I
A
E
S
I
n
t
e
r
n
at
io
n
al
Jou
r
n
al
of
A
r
t
if
ic
ia
l
I
n
t
e
ll
ig
e
n
c
e
(
I
J
-
AI
)
V
ol
.
14
, N
o.
3
,
J
une
2025
, pp.
2012
~
2025
I
S
S
N
:
2252
-
8938
,
D
O
I
:
10.11591/
ij
a
i.
v
14
.i
3
.pp
2012
-
2025
2012
Jou
r
n
al
h
om
e
page
:
ht
tp
:
//
ij
ai
.
ia
e
s
c
or
e
.c
om
E
n
se
m
b
l
e
m
od
e
l
-
b
ase
d
ar
r
h
yt
h
m
i
a
c
l
ass
i
f
i
c
at
i
on
w
i
t
h
l
oc
al
i
n
t
e
r
p
r
e
t
ab
l
e
m
od
e
l
-
agn
ost
i
c
e
xp
l
an
at
i
on
s
M
d
.
R
ab
iu
l
I
s
la
m
1
,
T
ap
an
K
u
m
ar
G
od
d
e
r
2
,
A
h
s
an
-
Ul
-
A
m
b
ia
3
,
F
e
r
d
ib
-
Al
-
I
s
la
m
4
,
A
n
in
d
ya
N
ag
4
,
B
u
lb
u
l
A
h
am
e
d
5
,
N
u
j
h
u
t
T
an
z
im
4
,
M
d
.
E
s
t
ia
k
A
h
m
e
d
6
1
D
e
pa
r
t
m
e
nt
of
B
i
om
e
di
c
a
l
E
ngi
ne
e
r
i
ng,
I
s
l
a
m
i
c
U
ni
ve
r
s
i
t
y,
K
us
ht
i
a
,
B
a
ngl
a
de
s
h
2
D
e
pa
r
t
m
e
nt
of
I
nf
or
m
a
t
i
on
a
nd
C
om
m
uni
c
a
t
i
on
T
e
c
hnol
ogy,
I
s
l
a
m
i
c
U
ni
ve
r
s
i
t
y,
K
us
ht
i
a
,
B
a
ngl
a
de
s
h
3
D
e
pa
r
t
m
e
nt
of
C
om
put
e
r
S
c
i
e
nc
e
a
nd
E
ngi
ne
e
r
i
ng,
I
s
l
a
m
i
c
U
ni
ve
r
s
i
t
y,
K
us
ht
i
a
,
B
a
ngl
a
de
s
h
4
D
e
pa
r
t
m
e
nt
of
C
om
put
e
r
S
c
i
e
nc
e
a
nd
E
ngi
ne
e
r
i
ng,
N
or
t
he
r
n
U
ni
ve
r
s
i
t
y
of
B
u
s
i
ne
s
s
a
nd
T
e
c
hnol
ogy
K
hul
na
,
K
hul
na
,
B
a
ngl
a
de
s
h
5
D
e
pa
r
t
m
e
nt
of
C
om
put
e
r
S
c
i
e
nc
e
a
nd
E
ngi
ne
e
r
i
ng,
S
ona
r
ga
on
U
ni
ve
r
s
i
t
y,
D
ha
ka
,
B
a
ngl
a
de
s
h
6
C
om
put
e
r
S
c
i
e
nc
e
a
nd
E
ngi
ne
e
r
i
ng
D
i
s
c
i
pl
i
ne
,
K
hul
na
U
ni
ve
r
s
i
t
y,
K
hul
na
,
B
a
ngl
a
de
s
h
A
r
t
ic
le
I
n
f
o
A
B
S
T
R
A
C
T
A
r
ti
c
le
h
is
to
r
y
:
R
e
c
e
iv
e
d
M
a
y 6, 2024
R
e
vi
s
e
d
D
e
c
22, 2024
A
c
c
e
pt
e
d
J
a
n 27, 2025
Arrhythmia
can
lead
to
he
art
failure,
stroke,
and
sudden
cardiac
arrest.
Prompt
diagnosis
of
arrhythmi
a
is
crucial
for
appropriat
e
treatment.
This
analysis
utilized
four
databases.
We
utilized
s
even
machine
learning
(ML)
algorit
hms
in
our
work.
These
algorit
hms
include
logistic
regression
(LR),
decision
tree
(DT),
extreme
gradient
boosting
(XGB),
k
-
nearest
nei
ghbo
rs
(KNN),
n
aïve
Bayes
(NB),
multilayer
perceptron
(MLP),
AdaBoost,
and
a
bagging
ensemble
of
these
approaches.
In
additio
n,
we
conduct
ed
an
analysis
on
a
stacking
ensemble
consist
ing
of
XGB
and
bagging
XG
B.
This
study
examines
various
arrhythmi
a
detection
techniques
using
both
a
single
base
dataset
and
a
composi
te
dataset.
The
objective
is
to
identify
the
o
ptimal
model
for
the
combined
dataset.
This
study
aims
to
evaluate
the
effic
acy
of
these
models
in
accurately
categorizin
g
normal
(N)
and
abnormal
(A)
heartbeats
as
binary
classes.
The
empirical
findings
demonstrated
t
hat
the
stacking
ensemble
approach
exhibit
ed
superior
accuracy
when
used
w
ith
the
combined
dataset.
Arrhythmia
classifi
cation
models
rely
on
this
as
a
crucial
component
.
The
binary
classifi
cation
achieved
an
accuracy
of
98.6
1%,
a
recall
of
97.66%,
and
a
precision
of
97.77%.
Subsequ
ently,
the
local
interpreta
ble
model
-
agnosti
c
explanati
ons
(LIME)
technique
is
emplo
yed
to
assess
the
prediction
capabilit
y
of
the
model.
K
e
y
w
o
r
d
s
:
A
r
r
hyt
hm
ia
E
le
c
tr
oc
a
r
di
ogr
a
m
E
xt
r
e
m
e
gr
a
di
e
nt
boos
ti
ng
L
I
M
E
M
a
c
hi
ne
le
a
r
ni
ng
This
is
an
open
access
article
under
the
CC
BY
-
SA
license.
C
or
r
e
s
pon
di
n
g
A
u
th
or
:
F
e
r
di
b
-
Al
-
I
s
la
m
D
e
pa
r
tm
e
nt
of
C
om
put
e
r
S
c
ie
nc
e
a
n
d
E
ngi
n
e
e
r
in
g
,
N
or
t
he
r
n
U
n
iv
e
r
s
it
y
of
B
us
in
e
s
s
a
n
d
T
e
c
hnol
og
y
K
h
ul
na
K
hul
na
-
9100,
B
a
ngl
a
de
s
h
E
m
a
il
:
f
e
r
di
b
.bs
m
r
s
tu
@
gm
a
il
.c
om
1.
I
N
T
R
O
D
U
C
T
I
O
N
A
r
r
hyt
hm
ia
s
,
w
hi
c
h
a
r
e
ir
r
e
gul
a
r
r
a
te
s
or
r
hyt
hm
s
of
he
a
r
tb
e
a
ts
,
r
e
ve
a
l
s
f
a
ul
ty
he
a
r
t
f
unc
ti
ons
.
