
i
t
y
9
8
.
4
2
9
4
.
6
9
9
9
.
3
3
9
9
.
7
7
9
9
.
3
4
9
8
.
3
1
P
r
e
c
i
s
i
o
n
9
0
.
5
0
9
8
.
2
7
8
4
.
4
8
9
5
.
0
0
7
5
.
6
8
8
8
.
7
8
F1
-
S
c
o
r
e
9
1
.
8
8
9
7
.
5
2
8
9
.
9
1
9
1
.
9
4
8
3
.
5
8
9
0
.
9
6
C
o
m
p
ar
ed
with
th
e
p
r
e
v
io
u
s
wo
r
k
[
2
0
]
in
ter
m
s
o
f
M
I
class
if
icatio
n
in
th
e
b
in
ar
y
ca
s
e,
u
n
id
ir
ec
tio
n
al
L
STM
p
er
f
o
r
m
ed
b
etter
th
an
GR
U.
I
n
co
n
tr
ast,
b
o
t
h
u
n
i
d
ir
ec
tio
n
al
an
d
b
id
ir
ec
tio
n
al
GR
U
was
b
etter
th
an
L
STM
o
v
er
all
r
eg
ar
d
in
g
th
e
m
u
lticlas
s
cla
s
s
if
ica
tio
n
ca
s
e.
G
R
U
i
s
co
m
p
u
tatio
n
ally
m
o
r
e
ef
f
icien
t
th
an
L
STM
,
d
u
e
to
h
av
i
n
g
o
n
l
y
two
g
ates
f
o
r
co
m
p
u
tin
g
.
I
n
tim
e
co
m
p
u
tatio
n
,
th
e
tr
ain
in
g
p
r
o
ce
s
s
o
f
GR
U
wa
s
f
aster
th
an
th
at
o
f
L
STM
an
d
d
id
n
o
t
h
av
e
m
u
ch
co
m
p
u
tatio
n
al
p
o
wer
.
No
tab
ly
,
in
th
is
s
tu
d
y
,
u
n
id
ir
ec
tio
n
al
L
STM
an
d
GR
U
d
o
n
o
t
h
a
v
e
d
if
f
er
en
t
tim
e
co
m
p
u
tatio
n
s
f
o
r
ea
ch
e
p
o
ch
'
s
tr
ain
in
g
.
Ho
wev
er
,
f
o
r
b
id
ir
e
ctio
n
al
L
STM
an
d
GR
U,
th
e
tim
e
eq
u
als
5
an
d
4
s
ec
o
n
d
s
,
r
esp
ec
tiv
ely
.
B
esid
es,
s
o
m
e
d
ee
p
lear
n
in
g
tec
h
n
iq
u
es
h
av
e
b
ee
n
im
p
lem
en
ted
.
Ho
wev
e
r
,
it
is
s
till
lim
ited
to
s
p
ec
if
ic
E
C
G
le
ad
s
with
f
ewe
r
class
es
an
d
o
n
ly
ap
p
licab
le
to
th
e
co
m
m
o
n
u
n
i
d
ir
ec
tio
n
al
m
o
d
el
.
Acc
o
r
d
i
n
g
to
th
e
c
o
m
p
a
r
is
o
n
b
etwe
en
u
n
id
ir
ec
ti
o
n
al
an
d
b
id
ir
ec
tio
n
al
GR
U,
th
is
s
tu
d
y
p
r
o
p
o
s
es
th
e
b
est
m
o
d
el,
i.e
.
,
u
n
id
i
r
ec
tio
n
al
GR
U.
T
h
e
u
n
i
d
ir
ec
tio
n
al
GR
U
p
er
f
o
r
m
an
ce
o
b
tain
ed
a
n
av
er
ag
e
a
cc
u
r
ac
y
,
s
en
s
itiv
ity
,
s
p
ec
if
icity
,
p
r
ec
is
io
n
,
an
d
F1
-
s
co
r
e
o
f
9
8
.
5
0
%,
9
5
.
5
4
%,
9
8
.
4
2
%,
8
9
.
9
3
%,
an
d
9
2
.
3
1
%,
r
esp
ec
tiv
ely
,
f
o
r
1
5
-
l
ea
d
s
o
f
E
C
G.
T
h
e
p
r
o
p
o
s
e
d
m
e
t
h
o
d
'
s
p
e
r
f
o
r
m
a
n
c
e
i
s
c
o
m
p
a
r
e
d
wi
t
h
t
h
e
p
r
e
v
i
o
u
s
l
i
t
e
r
at
u
r
e
i
n
r
e
c
e
n
t
y
e
a
r
s
,
as
i
l
l
u
s
t
r
a
t
e
d
i
n
T
a
b
l
e
4
.
T
r
i
p
at
h
y
e
t
a
l
.
[
3
6
]
p
r
o
p
o
s
e
d
t
h
e
l
e
a
s
t
s
q
u
a
r
e
S
V
M
w
it
h
a
n
a
c
c
u
r
a
c
y
o
f
8
9
.
9
3
%
,
9
3
.
9
5
%
,
9
3
.
0
3
%
,
9
0
.
0
9
%
,
a
n
d
8
5
.
2
9
%
f
o
r
H
C
,
M
I
,
C
,
D
,
a
n
d
h
y
p
e
r
t
r
o
p
h
y
.
T
h
e
a
v
e
r
a
g
e
a
c
c
u
r
a
c
y
is
9
0
.
3
4
%
i
n
a
l
l
c
l
as
s
es
.
I
n
a
n
o
t
h
e
r
l
i
t
e
r
a
t
u
r
e
,
A
c
h
a
r
y
a
et
a
l
.
[
3
7
]
i
d
e
n
t
i
f
i
e
d
t
h
r
e
e
c
l
as
s
es
o
f
N
o
r
m
a
l
,
M
I
,
a
n
d
o
t
h
e
r
C
Ds
,
b
a
s
e
d
o
n
d
i
s
c
r
et
e
c
o
s
i
n
e
t
r
a
n
s
f
o
r
m
(
DC
T
)
.