S
e
ve
r
e
a
r
r
hyt
hm
ia
s
can
r
e
s
ul
t
in
in
s
uf
f
ic
ie
nt
bl
ood
c
ir
c
ul
a
ti
o
n,
le
a
di
ng
to
pot
e
nt
ia
l
ha
r
m
to
th
e
br
a
in
a
nd
he
a
r
t,
a
nd
in
s
om
e
c
ir
c
um
s
ta
nc
e
s
,
a
br
upt
c
a
r
di
a
c
de
a
th
.
H
e
nc
e
,
it
is
c
r
uc
ia
l
to
m
oni
to
r
c
a
r
di
a
c
a
c
ti
vi
ti
e
s
a
nd
id
e
nt
if
y
a
r
r
hyt
hm
ia
s
f
or
th
e
s
a
ke
of
in
di
vi
dua
ls
'
w
e
lf
a
r
e
.
A
r
r
hyt
hm
ia
de
te
c
ti
on
can
be
ut
il
iz
e
d
to
pr
om
pt
ly
id
e
nt
if
y
th
e
ons
e
t
of
he
a
r
t
di
s
e
a
s
e
,
e
xpe
di
te
th
e
a
dm
in
is
tr
a
ti
o
n
of
in
i
ti
a
l
m
e
di
c
a
l
a
s
s
is
ta
nc
e
,
a
nd
ul
ti
m
a
te
ly
pr
e
s
e
r
ve
hum
a
n
li
ve
s
.
A
r
r
hyt
hm
ia
s
,
c
ha
r
a
c
te
r
iz
e
d
by
a
bno
r
m
a
l
he
a
r
t
r
hyt
hm
,
a
r
e
a
pr
e
va
le
nt
c
a
r
di
a
c
c
ondi
ti
on
th
a
t
im
pa
c
ts
a
s
ig
ni
f
ic
a
nt
num
be
r
of
in
di
vi
du
a
ls
gl
oba
ll
y
a
nd
h
a
s
th
e
pot
e
nt
ia
l
to
be
li
f
e
-
th
r
e
a
te
ni
ng.
B
a
s
e
d
on
da
ta
pr
ovi
de
d
by
th
e
W
or
ld
H
e
a
lt
h
O
r
ga
ni
z
a
ti
on
(
W
H
O
)
,
c
a
r
di
ova
s
c
ul
a
r
di
s
or
de
r
s
s
uc
h
as
s
tr
oke
a
nd
he
a
r
t
f
a
il
ur
e
,
ha
ve
be
e
n
th
e
pr
im
a
r
y
c
a
us
e
s
of
de
a
th
w
or
ld
w
id
e
in
r
e
c
e
nt
ye
a
r
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
E
ns
e
m
bl
e
m
ode
l
-
bas
e
d a
r
r
hy
th
m
ia
c
la
s
s
if
ic
at
io
n w
it
h l
oc
al
i
nt
e
r
pr
e
ta
bl
e
…
(
M
d. R
abi
ul
I
s
la
m
)
2013
A
r
r
hyt
hm
ia
s
can
m
a
ni
f
e
s
t
as
a
s
ig
n
of
di
ve
r
s
e
unde
r
ly
in
g
il
ln
e
s
s
e
s
,
w
it
h
c
a
r
di
ova
s
c
ul
a
r
di
s
e
a
s
e
be
in
g
a
m
ong
th
e
m
.
To
r
e
duc
e
th
e
r
is
k
of
po
te
nt
ia
ll
y
f
a
ta
l
c
ons
e
que
nc
e
s
s
uc
h
as
s
tr
oke
,
c
or
ona
r
y
a
r
te
r
y
di
s
e
a
s
e
,
a
nd
s
udde
n
c
a
r
di
a
c
a
r
r
e
s
t
a
nd
to
a
voi
d
th
e
ne
e
d
f
or
f
ut
ur
e
in
tr
us
iv
e
a
nd
de
m
a
ndi
ng
th
e
r
a
pi
e
s
,
e
a
r
ly
di
a
gno
s
is
of
a
r
r
hyt
hm
ia
s
is
e
s
s
e
nt
ia
l
[
1]
.
T
he
e
le
c
tr
oc
a
r
di
ogr
a
m
(
E
C
G
)
e
xa
m
in
e
s
th
e
he
a
r
t'
s
e
le
c
tr
ic
a
l
a
c
ti
vi
ty
th
r
oughout
time
as
r
e
pe
a
ti
ng
s
ig
na
ls
;
th
is
m
a
ke
s
it
an
e
s
s
e
nt
ia
l
to
ol
f
or
m
oni
to
r
in
g
c
a
r
di
a
c
f
unc
ti
oni
ng
a
nd
de
te
c
ti
ng
a
bnor
m
a
l
he
a
r
t
r
hyt
hm
s
.
P
e
r
io
ds
r
e
pr
e
s
e
nt
th
e
c
ha
ngi
ng
pa
tt
e
r
ns
of
r
hyt
hm
ic
a
c
ti
vi
ty
w
it
hi
n
a
s
in
gl
e
he
a
r
tb
e
a
t.
A
r
r
hyt
hm
ia
id
e
nt
if
ic
a
ti
on
in
vol
ve
s
c
la
s
s
if
yi
ng
c
a
r
di
a
c
c
yc
l
e
s
as
e
it
he
r
no
r
m
a
l
or
a
bnor
m
a
l.
T
he
f
or
m
e
r
de
s
c
r
ib
e
s
th
e
he
a
r
t
in
its
nor
m
a
l
f
unc
ti
oni
ng,
w
he
r
e
a
s
th
e
la
tt
e
r
de
s
c
r
ib
e
s
a
bn
or
m
a
l
he
a
r
t
c
yc
le
s
th
a
t
m
a
y
c
a
us
e
ha
r
m
to
th
e
he
a
r
t
or
e
ve
n
c
a
r
di
a
c
a
r
r
e
s
t.
R
a
pi
d
de
te
c
ti
on
of
a
bnor
m
a
l
c
a
r
di
a
c
c
yc
le
s
u
s
in
g
E
C
G
da
ta
is
our
pr
im
a
r
y
goa
l.
M
a
c
hi
ne
le
a
r
ni
ng
(
ML
)
ha
s
de
m
ons
tr
a
te
d
its
e
f
f
ic
a
c
y
in
a
ut
om
a
ti
ng
a
nd
im
pr
ovi
ng
th
e
a
c
c
ur
a
c
y
of
a
r
r
hyt
hm
ia
di
a
gnos
is
in
th
e
he
a
lt
hc
a
r
e
in
dus
tr
y.
H
e
a
r
tb
e
a
t
s
c
a
n
be
c
la
s
s
if
ie
d
us
in
g
m
a
ny
a
ppr
oa
c
he
s
,
w
it
h
ML
m
ode
ls
be
in
g
one
of
th
e
m
.
C
la
s
s
if
ie
r
s
tr
a
in
m
ode
ls
to
r
e
li
a
bl
y
di
vi
de
he
a
r
tb
e
a
ts
in
to
f
iv
e
c
a
te
gor
ie
s
:
nor
m
a
l
(
N
)
,
s
upr
a
ve
nt
r
ic
ul
a
r
e
c
to
pi
c
(
S
V
E
B
)
,
ve
nt
r
ic
ul
a
r
e
c
to
pi
c
(
V
E
B
)
,
f
us
io
n
(
F
)
,
a
nd
unknown
(
Q
)
.
H
ow
e
ve
r
,
th
e
obj
e
c
ti
ve
of
th
is
pr
o
je
c
t
is
to
de
ve
lo
p
c
la
s
s
if
ie
r
s
th
a
t
m
a
y
c
r
e
a
te
m
ode
ls
c
a
pa
bl
e
of
c
a
te
gor
iz
in
g
he
a
r
tb
e
a
ts
in
to
two
di
s
ti
nc
t
c
la
s
s
e
s
:
n
or
m
a
l
(
0)
a
nd
a
bnor
m
a
l
or
a
r
r
hyt
hm
ia
(
1
)
.
T
he
ba
s
is
da
ta
s
e
t
ut
il
iz
e
d
in
th
is
w
or
k
c
om
pr
is
e
s
a
s
ub
s
ta
nt
ia
l
pr
opor
ti
on
of
th
e
n
or
m
a
l
he
a
r
tb
e
a
t
c
la
s
s
w
hi
le
e
xhi
bi
ti
ng
m
in
im
a
l
pr
opor
ti
ons
of
S
V
E
B
,
V
E
B
,
F,
a
nd
Q
c
la
s
s
e
s
.
C
om
bi
ne
th
e
S
V
E
B
,
V
E
B
,
a
nd
F
c
la
s
s
e
s
in
to
a
s
in
gl
e
c
la
s
s
r
e
pr
e
s
e
nt
in
g
a
be
r
r
a
nt
he
a
r
tb
e
a
t
or
a
r
r
hyt
hm
ia
.
F
or
th
e
pur
pos
e
of
e
nha
nc
in
g
m
ode
l
pe
r
f
or
m
a
nc
e
,
th
is
w
or
k
in
te
gr
a
te
s
f
our
da
ta
s
e
ts
.
T
he
d
a
ta
s
e
t
ha
d
a
li
m
it
e
d
num
be
r
of
e
x
a
m
pl
e
s
f
or
th
e
S
V
E
B
,
V
E
B
,
F,
a
nd
Q
c
la
s
s
e
s
.
T
he
m
ode
l
f
a
c
e
s
gr
e
a
te
r
di
f
f
ic
ul
ty
in
di
s
c
e
r
ni
ng
th
e
di
s
ti
nc
ti
v
e
c
ha
r
a
c
te
r
is
ti
c
s
a
nd
pa
tt
e
r
ns
of
a
c
la
s
s
w
he
n
th
e
r
e
a
r
e
li
m
it
e
d
in
s
ta
nc
e
s
of
it.
T
hi
s
c
oul
d
im
pe
de
th
e
m
ode
l'
s
a
bi
li
ty
to
ge
ne
r
a
li
z
e
e
f
f
e
c
ti
ve
ly
a
nd
a
c
c
ur
a
te
ly
c
la
s
s
if
y
nove
l
e
xa
m
pl
e
s
of
th
a
t
c
la
s
s
.
T
hi
s
w
or
k
c
ont
r
ib
ut
e
s
to
th
e
pr
e
vi
ous
li
te
r
a
tu
r
e
[
2]
on
E
C
G
c
la
s
s
if
ic
a
ti
on
by
ut
il
iz
in
g
a
M
L
m
e
th
od
a
nd
e
ns
e
m
bl
e
le
a
r
n
in
g
m
ode
l
to
e
nha
nc
e
a
c
c
ur
a
c
y.
To
do
th
is
,
in
f
or
m
a
ti
on
pe
r
ta
in
in
g
to
th
e
S
V
E
B
,
V
E
B
,
a
nd
F
c
a
te
gor
i
e
s
w
a
s
e
xt
r
a
c
te
d
f
r
om
s
e
ve
r
a
l
s
our
c
e
s
a
nd
in
c
or
por
a
te
d
in
to
th
e
or
ig
in
a
l
f
ounda
ti
ona
l
da
ta
s
e
t.
U
pon
m
e
r
gi
ng
f
r
e
s
h
da
ta
in
to
th
e
da
ta
s
e
t,
th
e
S
V
E
B
,
V
E
B
,
a
nd
F
m
ul
ti
pl
e
c
la
s
s
e
s
w
e
r
e
tr
a
ns
f
or
m
e
d
in
to
an
a
bnor
m
a
l
he
a
r
tb
e
a
t
c
la
s
s
or
a
r
r
hyt
hm
ia
c
la
s
s
,
w
hi
le
th
e
nor
m
a
l
c
la
s
s
r
e
m
a
in
e
d
una
lt
e
r
e
d.
T
he
a
bnor
m
a
l
he
a
r
tb
e
a
t
c
la
s
s
now
ha
s
a
gr
e
a
te
r
num
be
r
of
in
s
ta
nc
e
s
.
E
ns
e
m
bl
in
g
m
ul
ti
pl
e
m
ode
ls
in
s
te
a
d
of
r
e
ly
in
g
on
a
s
in
gl
e
ML
m
ode
l
of
f
e
r
s
s
e
ve
r
a
l
be
ne
f
it
s
.
F
ir
s
tl
y,
it
r
e
duc
e
s
th
e
r
is
k
of
ove
r
f
i
tt
in
g
by
c
om
bi
ni
ng
p
r
e
di
c
ti
ons
f
r
o
m
di
f
f
e
r
e
nt
m
ode
ls
,
le
a
di
ng
to
m
or
e
r
obus
t
a
nd
ge
ne
r
a
li
z
a
bl
e
r
e
s
ul
ts
.
S
e
c
ondl
y,
e
ns
e
m
bl
e
s
can
c
a
pt
ur
e
a
w
id
e
r
r
a
nge
of
pa
tt
e
r
ns
a
nd
r
e
la
ti
ons
hi
ps
in
th
e
da
ta
,
e
nha
nc
in
g
pr
e
di
c
ti
ve
pe
r
f
or
m
a
nc
e
.
L
a
s
tl
y,
e
n
s
e
m
bl
in
g
can
he
lp
m
it
ig
a
te
th
e
w
e
a
kne
s
s
e
s
of
in
di
vi
dua
l
m
ode
ls
,
r
e
s
ul
ti
ng
in
b
e
tt
e
r
ove
r
a
ll
pe
r
f
or
m
a
nc
e
.
U
s
in
g
a
c
om
bi
ne
d
da
ta
s
e
t
c
om
pos
e
d
of
f
iv
e
da
ta
s
e
ts
can
a
ls
o
im
pr
ove
m
ode
l
pe
r
f
o
r
m
a
nc
e
.
By
m
e
r
gi
ng
m
ul
ti
pl
e
da
ta
s
e
ts
,
we
can
in
c
r
e
a
s
e
th
e
di
ve
r
s
it
y
a
nd
r
ic
hne
s
s
of
th
e
da
ta
,
e
na
bl
in
g
th
e
m
ode
l
to
le
a
r
n
m
or
e
c
om
pr
e
h
e
ns
iv
e
p
a
tt
e
r
ns
a
nd
r
e
la
ti
on
s
hi
ps
.
T
hi
s
a
ppr
oa
c
h
can
le
a
d
to
be
tt
e
r
ge
ne
r
a
li
z
a
ti
on
to
uns
e
e
n
da
ta
a
nd
im
pr
ove
d
m
od
e
l
pe
r
f
or
m
a
nc
e
c
om
pa
r
e
d
to
u
s
in
g
a
s
in
gl
e
da
ta
s
e
t
a
lo
ne
.