T
h
e
D
C
T
o
u
t
p
e
r
f
o
r
m
e
d
d
i
s
c
r
et
e
w
a
v
e
l
et
t
r
a
n
s
f
o
r
m
(
DW
T
)
an
d
e
m
p
i
r
i
c
a
l
m
o
d
e
d
e
c
o
m
p
o
s
i
t
i
o
n
(
E
M
D
)
.
T
h
e
a
c
c
u
r
a
c
y
,
s
e
n
s
i
t
i
v
i
t
y
,
a
n
d
s
p
e
c
i
f
i
ci
t
y
w
e
r
e
9
8
.
5
0
%
,
9
9
.
7
2
%
,
a
n
d
9
8
.
4
6
%,
r
e
s
p
e
c
t
i
v
el
y
.
I
n
t
h
e
s
a
m
e
y
ea
r
,
A
c
h
a
r
y
a
e
t
a
l
.
[
3
8
]
e
x
p
l
o
r
e
d
t
h
e
f
o
u
r
C
D
s
c
o
n
d
it
i
o
n
c
l
as
s
es
,
t
h
e
n
p
r
o
p
o
s
e
d
C
o
n
t
o
u
r
l
e
t
t
r
a
n
s
f
o
r
m
a
t
i
o
n
s
.
L
u
i
e
t
a
l
.
[
1
5
]
p
r
o
p
o
s
e
d
c
o
m
b
i
n
i
n
g
t
h
e
C
N
N
s
a
n
d
L
S
T
M
,
i
n
w
h
ic
h
t
h
e
r
e
c
u
r
r
e
n
t
l
a
y
e
r
i
n
c
r
e
a
s
e
d
t
h
e
s
e
n
s
i
t
i
v
it
y
t
o
c
l
a
s
s
i
f
ic
a
t
i
o
n
b
y
2
8
%
r
e
l
a
ti
v
e
t
o
t
h
e
C
NN
s
it
s
e
l
f
.
I
n
t
h
e
f
o
l
l
o
w
i
n
g
y
e
a
r
,
S
t
r
o
d
t
h
o
f
f
e
t
a
l
.
[
2
2
]
o
n
l
y
p
r
o
p
o
s
e
d
C
N
Ns
a
r
c
h
it
e
c
t
u
r
e
f
o
r
c
l
a
s
s
i
f
y
i
n
g
a
n
t
e
r
i
o
r
m
y
o
c
a
r
d
ia
l
i
n
f
a
r
c
t
i
o
n
(
a
M
I
)
,
i
n
f
e
r
i
o
r
m
y
o
c
a
r
d
i
a
l
i
n
f
a
r
c
t
i
o
n
,
(
i
M
I
)
,
a
n
d
h
e
a
l
t
h
y
c
o
n
t
r
o
l
.
U
s
in
g
1
0
-
f
o
l
d
c
r
o
s
s
-
v
a
l
i
d
a
ti
o
n
,
C
N
N
s
r
e
a
c
h
9
3
.
3
%
a
n
d
8
9
.
7
%
,
f
o
r
s
e
n
s
i
t
i
v
it
y
a
n
d
s
p
ec
if
i
c
i
t
y
,
r
es
p
e
c
ti
v
e
l
y
.
As
m
e
n
ti
o
n
e
d
e
a
r
l
i
e
r
,
f
r
o
m
t
h
e
b
e
n
c
h
m
a
r
k
s
t
u
d
y
w
it
h
t
h
e
s
a
m
e
d
a
t
a
s
et
i
m
p
l
e
m
e
n
t
at
i
o
n
,
o
u
r
p
r
o
p
o
s
e
d
u
n
i
d
i
r
e
c
t
i
o
n
a
l
GR
U
a
r
ch
i
t
e
c
t
u
r
e
c
a
n
i
m
p
r
o
v
e
9
8
.
5
%
ac
c
u
r
a
c
y
.
A
c
c
o
r
d
i
n
g
t
o
t
h
e
c
o
m
p
a
r
i
s
o
n
b
e
t
w
e
e
n
u
n
i
d
i
r
e
c
t
i
o
n
a
l
a
n
d
b
i
d
i
r
e
c
ti
o
n
a
l
,
th
i
s
s
t
u
d
y
p
r
o
p
o
s
e
s
t
h
e
b
e
s
t
m
o
d
e
l
,
u
n
i
d
i
r
e
c
t
i
o
n
a
l
G
R
U
.
T
h
e
p
e
r
f
o
r
m
a
n
c
e
o
f
u
n
i
d
i
r
e
c
t
i
o
n
a
l
G
R
U
o
b
t
a
i
n
e
d
a
n
a
v
e
r
a
g
e
A
C
C
,
S
E
N
,
S
P
E
,
P
R
E
,
a
n
d
F
1
-
s
c
o
r
e
o
f
9
8
.
5
0
%
,
9
5
.
5
4
%
,
9
8
.
4
2
%
,
8
9
.
9
3
%
,
a
n
d
9
2
.
3
1
%
,
r
e
s
p
e
c
t
i
v
e
l
y
,
f
o
r
1
5
-
l
e
a
d
s
o
f
E
C
G
.
I
n
c
o
n
t
r
a
s
t
,
h
i
t
ti
n
g
l
o
w
-
p
o
i
n
t
i
n
P
R
E
a
n
d
F
1
-
s
c
o
r
e
o
f
u
n
i
d
i
r
ec
t
i
o
n
a
l
R
N
N
l
e
d
t
o
a
p
e
r
f
o
r
m
an
c
e
o
f
8
5
%
a
n
d
8
7
.
2
7
%
,
r
e
s
p
ec
t
i
v
e
l
y
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6
9
3
0
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
,
Vo
l.
19
,
No
.
3
,
J
u
n
e
2
0
2
1
:
90
2
-
91
0
908
T
ab
le
4
.
B
en
ch
m
a
r
k
r
esu
lts
o
f
C
Ds cla
s
s
if
icatio
n
in
p
r
ev
io
u
s
liter
atu
r
e
A
u
t
h
o
r
s
M
e
t
h
o
d
Le
a
d
P
e
r
f
o
r
ma
n
c
e
(
%)
A
C
C
S
EN
SPE
P
R
E
F1
-
S
c
o
r
e
Tr
i
p
a
t
h
y
e
t
a
l
.