In
s
um
m
a
r
y,
e
n
s
e
m
bl
in
g
m
ode
l
s
a
nd
c
om
bi
ni
ng
da
ta
s
e
t
s
a
r
e
e
f
f
e
c
ti
ve
s
tr
a
te
gi
e
s
to
e
nha
nc
e
pr
e
di
c
ti
ve
pe
r
f
or
m
a
nc
e
,
in
c
r
e
a
s
e
m
ode
l
r
obus
tn
e
s
s
,
a
nd
im
pr
ove
ge
ne
r
a
li
z
a
ti
on
a
bi
li
ty
.
M
ode
ls
can
r
e
a
di
ly
id
e
nt
if
y
th
e
di
s
ti
nc
ti
ve
c
ha
r
a
c
te
r
is
ti
c
s
a
nd
pa
tt
e
r
n
s
of
a
c
e
r
ta
in
c
la
s
s
,
e
na
bl
i
ng
th
e
m
to
pr
e
c
is
e
ly
c
la
s
s
if
y
ne
w
in
s
ta
nc
e
s
of
th
a
t
c
la
s
s
.
N
e
xt
,
th
e
s
ynt
h
e
ti
c
m
in
or
it
y
ove
r
s
a
m
pl
in
g
te
c
hni
que
(
S
M
O
T
E
)
w
il
l
be
e
m
pl
oye
d
to
ge
n
e
r
a
te
m
or
e
s
ynt
he
ti
c
da
ta
by
ove
r
s
a
m
pl
in
g
th
e
m
in
or
it
y
c
la
s
s
.
T
h
e
in
te
gr
a
ti
on
of
f
our
da
ta
ba
s
e
s
a
nd
th
e
im
pl
e
m
e
nt
a
ti
on
of
S
M
O
T
E
ha
s
r
e
s
ul
te
d
in
e
nh
a
nc
e
d
ove
r
a
ll
a
c
c
ur
a
c
y,
r
e
c
a
ll
,
a
n
d
pr
e
c
is
io
n
m
e
tr
ic
s
.
T
hr
oughout
th
e
ye
a
r
s
,
num
e
r
ous
r
e
s
e
a
r
c
h
e
r
s
ha
ve
e
m
pl
oye
d
v
a
r
io
us
m
e
th
odol
ogi
e
s
to
id
e
nt
if
y
th
e
pr
e
s
e
nc
e
of
a
r
r
hyt
hm
ia
di
s
e
a
s
e
.
T
h
e
f
in
is
he
d
w
or
ks
a
r
e
s
uc
c
in
c
tl
y
s
ta
te
d
in
th
e
f
ol
lo
w
in
g
te
r
m
s
.
A
m
ode
l
f
o
r
E
C
G
he
a
r
tb
e
a
t
c
la
s
s
if
ic
a
ti
on
pr
opos
e
d
by
Al
-
M
ous
a
et
al
.
[
1]
im
pr
ove
s
r
e
c
ol
le
c
ti
on
f
or
c
a
te
gor
ie
s
F
a
nd
Q
w
hi
le
m
a
in
ta
in
in
g
th
e
s
a
m
e
r
e
c
a
ll
f
or
th
e
ot
he
r
ki
nds
.
T
hi
s
a
n
a
ly
s
is
r
e
li
e
d
on
th
e
M
a
s
s
a
c
hus
e
tt
s
I
ns
ti
tu
te
of
T
e
c
hnol
ogy
-
B
e
th
I
s
r
a
e
l
H
os
pi
ta
l
a
r
r
hyt
hm
ia
da
ta
ba
s
e
(
M
I
T
-
B
I
H
)
s
upr
a
ve
nt
r
ic
ul
a
r
da
ta
ba
s
e
as
its
ba
s
ic
da
ta
s
e
t.
To
im
pr
ove
th
e
r
e
c
a
ll
f
or
th
e
F
a
nd
Q
c
la
s
s
e
s
,
th
e
a
u
th
or
s
c
om
bi
ne
d
da
ta
f
r
o
m
ot
he
r
da
ta
s
e
ts
a
nd
a
dde
d
it
to
th
e
f
unda
m
e
nt
a
l
da
t
a
s
e
t.
F
or
th
is
c
om
bi
ne
d
da
ta
s
e
t,
th
e
y
us
e
d
S
M
O
T
E
to
a
c
hi
e
v
e
ba
la
nc
e
.
T
h
e
r
a
ndom
f
or
e
s
t
(
R
F
)
a
lg
or
it
hm
out
pe
r
f
or
m
e
d
a
ll
ot
he
r
s
w
it
h
an
a
c
c
ur
a
c
y
r
a
te
of
97%
a
nd
r
e
c
a
ll
va
lu
e
s
of
93,
95,
95,
a
nd
30%
f
or
N,
S
V
E
B
,
V
E
B
,
F,
a
nd
Q,
r
e
s
pe
c
ti
ve
ly
.
S
a
ki
b
et
al
.
[
2]
di
s
c
us
s
e
d
th
e
di
f
f
ic
ul
ty
of
in
c
or
por
a
ti
ng
a
r
ti
f
ic
ia
l
in
te
ll
ig
e
nc
e
(
A
I
)
in
to
a
dva
nc
e
d
in
te
r
ne
t
of
th
in
gs
(
I
oT
)
s
e
n
s
or
s
to
de
te
c
t
ir
r
e
gul
a
r
he
a
r
t
r
hyt
hm
s
us
in
g
E
C
G
da
ta
.
T
he
a
ut
hor
s
la
id
f
or
th
a
m
e
th
od
f
or
a
r
r
hyt
hm
ia
c
la
s
s
if
ic
a
ti
on
vi
a
a
li
ght
w
e
ig
h
t
de
e
p
le
a
r
ni
ng
(
DL
)
s
tr
a
te
gy.
T
hi
s
te
c
hni
que
c
la
s
s
if
ie
d
f
our
di
s
ti
nc
t
ki
nds
of
he
a
r
tb
e
a
ts
us
in
g
a
1D
c
onvolut
io
na
l
ne
ur
a
l
ne
twor
k
(
C
N
N
)
de
s
ig
n.
F
our
s
e
pa
r
a
te
P
h
ys
io
N
e
t
da
ta
s
e
ts
w
e
r
e
us
e
d
f
or
th
is
e
va
lu
a
ti
on.
C
om
pa
r
e
d
to
w
e
ll
-
e
s
ta
bl
is
he
d
a
ppr
oa
c
he
s
li
ke
RF
,
k
-
ne
a
r
e
s
t
ne
ig
hbor
s
(
KNN
)
,
a
nd
de
l
a
y
nonl
in
e
a
r
e
qua
ti
on
-
ba
s
e
d
opt
im
iz
a
ti
on,
th
e
pr
opos
e
d
DL
s
tr
a
te
gy
p
e
r
f
or
m
e
d
be
tt
e
r
in
th
e
e
xpe
r
im
e
nt
s
.
W
he
n
e
m
pl
oye
d
on
vi
r
tu
a
li
z
e
d
m
ic
r
oc
ont
r
ol
le
r
s
c
onne
c
te
d
to
I
oT
s
e
ns
or
s
,
th
e
DL
m
ode
l
s
how
s
out
s
ta
ndi
ng
pe
r
f
or
m
a
nc
e
in
te
r
m
s
of
pr
oc
e
s
s
in
g
time
a
nd
m
e
m
or
y
us
e
.