[
3
6
]
S
u
p
p
o
r
t
V
e
c
t
o
r
M
a
c
h
i
n
e
12
-
l
e
a
d
s
9
0
.
3
4
-
-
-
-
A
c
h
a
r
y
a
e
t
a
l
.
[
3
7
]
D
i
scret
e
C
o
s
i
n
e
Tr
a
n
sf
o
r
m
1
-
l
e
a
d
(
Le
a
d
I
I
)
9
8
.
5
0
9
9
.
7
2
9
8
.
4
6
-
-
A
c
h
a
r
y
a
e
t
a
l
.
[
3
8
]
C
o
n
t
o
u
r
l
e
t
t
r
a
n
sf
o
r
ma
t
i
o
n
s
1
-
l
e
a
d
(
Le
a
d
I
I
)
9
9
.
5
5
9
9
.
9
3
9
9
.
2
4
-
-
Lu
i
e
t
a
l
.
[
1
5
]
C
o
n
v
o
l
u
t
i
o
n
a
l
-
R
e
c
u
r
r
e
n
t
N
e
u
r
a
l
N
e
t
w
o
r
k
12
-
l
e
a
d
s
-
9
2
.
4
9
7
.
7
-
9
4
.
6
S
t
r
o
d
t
h
o
f
f
e
t
a
l
.
[
2
2
]
C
o
n
v
o
l
u
t
i
o
n
a
l
N
e
u
r
a
l
N
e
t
w
o
r
k
s
8
-
l
e
a
d
s
(
I
,
I
I
,
V
1
,
V
2
,
V
3
,
V
4
,
V
5
,
V
6
)
-
9
3
.
3
8
9
.
7
-
-
Pr
o
p
o
sed
m
e
t
h
o
d
U
n
i
d
i
r
e
c
t
i
o
n
a
l
GR
U
15
-
L
e
a
d
s
9
8
.
5
0
9
5
.
5
4
9
8
.
4
2
8
9
.
9
3
9
2
.
3
1
4.
CO
NCLU
SI
O
N
E
C
G
class
if
icatio
n
p
r
o
ce
s
s
h
a
s
th
e
g
r
ea
test
r
o
le
i
n
th
e
clin
i
ca
l
d
iag
n
o
s
is
o
f
C
Ds.
A
h
ig
h
n
u
m
b
er
o
f
E
C
G
s
ig
n
al
cla
s
s
e
s
m
ay
h
av
e
d
if
f
er
en
t
s
lo
p
es
o
f
s
ig
n
al,
tim
in
g
,
an
d
am
p
litu
d
e
,
wh
ich
ch
an
g
e
th
e
E
C
G
wav
ef
o
r
m
s
.
E
C
G
m
u
lticlas
s
class
if
icatio
n
is
n
o
t
as
s
im
p
l
e
as
b
in
ar
y
ca
s
e
p
r
o
b
lem
s
.
Sev
er
al
co
n
v
en
tio
n
al
m
ac
h
in
e
lear
n
in
g
tech
n
iq
u
es
h
av
e
b
ee
n
p
r
o
p
o
s
ed
in
p
r
e
v
io
u
s
liter
atu
r
e
f
o
r
s
p
ec
if
ic
m
u
lticla
s
s
ca
s
e
s
.
Ho
wev
er
,
th
e
m
ain
d
r
awb
ac
k
o
f
m
ac
h
in
e
lear
n
in
g
is
s
till
ab
o
u
t
f
e
atu
r
e
d
ef
in
itio
n
an
d
r
ep
r
esen
tatio
n
.
T
h
is
p
ap
er
s
u
cc
ess
f
u
lly
d
esig
n
s
a
s
u
p
er
v
is
ed
d
ee
p
lear
n
in
g
m
o
d
el
b
y
co
m
p
ar
in
g
u
n
id
ir
ec
tio
n
al
an
d
b
id
ir
ec
tio
n
al
r
ec
u
r
r
en
t
n
etwo
r
k
alg
o
r
ith
m
s
f
o
r
m
u
lticlas
s
ca
s
es
to
f
ig
u
r
e
o
u
t
th
e
b
est
m
o
d
el
p
er
f
o
r
m
an
ce
.
Fr
o
m
th
is
s
tu
d
y
,
th
e
u
n
id
ir
ec
tio
n
al
an
d
b
id
ir
ec
tio
n
a
l
p
h
ase
ar
e
co
m
p
ar
ed
with
d
ee
p
lear
n
in
g
f
o
r
th
e
C
Ds
m
u
lticl
ass
cla
s
s
if
icatio
n
in
E
C
G
1
5
-
lead
s
.
Desp
ite
a
lar
g
e
in
cr
ea
s
e
in
th
e
n
u
m
b
e
r
o
f
p
ar
a
m
eter
s
,
th
e
f
u
tu
r
e
in
p
u
ts
o
f
u
n
id
ir
ec
tio
n
al
ca
n
s
till
b
e
ac
ce
s
s
ed
b
y
win
d
o
win
g
,
lo
o
k
in
g
ah
ea
d
,
o
r
d
elay
in
g
o
u
tp
u
t.
Yet,
in
s
o
m
e
ca
s
es,
th
e
b
id
ir
ec
tio
n
al
n
etwo
r
k
s
o
u
tp
er
f
o
r
m
u
n
id
ir
ec
tio
n
al
o
n
es,
m
u
ch
f
aster
an
d
m
o
r
e
a
cc
u
r
ate
th
an
s
tan
d
ar
d
r
ec
u
r
r
en
t
n
etwo
r
k
s
an
d
tim
e
-
win
d
o
wed
m
u
ltil
ay
er
p
e
r
ce
p
tr
o
n
s
(
ML
Ps
)
.
Ov
er
all,
b
o
t
h
u
n
id
i
r
ec
tio
n
al
an
d
b
id
i
r
ec
tio
n
al
GR
U
is
b
etter
th
an
L
STM
in
m
u
lticlas
s
ca
s
e
s
.
B
es
id
es,
u
n
id
ir
ec
tio
n
al
GR
U
s
h
o
ws th
e
b
est p
er
f
o
r
m
an
ce
m
o
d
el
an
d
ac
h
iev
es
an
av
er
ag
e
ac
cu
r
ac
y
,
s
en
s
itiv
ity
,
s
p
ec
if
icity
,
p
r
ec
is
io
n
,
an
d
F
1
-
s
co
r
e
o
f
9
8
.