W
he
n
te
s
te
d
on
th
e
M
I
T
-
B
I
H
s
upr
a
ve
nt
r
ic
ul
a
r
da
ta
s
e
t
(
94.12%
a
c
c
ur
a
c
y)
,
th
e
M
I
T
-
B
I
H
a
r
r
hyt
hm
ia
da
ta
s
e
t
(
94.97%
a
c
c
ur
a
c
y)
,
th
e
I
ns
ti
tu
te
of
C
a
r
di
ol
ogi
c
a
l
T
e
c
hni
c
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
. 14, No. 3, J
une
2025
:
2012
-
2025
2014
(
I
N
C
A
R
T
)
12
-
le
a
d
a
r
r
hyt
hm
ia
da
ta
s
e
t
(
94.97%
a
c
c
ur
a
c
y)
,
a
n
d
th
e
S
udde
n
C
a
r
di
a
c
D
e
a
th
H
ol
te
r
(
S
C
D
H
)
(
96.67%
a
c
c
ur
a
c
y)
,
th
e
pr
opos
e
d
m
ode
l
pe
r
f
or
m
s
w
e
ll
.
A
ne
w
m
e
th
od
f
or
id
e
nt
if
yi
ng
i
r
r
e
gul
a
r
he
a
r
tb
e
a
ts
w
a
s
pr
e
s
e
nt
e
d
by
W
a
ng
et
al
.
[
3]
th
a
t
m
a
ke
s
us
e
of
th
e
E
a
s
yE
ns
e
m
bl
e
a
lg
or
it
hm
in
c
onj
unc
ti
on
w
it
h
gl
oba
l
he
a
r
tb
e
a
t
da
ta
obt
a
in
e
d
f
r
om
th
e
M
I
T
-
B
I
H
a
r
r
hyt
hm
ia
da
ta
ba
s
e
.
In
a
ddi
ti
on
to
out
pe
r
f
or
m
in
g
th
e
c
om
pe
ti
ti
on
ove
r
a
ll
,
th
e
ir
s
ugge
s
te
d
s
tr
a
te
gy
boos
t
s
m
in
or
it
y
c
a
te
gor
y
pe
r
f
or
m
a
nc
e
w
hi
le
ke
e
pi
ng
m
a
jo
r
it
y
c
a
te
gor
y
pe
r
f
or
m
a
nc
e
hi
gh.
On
a
ve
r
a
ge
,
th
e
s
ugge
s
te
d
m
ode
l
a
c
hi
e
ve
d
a
r
e
c
a
ll
r
a
te
of
55.4%
f
or
ty
pe
F,
an
a
c
c
ur
a
c
y
of
91.7%
f
or
ty
pe
N,
89.9%
f
or
ty
pe
V
E
B
,
a
nd
87.8%
f
o
r
ty
pe
S
V
E
B
.
To
c
l
a
s
s
i
f
y
E
C
G
d
a
ta
,
K
h
a
n
et
al
.
[
4]
r
e
v
e
a
l
e
d
a
t
a
il
or
e
d
C
N
N
m
o
d
e
l
.
T
h
e
a
ut
ho
r
s
m
a
d
e
u
s
e
of
P
h
y
s
i
o
N
e
t'
s
M
I
T
-
B
I
H
a
r
r
h
y
th
m
i
a
d
a
t
a
b
a
s
e
.
W
it
h
an
a
v
e
r
a
g
e
r
e
c
a
l
l
of
95
.
40%
a
n
d
a
to
t
a
l
a
c
c
ur
a
c
y
of
9
5
.2
%
,
t
h
e
i
r
pr
o
po
s
e
d
m
o
d
e
l
s
u
c
c
e
s
s
f
u
ll
y
c
a
t
e
g
or
iz
e
s
a
r
r
h
yt
hm
i
a
.
A
m
od
e
l
t
h
a
t
c
h
o
o
s
e
s
th
e
be
s
t
s
u
b
s
e
t
s
of
f
e
a
t
ur
e
s
to
d
i
s
ti
ng
ui
s
h
one
c
la
s
s
f
r
o
m
a
n
ot
he
r
w
a
s
pr
e
s
e
nt
e
d
in
[
5]
.
A
s
u
pp
or
t
v
e
c
t
or
m
a
c
hi
n
e
(
S
V
M
)
b
in
a
r
y
c
l
a
s
s
i
f
i
e
r
is
u
s
e
d
to
a
c
c
o
m
p
li
s
h
t
hi
s
t
a
s
k
by
c
o
m
p
a
r
i
ng
t
w
o
s
e
t
s
of
d
a
t
a
one
a
f
t
e
r
th
e
o
th
e
r
.
u
ti
li
z
i
n
g
th
e
M
I
T
-
B
I
H
a
r
r
hy
th
m
i
a
d
a
t
a
b
a
s
e
,
th
e
p
r
o
p
o
s
e
d
f
e
a
t
ur
e
s
e
l
e
c
ti
o
n
m
e
t
ho
d
w
a
s
a
s
s
e
s
s
e
d
.
C
l
a
s
s
if
ic
a
ti
on
a
c
c
ur
a
c
y
is
8
6.
6
6%
on
a
v
e
r
a
g
e
w
h
e
n
u
s
in
g
th
e
pr
o
p
o
s
e
d
f
e
a
t
ur
e
s
e
l
e
c
t
io
n
m
e
t
h
od
.
F
o
r
th
e
c
l
a
s
s
e
s
N,
S
V
E
B
,
V
E
B
,
a
nd
F,
it
a
c
h
i
e
v
e
d
r
e
c
a
ll
r
a
te
s
of
8
8.
9
4,
79
.0
6
,
85
.
48
,
a
n
d
9
3.
81
%
,
r
e
s
p
e
c
ti
v
e
l
y.
In
t
h
e
ir
in
v
e
s
t
ig
a
t
io
n
,
A
la
r
s
a
n
a
n
d
Y
o
un
e
s
[
6]
u
s
e
d
a
c
o
m
bi
ne
d
d
a
t
a
b
a
s
e
of
th
e
M
I
T
-
B
I
H
s
up
r
a
v
e
n
tr
i
c
u
la
r
a
r
r
h
yt
h
m
i
a
a
n
d
th
e
M
I
T
-
B
I
H
a
r
r
h
yt
hm
i
a
d
a
t
a
b
a
s
e
s
,
a
n
d
t
he
y
u
s
e
d
t
hr
e
e
m
o
d
e
l
s
to
it
:
d
e
c
i
s
i
o
n
tr
e
e
(
DT
)
,
RF
,
a
nd
gr
a
di
e
nt
b
oo
s
ti
n
g
tr
e
e
s
.
U
s
in
g
RF
f
o
r
m
ul
ti
c
la
s
s
if
i
c
a
ti
o
n,
th
e
a
u
t
hor
s
a
c
h
i
e
v
e
d
a
hi
gh
a
c
c
ur
a
c
y
of
98
.0
3%
.
A
di
a
gn
o
s
t
i
c
m
o
d
e
l
d
e
v
e
l
op
e
d
f
o
r
th
e
pur
p
o
s
e
of
d
e
t
e
c
ti
n
g
c
a
r
di
a
c
a
r
r
h
yt
hm
i
a
w
a
s
pu
b
li
s
he
d
by
S
i
ng
h
a
n
d
S
i
ng
h
[
7
]
.