5
0
%,
9
5
.
5
4
%,
9
8
.
4
2
%,
8
9
.
9
3
%,
an
d
9
2
.
3
1
%,
r
esp
ec
tiv
ely
.
RE
F
E
R
E
NC
E
S
[1
]
B.
P
e
n
e
li
ti
a
n
a
n
d
P
.
Ke
se
h
a
tan
,
“
Th
e
m
a
in
p
o
in
ts
o
f
t
h
e
2
0
1
3
Ris
k
e
sd
a
s
In
d
o
n
e
sia
(In
Ba
h
a
sa
:
P
o
k
o
k
-
p
o
k
o
k
h
a
sil
Risk
e
sd
a
s In
d
o
n
e
sia
tah
u
n
2
0
1
3
)
,
”
J
a
k
a
rta
L
e
mb
.
Pen
e
rb
i
t
B
a
li
t
b
a
n
g
k
e
s
,
2
0
1
4
.
[2
]
T.
G
a
z
ian
o
,
K.
S
.
Re
d
d
y
,
F
.
P
a
c
c
a
u
d
,
S
.
Ho
rt
o
n
,
a
n
d
V.
C
h
a
tu
r
v
e
d
i,
“
Ca
rd
i
o
v
a
sc
u
lar
d
ise
a
se
,
”
in
Dise
a
se
Co
n
tro
l
Prio
riti
e
s
in
De
v
e
lo
p
i
n
g
C
o
u
n
tri
e
s.
2
n
d
e
d
i
ti
o
n
,
T
h
e
In
tern
a
ti
o
n
a
l
Ba
n
k
fo
r
Re
c
o
n
str
u
c
ti
o
n
a
n
d
De
v
e
lo
p
m
e
n
t
/T
h
e
Wo
rld
Ba
n
k
,
2
0
0
6
.
[3
]
Y.
Ji,
S
.
Zh
a
n
g
,
a
n
d
W.
Xi
a
o
,
“
El
e
c
tro
c
a
rd
io
g
ra
m
Clas
sifica
ti
o
n
Ba
se
d
o
n
F
a
ste
r
Re
g
io
n
s
with
Co
n
v
o
lu
ti
o
n
a
l
Ne
u
ra
l
Ne
two
rk
,
”
S
e
n
s
o
rs
,
v
o
l.
1
9
,
n
o
.
1
1
,
2
0
1
9
.
[4
]
F
.
A.
El
h
a
j,
N.
S
a
li
m
,
A.
R.
Ha
r
ris,
T.
T
.
S
we
e
,
a
n
d
T.
Ah
m
e
d
,
“
Arrh
y
t
h
m
ia
re
c
o
g
n
i
ti
o
n
a
n
d
c
las
sifica
ti
o
n
u
sin
g
c
o
m
b
in
e
d
l
i
n
e
a
r
a
n
d
n
o
n
li
n
e
a
r
fe
a
tu
re
s
o
f
ECG
sig
n
a
ls,”
C
o
mp
u
t
.
M
e
th
o
d
s
Pro
g
ra
ms
Bi
o
me
d
.
,
v
o
l
.
1
2
7
,
p
p
.
5
2
-
6
3
,
2
0
1
6
.
[5
]
S
.
Nu
rm
a
in
i
e
t
a
l.
,
“
Ro
b
u
st
d
e
te
c
ti
o
n
o
f
a
tri
a
l
fib
r
il
latio
n
fr
o
m
sh
o
rt
-
term
e
lec
tro
c
a
rd
io
g
ra
m
u
sin
g
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
tw
o
rk
s,”
F
u
tu
r.
Ge
n
e
r.
Co
mp
u
t
.
S
y
st.
,
v
o
l.
1
1
3
,
p
p
.
3
0
4
-
3
1
7
,
2
0
2
0
.
[6
]
C.
C.
Ag
g
a
rwa
l,
"
Da
ta cla
ss
ifi
c
a
ti
o
n
:
a
l
g
o
ri
th
m
s a
n
d
a
p
p
li
c
a
ti
o
n
s
,"
CRC
p
re
ss
,
2
0
1
4
.
[7
]
S
.
H.
Ja
m
b
u
k
ia,
V.
K.
Da
b
h
i,
a
n
d
H.
B.
P
ra
jap
a
ti
,
“
Clas
sifica
ti
o
n
o
f
ECG
sig
n
a
ls
u
sin
g
m
a
c
h
in
e
lea
rn
i
n
g
tec
h
n
iq
u
e
s:
A
su
rv
e
y
,
”
in
2
0
1
5
In
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
Ad
v
a
n
c
e
s
in
Co
mp
u
ter
En
g
i
n
e
e
rin
g
a
n
d
A
p
p
li
c
a
ti
o
n
s
,
2
0
1
5
,
p
p
.
7
1
4
-
7
2
1
.
[8
]
Q.
Li
,
C.
Ra
jag
o
p
a
lan
,
a
n
d
G
.
D.
Cli
ffo
rd
,
“
A
m
a
c
h
in
e
lea
rn
in
g
a
p
p
ro
a
c
h
t
o
m
u
lt
i
-
lev
e
l
EC
G
sig
n
a
l
q
u
a
li
t
y
c
las
sifica
ti
o
n
,
”
Co
mp
u
t.
M
e
t
h
o
d
s
Pro
g
ra
ms
B
io
me
d
.
,
v
o
l.
1
1
7
,
n
o
.
3
,
p
p
.
4
3
5
-
4
4
7
,
2
0
1
4
.
[9
]
T.
Lan
g
,
F
.
F
lac
h
se
n
b
e
rg
,
U.
v
o
n
Lu
x
b
u
rg
,
a
n
d
M
.
Ra
re
y
,
“
F
e
a
sib
il
it
y
o
f
a
c
ti
v
e
m
a
c
h
in
e
lea
rn
in
g
fo
r
m
u
l
ti
c
las
s
c
o
m
p
o
u
n
d
c
las
sifica
ti
o
n
,
”
J
.
C
h
e
m.
In
f.
M
o
d
e
l.
,
v
o
l.