To
f
i
nd
t
h
e
m
o
s
t
im
po
r
t
a
nt
c
h
a
r
a
c
t
e
r
i
s
ti
c
s
,
th
i
s
s
tu
d
y
u
s
e
d
t
hr
e
e
f
i
lt
e
r
-
b
a
s
e
d
m
e
t
h
od
s
f
o
r
s
e
l
e
c
t
in
g
f
e
a
tu
r
e
s
.
To
t
e
s
t
h
ow
w
e
l
l
t
h
e
f
e
a
t
ur
e
s
e
l
e
c
ti
o
n
m
e
th
od
w
or
k
e
d
,
th
e
w
r
i
t
e
r
s
u
s
e
d
t
hr
e
e
s
e
p
a
r
a
t
e
m
o
de
l
s
:
J
R
i
p,
l
in
e
a
r
S
V
M
,
a
nd
R
F
.
U
s
i
n
g
a
ga
i
n
r
a
t
io
s
e
l
e
c
t
e
d
f
e
a
t
ur
e
s
tr
a
t
e
gy
w
it
h
a
s
e
l
e
c
te
d
gr
o
up
of
30
f
e
a
tu
r
e
s
a
n
d
a
RF
c
l
a
s
s
if
i
e
r
,
t
h
e
r
e
s
e
a
r
c
h
a
tt
a
in
e
d
a
m
a
x
im
um
a
c
c
ur
a
c
y
of
8
5.
58
%
.
A
bde
lm
one
e
m
et
al
.
[
8]
pr
e
s
e
nt
e
d
a
hi
ghl
y
e
f
f
e
c
ti
ve
a
lg
or
it
hm
f
or
de
te
c
ti
ng
c
a
r
di
a
c
a
r
r
hyt
hm
ia
.
T
he
y
e
xpl
or
e
d
di
f
f
e
r
e
nt
ove
r
s
a
m
pl
in
g
te
c
hni
que
s
to
a
ddr
e
s
s
th
e
is
s
ue
of
im
ba
la
nc
e
d
da
ta
s
e
ts
.
T
h
e
e
ns
e
m
bl
e
c
la
s
s
if
ie
r
,
S
V
M
,
a
nd
RF
w
it
h
r
a
ndom
s
a
m
pl
in
g
obt
a
in
e
d
a
r
e
m
a
r
ka
bl
e
a
c
c
ur
a
c
y
of
98.18%
.
T
hi
s
r
e
s
e
a
r
c
h
a
ls
o
in
tr
oduc
e
d
a
m
obi
le
s
ys
te
m
de
s
ig
n
th
a
t
in
c
or
por
a
te
s
an
a
lg
or
it
hm
f
or
di
a
gnos
in
g
a
nd
c
a
te
gor
iz
in
g
c
a
r
di
a
c
a
r
r
hyt
hm
ia
il
ln
e
s
s
e
s
.
M
a
nj
u
a
nd
N
a
ir
[
9]
e
s
ta
bl
is
he
d
a
m
ode
l
th
a
t
c
la
s
s
if
ie
s
a
r
r
hyt
hm
ia
s
in
to
te
n
c
a
te
gor
ie
s
,
w
it
h
one
c
a
te
gor
y
r
e
pr
e
s
e
nt
in
g
nor
m
a
l
c
ondi
ti
ons
a
nd
th
e
ot
he
r
s
r
e
pr
e
s
e
nt
in
g
di
s
ti
nc
t
f
or
m
s
of
a
r
r
hyt
hm
ia
s
.
T
he
a
ut
hor
s
de
r
iv
e
c
ha
r
a
c
te
r
is
ti
c
s
f
r
om
12
-
le
a
d
E
C
G
da
ta
.
T
hi
s
s
tu
dy
e
m
pl
oye
d
th
e
e
xt
r
e
m
e
gr
a
di
e
nt
boos
ti
ng
(
X
G
B
)
a
lg
or
it
hm
to
do
f
e
a
tu
r
e
r
e
duc
ti
on
a
nd
ba
la
nc
e
d
th
e
da
ta
s
e
t
us
in
g
th
e
S
M
O
T
E
e
di
te
d
ne
a
r
e
s
t
ne
ig
hbor
s
(
E
N
N
)
te
c
hni
que
.
T
he
y
e
m
pl
oye
d
m
ul
ti
pl
e
s
upe
r
vi
s
e
d
ML
m
e
th
ods
f
or
c
la
s
s
if
ic
a
ti
on.
T
h
e
r
e
s
ul
ts
de
m
on
s
tr
a
te
d
th
a
t
th
e
pr
opos
e
d
m
ode
l
e
f
f
e
c
ti
ve
ly
c
a
te
gor
iz
e
s
di
f
f
e
r
e
nt
f
or
m
s
of
a
r
r
hyt
hm
ia
w
it
h
a
r
e
m
a
r
ka
bl
e
le
ve
l
of
p
r
e
c
is
io
n,
r
e
a
c
hi
ng
an
a
c
c
ur
a
c
y
r
a
te
of
97.4
8%
.
T
hi
s
w
or
k
pr
e
s
e
nt
e
d
a
pr
a
c
ti
c
a
l
a
ppr
oa
c
h
to
a
c
c
ur
a
te
ly
a
nd
e
f
f
ic
ie
nt
ly
c
la
s
s
if
y
a
r
r
hyt
hm
ia
s
by
ut
il
iz
in
g
s
ophi
s
ti
c
a
te
d
da
t
a
pr
e
pr
oc
e
s
s
in
g
te
c
hni
que
s
a
nd
ML
a
lg
or
it
hm
s
.
P
e
im
a
nka
r
et
al
.
[
10]
in
tr
oduc
e
d
an
e
ns
e
m
bl
e
le
a
r
ni
ng
m
e
th
od
to
a
ut
om
a
ti
c
a
ll
y
c
la
s
s
if
y
c
om
m
on
c
a
r
di
a
c
a
r
r
hyt
hm
ia
.
T
he
s
tu
dy
e
m
pl
oye
d
th
r
e
e
c
la
s
s
if
ic
a
ti
on
a
lg
or
it
hm
s
,
na
m
e
ly
R
F
,
A
da
B
oos
t
,
a
nd
a
r
ti
f
ic
ia
l
ne
ur
a
l
ne
twor
k
(
A
N
N
)
,
ut
il
iz
in
g
twe
nt
y
-
s
ix
f
e
a
tu
r
e
s
c
ol
le
c
te
d
f
r
om
E
C
G
da
ta
.
U
pon
e
va
lu
a
ti
ng
44
r
e
c
or
di
ngs
f
r
om
th
e
M
I
T
-
B
I
H
a
r
r
hyt
hm
ia
da
ta
ba
s
e
,
it
w
a
s
d
is
c
ove
r
e
d
th
a
t
th
e
R
F
,
A
da
B
oos
t,
a
nd
ANN
a
lg
or
it
hm
s
ha
d
hi
gh
in
di
vi
dua
l
a
c
c
ur
a
c
y
r
a
te
s
of
96.16,
96.16,
a
nd
94.49
%
,
r
e
s
pe
c
ti
ve
ly
.
R
e
m
a
r
ka
bl
y,
th
e
e
ns
e
m
bl
e
m
ode
l
a
c
hi
e
ve
s
a
r
e
m
a
r
ka
bl
e
96.18%
in
c
r
e
a
s
e
in
to
ta
l
a
c
c
ur
a
c
y.
T
he
s
tu
dy
in
di
c
a
te
d
th
a
t
th
e
c
la
s
s
if
ic
a
ti
on
of
a
r
r
hyt
hm
ia
s
us
in
g
an
e
n
s
e
m
bl
e
le
a
r
ni
ng
a
ppr
oa
c
h
is
bot
h
r
e
li
a
bl
e
a
nd
us
e
r
-
f
r
ie
ndl
y.
S
r
a
it
ih
et
al
.
[
11]
in
tr
oduc
e
d
a
nove
l
E
C
G
a
r
r
hyt
hm
ia
c
la
s
s
if
ic
a
ti
on
s
ys
te
m
th
a
t
ut
il
iz
e
s
a
l
a
r
ge
E
C
G
da
ta
ba
s
e
w
it
h
an
in
te
r
-
pa
ti
e
nt
pa
r
a
di
gm
.
T
he
obj
e
c
ti
ve
is
to
im
pr
ove
th
e
id
e
nt
if
ic
a
ti
on
of
le
s
s
c
om
m
on
a
r
r
hyt
hm
ia
c
a
te
gor
ie
s
w
it
hout
ut
il
iz
in
g
f
e
a
tu
r
e
e
xt
r
a
c
ti
on.
F
our
s
upe
r
vi
s
e
d
M
L
m
ode
ls
,
na
m
e
ly
S
V
M
,
KNN,
R
F
,
a
nd
an
e
ns
e
m
bl
e
of
th
e
s
e
th
r
e
e
m
ode
ls
,
w
e
r
e
e
m
pl
oye
d.
T
h
e
m
ode
ls
unde
r
w
e
nt
te
s
ti
ng
us
in
g
a
c
tu
a
l
in
te
r
-
pa
ti
e
nt
E
C
G
r
e
c
or
ds
f
r
o
m
M
I
T
-
da
ta
ba
s
e
(
M
I
T
-
DB
)
.
P
r
io
r
to
te
s
ti
ng,
th
e
da
ta
w
a
s
s
e
gm
e
nt
e
d
a
nd
nor
m
a
li
z
e
d.
T
he
f
oc
u
s
of
th
e
t
e
s
ti
ng
w
a
s
on
c
la
s
s
if
yi
ng
th
e
f
ol
lo
w
in
g
ty
pe
s
of
be
a
ts
:
nor
m
a
l
be
a
t
(
N
O
R
)
,
le
f
t
bundle
br
a
nc
h
bl
oc
k
be
a
t
(
L
B
B
B
)
,
r
ig
ht
bundle
br
a
nc
h
bl
o
c
k
be
a
t
(
R
B
B
B
)
,
a
nd
pr
e
m
a
tu
r
e
a
tr
ia
l
c
ont
r
a
c
ti
on
(
P
A
C
)
.
T
he
r
e
s
ul
t
s
de
m
ons
tr
a
te
th
a
t
S
V
M
s
ur
pa
s
s
e
d
ot
he
r
a
ppr
oa
c
he
s
in
a
ll
c
r
it
e
r
ia
,
obt
a
in
in
g
an
a
c
c
ur
a
c
y
of
0.83.
F
ur
th
e
r
m
or
e
,
th
e
S
V
M
m
ode
l
s
ho
w
n
e
f
f
ic
a
c
y
in
te
r
m
s
of
c
om
put
a
ti
ona
l
e
xpe
ndi
tu
r
e
,
a
pi
vot
a
l
a
s
pe
c
t
in
th
e
im
pl
e
m
e
nt
a
ti
on
of
E
C
G
a
r
r
hyt
hm
ia
c
la
s
s
if
ic
a
ti
on
a
lg
or
it
hm
s
.
G
uo
a
nd
L
in
[
12]
in
tr
oduc
e
d
an
AI
f
r
a
m
e
w
or
k
to
pr
e
c
is
e
ly
id
e
nt
if
y
a
tr
ia
l
f
ib
r
il
la
ti
on
by
a
na
ly
z
in
g
E
C
G
s
ig
na
ls
.
By
e
m
pl
oyi
ng
f
e
a
tu
r
e
e
xt
r
a
c
ti
on
a
nd
e
ns
e
m
bl
e
le
a
r
ni
ng
te
c
hni
que
s
,
th
e
s
ys
t
e
m
a
tt
a
in
e
d
an
im
pr
e
s
s
iv
e
a
c
c
ur
a
c
y
r
a
te
of
92%
.
F
or
m
ode
l
t
r
a
in
in
g
a
nd
to
de
m
ons
tr
a
te
th
e
e
f
f
ic
ie
nc
y
of
th
is
pa
r
a
m
e
te
r
c
om
bi
na
ti
on,
th
e
s
c
ie
nt
is
ts
us
e
d
an
a
r
r
a
nge
m
e
nt
of
P
w
a
ve
m
or
phol
ogy
a
nd
he
a
r
t
r
a
te
va
r
ia
bi
li
t
y
c
ha
r
a
c
te
r
is
ti
c
s
.
AI
e
n
s
e
m
bl
e
le
a
r
ni
ng
m
e
th
ods
li
ke
B
a
ggi
ng,
A
da
B
oos
t,
a
nd
s
ta
c
ki
ng
w
e
r
e
us
e
d
by
th
e
w
r
it
e
r
s
of
th
is
s
tu
dy.
W
he
n
c
om
bi
ne
d
w
it
h
m
a
ny
m
ode
ls
,
th
e
s
ta
c
ki
ng
e
ns
e
m
bl
e
le
a
r
ni
ng
m
e
th
od
pr
oduc
e
d
th
e
m
os
t
a
c
c
ur
a
te
pr
e
di
c
ti
on
s
.
A
lo
ng
w
it
h
an
F1
s
c
or
e
of
92.31%
a
nd
an
a
r
e
a
und
e
r
th
e
c
ur
ve
(
AUC
)
v
a
lu
e
of
91.10%
,
th
e
f
in
di
ngs
in
c
lu
de
a
s
e
ns
it
iv
it
y
r
a
te
of
88%
a
nd
a
s
pe
c
if
ic
it
y
r
a
te
of
96%
.
By
c
om
bi
ni
ng
a
bi
di
r
e
c
ti
ona
l
lo
ng
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
E
ns
e
m
bl
e
m
ode
l
-
bas
e
d a
r
r
hy
th
m
ia
c
la
s
s
if
ic
at
io
n w
it
h l
oc
al
i
nt
e
r
pr
e
ta
bl
e
…
(
M
d. R
abi
ul
I
s
la
m
)
2015
s
hor
t
-
te
r
m
m
e
m
or
y
w
it
h
a
C
N
N
,
L
u
et
al
.
[
13]
pr
e
s
e
nt
e
d
a
nov
e
l
DL
a
r
c
hi
te
c
tu
r
e
f
or
a
ut
onomous
a
r
r
hyt
hm
ia
c
a
te
gor
iz
a
ti
on.
T
h
e
M
I
T
-
B
I
H
a
nd
St
-
P
e
te
r
s
bur
g
da
ta
s
e
ts
a
r
e
u
s
e
d
f
or
tr
a
in
in
g
a
nd
e
v
a
lu
a
ti
on
of
th
e
m
ode
l.
T
he
M
I
T
-
B
I
H
da
ta
s
e
t
yi
e
ld
e
d
a
tr
a
in
in
g
a
c
c
ur
a
c
y
of
100%
,
a
va
li
da
ti
on
a
c
c
ur
a
c
y
of
98%
,
a
nd
a
te
s
ti
ng
a
c
c
ur
a
c
y
of
98%
.