5
6
,
n
o
.
1
,
p
p
.
1
2
-
2
0
,
2
0
1
6
.
[1
0
]
M
.
Kr
o
p
f,
D.
Ha
y
n
,
a
n
d
G
.
S
c
h
re
i
e
r,
“
ECG
c
las
sifica
ti
o
n
b
a
s
e
d
o
n
t
ime
a
n
d
fre
q
u
e
n
c
y
d
o
m
a
in
fe
a
tu
re
s
u
sin
g
ra
n
d
o
m
fo
re
sts,”
in
2
0
1
7
C
o
mp
u
ti
n
g
i
n
C
a
rd
io
lo
g
y
(Cin
C)
,
2
0
1
7
,
p
p
.
1
–
4.
[1
1
]
Q.
Qin
,
J.
Li
,
L.
Z
h
a
n
g
,
Y.
Y
u
e
,
a
n
d
C
.
Li
u
,
“
Co
m
b
in
i
n
g
lo
w
-
d
ime
n
sio
n
a
l
wa
v
e
let
fe
a
tu
re
s
a
n
d
su
p
p
o
rt
v
e
c
to
r
m
a
c
h
in
e
fo
r
a
rrh
y
th
m
ia
b
e
a
t
c
las
sifica
ti
o
n
,
”
S
c
i.
Re
p
.
,
v
o
l
.
7
,
n
o
.
1
,
2
0
1
7
.
[1
2
]
B.
C.
S
.
Lo
h
a
n
d
P
.
H.
H.
T
h
e
n
,
“
De
e
p
lea
rn
in
g
fo
r
c
a
rd
iac
c
o
m
p
u
t
e
r
-
a
id
e
d
d
iag
n
o
sis: b
e
n
e
fi
ts,
issu
e
s &
so
lu
ti
o
n
s,
”
M
h
e
a
l
th
,
v
o
l.
3
,
n
o
.
1
0
,
2
0
1
7
.
[1
3
]
S
.
P
a
rv
a
n
e
h
,
J.
Ru
b
in
,
A.
Ra
h
m
a
n
,
B.
Co
n
ro
y
,
a
n
d
S
.
Ba
b
a
e
iza
d
e
h
,
“
An
a
ly
z
i
n
g
si
n
g
le
-
lea
d
sh
o
rt
E
CG
re
c
o
rd
in
g
s
u
sin
g
d
e
n
se
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
s
a
n
d
fe
a
t
u
re
-
b
a
se
d
p
o
st
-
p
ro
c
e
ss
in
g
to
d
e
tec
t
a
tri
a
l
fib
r
il
l
a
ti
o
n
,
”
P
h
y
sio
l.
M
e
a
s.
,
v
o
l
.
3
9
,
n
o
.
8
,
2
0
1
8
.
[1
4
]
M.
-
L.
H
u
a
n
g
a
n
d
Y.
-
S
.
Wu
,
“
Cl
a
ss
ifi
c
a
ti
o
n
o
f
a
tri
a
l
fi
b
ril
lat
io
n
a
n
d
n
o
rm
a
l
sin
u
s
rh
y
th
m
b
a
se
d
o
n
c
o
n
v
o
lu
t
io
n
a
l
n
e
u
ra
l
n
e
tw
o
rk
,
”
B
io
me
d
.
En
g
.
L
e
tt
.
,
v
o
l.
1
8
,
p
p
.
1
8
3
-
1
9
3
,
2
0
2
0
.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
Un
id
ir
ec
tio
n
a
l
-
b
id
ir
ec
tio
n
a
l r
ec
u
r
r
en
t n
etw
o
r
ks fo
r
ca
r
d
ia
c
d
is
o
r
d
ers
cla
s
s
ifica
tio
n
(
A
n
n
is
a
Da
r
ma
w
a
h
yu
n
i
)
909
[1
5
]
H.
W.
L
u
i
a
n
d
K.
L.
Ch
o
w,
“
M
u
lt
i
c
las
s
c
las
sifica
ti
o
n
o
f
m
y
o
c
a
rd
ial
i
n
fa
rc
ti
o
n
wit
h
c
o
n
v
o
l
u
ti
o
n
a
l
a
n
d
r
e
c
u
rre
n
t
n
e
u
ra
l
n
e
two
rk
s f
o
r
p
o
rtab
le E
CG
d
e
v
ice
s,”
In
fo
rm
a
ti
c
s
M
e
d
.
U
n
lo
c
k
e
d
,
v
o
l.
1
3
,
p
p
.
2
6
-
3
3
,
2
0
1
8
.
[1
6
]
S
.
L
.
O
h
,
E.
Y.
K.
Ng
,
R.
S
a
n
Ta
n
,
a
n
d
U.
R
.
Ac
h
a
r
y
a
,
“
Au
t
o
m
a
ted
d
iag
n
o
sis
o
f
a
rrh
y
t
h
m
ia
u
sin
g
c
o
m
b
in
a
ti
o
n
o
f
CNN
a
n
d
LS
T
M
tec
h
n
iq
u
e
s with
v
a
riab
le l
e
n
g
th
h
e
a
rt
b
e
a
ts,”
Co
m
p
u
t.
Bi
o
l.
M
e
d
.
,
v
o
l.
1
0
2
,
p
p
.
2
7
8
-
2
8
7
,
2
0
1
8
.
[1
7
]
S
.
M
.
M
a
t
h
e
ws
,
C.
Ka
m
b
h
a
m
e
tt
u
,
a
n
d
K.
E.
Ba
rn
e
r,
“
A
n
o
v
e
l
a
p
p
li
c
a
ti
o
n
o
f
d
e
e
p
lea
rn
i
n
g
f
o
r
sin
g
le
-
lea
d
ECG
c
las
sifica
ti
o
n
,
”
Co
mp
u
t.
Bi
o
l
.
M
e
d
.
,
v
o
l
.
9
9
,
p
p
.
5
3
-
6
2
,
2
0
1
8
.
[1
8
]
C.
Xin
g
,
L
.
M
a
,
a
n
d
X.
Ya
n
g
,
“
S
tac
k
e
d
d
e
n
o
ise
a
u
to
e
n
c
o
d
e
r
b
a
se
d
fe
a
tu
re
e
x
trac
ti
o
n
a
n
d
c
l
a
ss
ifi
c
a
ti
o
n
fo
r
h
y
p
e
rs
p
e
c
tral
ima
g
e
s,”
J
.
S
e
n
so
rs
,
v
o
l.
2
0
1
6
,
p
p
.
1
-
1
0
,
2
0
1
6
.
[1
9
]
S
.
Nu
rm
a
i
n
i,
A.
G
a
n
i,
a
n
d
o
t
h
e
rs,
“
Ca
rd
iac
Arrh
y
th
m
ias
Clas
sifica
ti
o
n
Us
in
g
De
e
p
Ne
u
ra
l
Ne
two
r
k
s
a
n
d
P
ri
n
c
ip
le
Co
m
p
o
n
e
n
t
A
n
a
ly
sis Alg
o
rit
h
m
.
,
”
In
t.
J
.
Ad
v
.
S
o
f
t
Co
m
p
u
t.
Its
A
p
p
l.
,
v
o
l.
1
0
,
n
o
.
2
,
p
p
.
1
4
-
3
2
,
2
0
1
8
.
[2
0
]
A.
Da
rm
a
w
a
h
y
u
n
i,
e
t
a
l.
,
“
De
e
p
Lea
rn
in
g
with
a
Re
c
u
rre
n
t
N
e
two
rk
S
tr
u
c
tu
re
in
t
h
e
S
e
q
u
e
n
c
e
M
o
d
e
li
n
g
o
f
Im
b
a
lan
c
e
d
Da
ta fo
r
ECG
-
Rh
y
t
h
m
Clas
sifier,”
Al
g
o
rit
h
ms
,
v
o
l.
1
2
,
n
o
.
6
,
2
0
1
9
.
[2
1
]
S
.
Nu
rm
a
in
i
e
t
a
l.
,
“
De
e
p
Lea
rn
i
n
g
-
Ba
se
d
S
tac
k
e
d
De
n
o
isin
g
a
n
d
Au
to
e
n
c
o
d
e
r
f
o
r
ECG
He
a
rtb
e
a
t
Clas
sifica
ti
o
n
,
”
El
e
c
tro
n
ics
,
v
o
l.
9
,
n
o
.
1
,
p
.
1
-
18
,
2
0
2
0
.
[2
2
]
N.
S
tro
d
th
o
ff
a
n
d
C.
S
tr
o
d
t
h
o
ff
,
“
De
tec
ti
n
g
a
n
d
i
n
terp
re
ti
n
g
m
y
o
c
a
rd
ial
in
fa
rc
ti
o
n
u
sin
g
fu
ll
y
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
s,”
P
h
y
sio
l
.
M
e
a
s
.
,
v
o
l.
4
0
,
n
o
.
1
,
2
0
1
9
.
[2
3
]
A.
G
ra
v
e
s
a
n
d
J.
S
c
h
m
id
h
u
b
e
r,
“
F
ra
m
e
wise
p
h
o
n
e
m
e
c
las
sifica
ti
o
n
wit
h
b
id
irec
ti
o
n
a
l
LS
TM
a
n
d
o
th
e
r
n
e
u
ra
l
n
e
two
rk
a
rc
h
it
e
c
tu
re
s,”
Ne
u
ra
l
N
e
two
rk
s
,
v
o
l
.
1
8
,
n
o
.
5
-
6
,
p
p
.
6
0
2
-
6
1
0
,
2
0
0
5
.
[2
4
]
A.
Da
rm
a
wa
h
y
u
n
i
,
S
.
N
u
rm
a
in
i,
a
n
d
o
th
e
rs,
“
De
e
p
Lea
rn
in
g
wit
h
L
o
n
g
S
h
o
rt
-
Term
M
e
m
o
r
y
f
o
r
En
h
a
n
c
e
m
e
n
t
M
y
o
c
a
rd
ial
I
n
fa
rc
ti
o
n
Clas
sifica
ti
o
n
,
”
in
2
0
1
9
6
th
I
n
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
In
stru
me
n
t
a
ti
o
n
,
Co
n
tr
o
l,
a
n
d
Au
to
m
a
ti
o
n
(ICA)
,
2
0
1
9
,
p
p
.
1
9
-
2
3
.
[2
5
]
S
.
S
in
g
h
,
S
.
K.
P
a
n
d
e
y
,
U.
P
a
wa
r,
a
n
d
R
.
R.
Ja
n
g
h
e
l,
“
Clas
sifica
ti
o
n
o
f
ECG
Arrh
y
th
m
ia
u
sin
g
R
e
c
u
rre
n
t
Ne
u
ra
l
Ne
two
rk
s,”
Pro
c
e
d
i
a
Co
m
p
u
t
.
S
c
i
.
,
v
o
l.
1
3
2
,
p
p
.
1
2
9
0
-
1
2
9
7
,
2
0
1
8
.
[2
6
]
M
.
S
c
h
u
ste
r,
“
On
su
p
e
rv
ise
d
lea
rn
in
g
fr
o
m
se
q
u
e
n
ti
a
l
d
a
ta wi
th
a
p
p
li
c
a
ti
o
n
s fo
r
sp
e
e
c
h
re
c
o
g
n
it
i
o
n
,
”
Na
ra
I
n
st.
S
c
i.
T
e
c
h
n
o
l
.
(P
h
D Diss
.
,
1
9
9
9
.
[2
7
]
H.
S
a
leh
i
n
e
jad
,
S
.
S
a
n
k
a
r,
J.
Ba
rf
e
tt
,
E.
Co
lak
,
a
n
d
S
.
Va
lae
e
,
“
Re
c
e
n
t
a
d
v
a
n
c
e
s
in
re
c
u
rre
n
t
n
e
u
ra
l
n
e
two
rk
s,”
a
rXiv
Pre
p
r.
a
rXiv1
8
0
1
.
0
1
0
7
8
,
2
0
1
7
.
[2
8
]
Ö.
Yild
iri
m
,
“
A
n
o
v
e
l
wa
v
e
let
se
q
u
e
n
c
e
b
a
se
d
o
n
d
e
e
p
b
i
d
irec
ti
o
n
a
l
L
S
TM
n
e
two
rk
m
o
d
e
l
fo
r
EC
G
sig
n
a
l
c
las
sifica
ti
o
n
,
”
Co
mp
u
t.
Bi
o
l
.
M
e
d
.
,
v
o
l
.
9
6
,
p
p
.
1
8
9
-
2
0
2
,
2
0
1
8
.
[2
9
]
H.
Zen
a
n
d
H.
S
a
k
,
“
U
n
id
irec
ti
o
n
a
l
lo
n
g
sh
o
rt
-
term
m
e
m
o
ry
re
c
u
rr
e
n
t
n
e
u
ra
l
n
e
two
r
k
wit
h
re
c
u
rre
n
t
o
u
t
p
u
t
lay
e
r
f
o
r
lo
w
-
late
n
c
y
s
p
e
e
c
h
sy
n
th
e
sis,”
i
n
Aco
u
stics
,
S
p
e
e
c
h
a
n
d
S
ig
n
a
l
Pro
c
e
ss
in
g
(ICA
S
S
P),
2
0
1
5
IEE
E
In
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
,
2
0
1
5
,
p
p
.
4
4
7
0
-
4
4
7
4
.
[3
0
]
R.
Bo
u
ss
e
lj
o
t,
D.
Kre
ise
ler,
a
n
d
A.
S
c
h
n
a
b
e
l
,
“
Nu
tzu
n
g
d
e
r
EKG
-
S
ig
n
a
l
d
a
ten
b
a
n
k
CARD
IOD
AT
d
e
r
P
TB
ü
b
e
r
d
a
s
In
tern
e
t,
”
Bi
o
me
d
.
T
e
c
h
.
E
n
g
.
,
v
o
l.
4
0
,
n
o
.
s1
,
p
p
.
3
1
7
-
3
1
8
,
1
9
9
5
.
[3
1
]
A.
G
ra
v
e
s,
“
S
u
p
e
rv
ise
d
se
q
u
e
n
c
e
lab
e
ll
in
g
,
”
i
n
S
u
p
e
rv
ise
d
se
q
u
e
n
c
e
la
b
e
ll
in
g
wit
h
re
c
u
rr
e
n
t
n
e
u
ra
l
n
e
tw
o
rk
s
,
S
p
rin
g
e
r,
2
0
1
2
,
p
p
.
5
-
1
3
.
[3
2
]
Y.
Be
n
g
i
o
,
P
.
S
ima
rd
,
P
.
F
ra
sc
o
n
i
,
a
n
d
o
t
h
e
rs,
“
Lea
rn
in
g
l
o
n
g
-
term
d
e
p
e
n
d
e
n
c
ies
with
g
ra
d
ien
t
d
e
sc
e
n
t
is
d
i
fficu
l
t,
”
IEE
E
T
ra
n
s.
n
e
u
ra
l
n
e
two
rk
s
,
v
o
l
.
5
,
n
o
.
2
,
p
p
.
1
5
7
-
1
6
6
,
1
9
9
4
.
[3
3
]
M
.
S
c
h
u
ste
r
a
n
d
K.
K.
P
a
li
wa
l,
“
Bid
irec
ti
o
n
a
l
re
c
u
rre
n
t
n
e
u
ra
l
n
e
two
rk
s,”
IEE
E
T
ra
n
s.
S
ig
n
a
l
Pro
c
e
ss
.
,
v
o
l
.
4
5
,
n
o
.
1
1
,
p
p
.
2
6
7
3
-
2
6
8
1
,
1
9
9
7
.
[3
4
]
K.
Ch
o
,
B.
Va
n
M
e
rriën
b
o
e
r,
D.
B
a
h
d
a
n
a
u
,
a
n
d
Y.
Be
n
g
io
,
“
On
th
e
p
ro
p
e
rti
e
s
o
f
n
e
u
ra
l
m
a
c
h
in
e
tran
s
latio
n
:
E
n
c
o
d
e
r
-
d
e
c
o
d
e
r
a
p
p
ro
a
c
h
e
s,”
a
rXiv
Pre
p
r.
a
rXiv1
4
0
9
.
1
2
5
9
,
2
0
1
4
.
[3
5
]
H.
Jia
,
Y
.
De
n
g
,
P
.
Li
,
X
.
Qi
u
,
a
n
d
Y.
Tao
,
“
Re
se
a
rc
h
a
n
d
Re
a
li
z
a
ti
o
n
o
f
ECG
Clas
sifica
ti
o
n
b
a
se
d
o
n
G
a
ted
Re
c
u
r
re
n
t
Un
it
,
”
i
n
2
0
1
8
Ch
in
e
se
Au
t
o
ma
t
io
n
Co
n
g
re
ss
(CAC)
,
2
0
1
8
,
p
p
.
2
1
8
9
-
2
1
9
3
.
[3
6
]
R.
K.
Tri
p
a
th
y
,
L
.
N.
S
h
a
rm
a
,
a
n
d
S
.
Da
n
d
a
p
a
t
,
“
A
n
e
w
wa
y
o
f
q
u
a
n
ti
f
y
in
g
d
iag
n
o
stic
i
n
fo
rm
a
ti
o
n
fro
m
m
u
lt
il
e
a
d
e
lec
tro
c
a
rd
io
g
ra
m
fo
r
c
a
rd
iac
d
is
e
a
se
c
las
sifica
ti
o
n
,
”
He
a
lt
h
c
.
T
e
c
h
n
o
l.
L
e
tt
.
,
v
o
l
.
1
,
n
o
.
4
,
p
p
.
9
8
-
1
0
3
,
2
0
1
4
.
[3
7
]
U.
R.
Ac
h
a
ry
a
,
e
t
a
l.
,
“
Au
t
o
m
a
ted
c
h
a
ra
c
teriz
a
ti
o
n
a
n
d
c
las
sific
a
ti
o
n
o
f
c
o
r
o
n
a
ry
a
rtery
d
ise
a
se
a
n
d
m
y
o
c
a
rd
ial
in
fa
rc
ti
o
n
b
y
d
e
c
o
m
p
o
siti
o
n
o
f
E
CG
sig
n
a
ls:
A co
m
p
a
ra
ti
v
e
stu
d
y
,
”
In
f.
S
c
i.
(Ny
).
,
v
o
l.
3
7
7
,
p
p
.
1
7
-
2
9
,
2
0
1
7
.
[3
8
]
U.
R.
Ac
h
a
r
y
a
,
e
t
a
l.
,
“
Au
t
o
m
a
ted
c
h
a
ra
c
teriz
a
ti
o
n
o
f
c
o
r
o
n
a
r
y
a
rter
y
d
ise
a
se
,
m
y
o
c
a
rd
ial
in
fa
rc
ti
o
n
,
a
n
d
c
o
n
g
e
sti
v
e
h
e
a
rt
fa
il
u
re
u
sin
g
c
o
n
to
u
rlet
a
n
d
sh
e
a
rlet
tran
sfo
rm
s
o
f
e
lec
tro
c
a
rd
i
o
g
ra
m
sig
n
a
l,
”
K
n
o
wle
d
g
e
-
Ba
se
d
S
y
st.
,
v
o
l.
1
3
2
,
p
p
.
1
5
6
-
1
6
6
,
2
0
1
7
.
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
Ann
isa
Da
r
m
a
wa
h
y
u
n
i
is
c
u
rre
n
tl
y
a
lec
tu
re
r
a
n
d
re
se
a
rc
h
a
ss
istan
t
at
In
telli
g
e
n
t
S
y
ste
m
Re
se
a
rc
h
G
ro
u
p
,
F
a
c
u
lt
y
o
f
C
o
m
p
u
ter
S
c
ien
c
e
,
Un
i
v
e
rsitas
S
riwij
a
y
a
,
In
d
o
n
e
sia
.
He
r
re
se
a
rc
h
in
tere
st
in
c
lu
d
e
s
Bio
m
e
d
ica
l
S
ig
n
a
l
a
n
d
E
n
g
i
n
e
e
rin
g
,
De
e
p
Lea
rn
in
g
,
a
n
d
M
a
c
h
i
n
e
Lea
rn
in
g
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6
9
3
0
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
,
Vo
l.
19
,
No
.
3
,
J
u
n
e
2
0
2
1
:
90
2
-
91
0
910
S
iti
Nurm
a
in
i
is
c
u
rre
n
tl
y
a
p
r
o
fe
ss
o
r
in
t
h
e
F
a
c
u
lt
y
o
f
Co
m
p
u
ter
S
c
ien
c
e
,
Un
iv
e
rsita
s
S
riwij
a
y
a
.
S
h
e
wa
s
re
c
e
iv
e
d
h
e
r
M
a
ste
r'
s
d
e
g
re
e
in
Co
n
tr
o
l
sy
ste
m
,
In
stit
u
t
Te
k
n
o
lo
g
i
Ba
n
d
u
n
g
–
In
d
o
n
e
sia
(IT
B),
in
1
9
9
8
,
a
n
d
th
e
P
h
D
d
e
g
re
e
in
Co
m
p
u
ter
S
c
ien
c
e
,
Un
i
v
e
rsiti
Tek
n
o
lo
g
i
M
a
lay
sia
(UTM
),
a
t
2
0
1
1
.
He
r
re
se
a
rc
h
in
tere
st
,
in
c
lu
d
i
n
g
Bi
o
m
e
d
ica
l
En
g
i
n
e
e
rin
g
,
De
e
p
Lea
rn
in
g
,
M
a
c
h
in
e
Lea
rn
i
n
g
,
Im
a
g
e
P
r
o
c
e
ss
in
g
,
C
o
n
tr
o
l
S
y
ste
m
s,
a
n
d
R
o
b
o
ti
c
s.
Mu
h
a
m
m
a
d
Na
u
f
a
l
R
a
c
h
m
a
t
u
ll
a
h
is
c
u
rre
n
tl
y
a
lec
tu
re
r
a
n
d
re
se
a
rc
h
a
ss
istan
t
o
f
In
telli
g
e
n
t
S
y
ste
m
Re
se
a
rc
h
G
ro
u
p
,
F
a
c
u
lt
y
o
f
Co
m
p
u
ter
S
c
ien
c
e
,
Un
iv
e
rsitas
S
riwij
a
y
a
,
In
d
o
n
e
sia
.
His
re
se
a
rc
h
in
tere
s
t
in
c
l
u
d
e
s
M
e
d
ica
l
Im
a
g
in
g
,
Bio
m
e
d
ica
l
S
i
g
n
a
l
a
n
d
En
g
i
n
e
e
rin
g
,
De
e
p
Lea
rn
i
n
g
,
a
n
d
M
a
c
h
in
e
Lea
rn
i
n
g
.
Fi
r
d
a
u
s
Fi
r
d
a
u
s
i
s
c
u
rre
n
tl
y
is
a
lec
tu
re
r
a
n
d
re
se
a
rc
h
e
r
in
I
n
tel
li
g
e
n
t
S
y
ste
m
Re
se
a
rc
h
G
ro
u
p
,
F
a
c
u
lt
y
o
f
C
o
m
p
u
ter
S
c
ien
c
e
,
Un
iv
e
rsitas
S
riwij
a
y
a
,
I
n
d
o
n
e
sia
.
His res
e
a
rc
h
in
tere
st
in
c
lu
d
e
s T
e
x
t
p
ro
c
e
ss
in
g
,
De
e
p
L
e
a
rn
in
g
,
a
n
d
M
a
c
h
in
e
Lea
rn
in
g
.
Ba
m
b
a
n
g
Tu
tu
k
o
is
c
u
rre
n
tl
y
is
a
lec
tu
re
r
a
n
d
re
se
a
rc
h
e
r
in
In
te
ll
ig
e
n
t
S
y
ste
m
Re
se
a
rc
h
G
ro
u
p
,
F
a
c
u
lt
y
o
f
C
o
m
p
u
ter
S
c
ien
c
e
,
Un
iv
e
rsitas
S
riwij
a
y
a
,
I
n
d
o
n
e
sia
.
His res
e
a
rc
h
in
tere
st
in
c
lu
d
e
s Ro
b
o
ti
c
s,
De
e
p
Lea
rn
i
n
g
,
a
n
d
M
a
c
h
in
e
Lea
rn
in
g
.
Evaluation Warning : The document was created with Spire.PDF for Python.