T
he
tr
a
in
in
g,
va
li
da
ti
on,
a
nd
te
s
ti
ng
a
c
c
ur
a
c
y
s
c
or
e
s
f
or
th
e
St
-
P
e
te
r
s
bur
g
da
ta
s
e
t
w
e
r
e
98,
95,
a
nd
95%
,
r
e
s
pe
c
ti
ve
ly
.
G
e
tt
in
g
ve
r
y
a
c
c
ur
a
te
r
e
s
ul
ts
,
p
a
r
ti
c
ul
a
r
ly
w
he
n
de
a
li
ng
w
it
h
th
e
M
I
T
-
B
I
H
da
ta
c
ol
le
c
ti
on.
By
c
om
pa
r
in
g
th
e
m
ode
ls
'
pe
r
f
or
m
a
nc
e
to
th
a
t
of
pr
e
e
xi
s
ti
ng
m
ode
ls
,
we
can
s
e
e
th
a
t
th
is
one
pe
r
f
or
m
s
be
tt
e
r
on
th
e
M
I
T
-
B
I
H
da
t
a
s
e
t.
H
a
s
s
a
n
et
al
.
[
1
4]
in
tr
oduc
e
d
a
ML
a
lg
or
it
hm
de
s
ig
ne
d
to
a
ut
om
a
ti
c
a
ll
y
de
te
c
t
h
e
a
r
t
di
s
e
a
s
e
.
In
th
is
in
qui
r
y,
unba
la
n
c
e
d
E
C
G
s
a
m
pl
e
s
a
r
e
u
s
e
d
to
tr
a
in
S
V
M
,
lo
gi
s
ti
c
r
e
gr
e
s
s
io
n
(
LR
)
,
a
nd
A
da
B
oos
t
c
la
s
s
if
ie
r
s
.
A
da
B
oo
s
t
a
nd
LR
a
r
e
r
a
nke
d
as
th
e
to
p
-
pe
r
f
or
m
in
g
m
ode
ls
a
nd
a
r
e
c
om
bi
ne
d
to
ge
th
e
r
to
e
nha
n
c
e
th
e
ir
p
e
r
f
or
m
a
nc
e
th
r
ough
e
ns
e
m
bl
in
g.
T
h
e
e
n
s
e
m
bl
e
m
ode
l
de
m
ons
tr
a
te
s
s
upe
r
io
r
HD
id
e
nt
if
ic
a
ti
on
a
bi
li
ty
on
th
e
P
T
B
-
E
C
G
a
nd
M
I
T
-
B
I
H
da
ta
s
e
ts
,
a
c
hi
e
vi
ng
hi
gh
a
c
c
ur
a
c
y,
F1
-
s
c
or
e
,
a
nd
AUC
va
lu
e
s
.
T
h
e
e
ns
e
m
bl
e
m
od
e
l
a
c
hi
e
ve
d
a
c
c
ur
a
c
y
s
c
or
e
s
of
94
.
6
0
,
94
.
9
0
,
a
nd
95
.
1
0%
f
or
th
e
P
T
B
-
E
C
G
da
ta
s
e
t,
a
nd
92
.
1
0
,
92
.
6
0
,
a
nd
95
%
f
or
th
e
M
I
T
-
B
I
H
da
ta
s
e
t,
in
te
r
m
s
of
a
c
c
ur
a
c
y,
F1
-
s
c
or
e
,
a
nd
A
U
C
,
r
e
s
pe
c
ti
ve
ly
.
T
he
f
ol
lo
w
in
g
a
r
e
th
e
ke
y
c
ont
r
ib
ut
io
ns
to
th
is
r
e
s
e
a
r
c
h.
‒
By
c
om
bi
ni
ng
da
ta
f
r
om
f
our
di
s
ti
nc
t
da
ta
ba
s
e
s
,
our
s
tu
dy
in
c
r
e
a
s
e
d
th
e
di
ve
r
s
it
y
a
nd
r
ic
hne
s
s
of
th
e
da
ta
s
e
t,
le
a
di
ng
to
be
tt
e
r
ge
n
e
r
a
li
z
a
ti
on
a
nd
m
or
e
a
c
c
ur
a
te
c
la
s
s
if
ic
a
ti
on
of
a
r
r
hyt
hm
ia
s
.
‒
We
a
ls
o
us
e
e
ns
e
m
bl
e
le
a
r
ni
ng
te
c
hni
qu
e
s
,
s
pe
c
if
ic
a
ll
y
a
s
ta
c
ki
ng
e
ns
e
m
bl
e
of
X
G
B
oos
t
a
nd
ba
ggi
ng
X
G
B
oos
t
(
E
B
X
G
B
)
,
w
hi
c
h
s
ig
ni
f
ic
a
nt
ly
im
pr
ove
d
th
e
m
od
e
l'
s
pe
r
f
or
m
a
nc
e
a
nd
c
om
pa
r
e
d
m
ode
l
pe
r
f
or
m
a
nc
e
w
it
h
tr
a
di
ti
ona
l
ML
a
lg
or
it
hm
s
.
‒
I
m
pl
e
m
e
nt
in
g
th
e
S
M
O
T
E
he
lp
e
d
in
ge
ne
r
a
ti
ng
s
ynt
he
ti
c
d
a
ta
to
ba
la
nc
e
th
e
m
in
or
it
y
c
la
s
s
e
s
,
f
ur
th
e
r
e
nha
nc
in
g
th
e
m
ode
l'
s
a
c
c
ur
a
c
y,
r
e
c
a
ll
,
a
nd
pr
e
c
is
io
n.
‒
We
ut
il
iz
e
d
lo
c
a
l
in
te
r
pr
e
ta
bl
e
m
ode
l
-
a
gnos
ti
c
e
xpl
a
na
ti
ons
(
L
I
M
E
)
to
de
te
r
m
in
e
th
e
im
pa
c
t
of
th
e
f
e
a
tu
r
e
s
on
th
e
m
ode
l’
s
out
c
om
e
.
T
he
r
e
m
a
in
in
g
por
ti
ons
of
th
e
pa
pe
r
a
r
e
or
ga
ni
z
e
d
in
th
e
f
ol
l
ow
in
g
m
a
nne
r
:
s
e
c
ti
on
2
out
li
ne
s
th
e
s
tr
a
te
gy
us
e
d
in
th
is
r
e
s
e
a
r
c
h.
S
e
c
ti
on
3
pr
e
s
e
nt
s
th
e
out
c
om
e
s
of
our
r
e
s
e
a
r
c
h
a
nd
c
om
p
a
r
e
s
our
w
or
k
w
it
h
e
a
r
li
e
r
s
tu
di
e
s
. L
a
s
tl
y, s
e
c
ti
on 4 pr
ovi
de
s
t
he
c
onc
lu
s
io
n.
2.
M
E
T
H
O
D
In
th
is
s
e
c
ti
on,
th
e
pr
opos
e
d
w
or
kf
lo
w
of
our
r
e
s
e
a
r
c
h
m
e
th
od
ha
s
be
e
n
di
s
c
u
s
s
e
d
in
de
ta
il
.
F
ig
ur
e
1
s
how
s
th
e
pr
opos
e
d
w
or
kf
lo
w
.
I
n
F
ig
ur
e
1,
da
ta
pr
e
pa
r
a
ti
on
a
nd
m
ode
l
tr
a
in
in
g
w
or
kf
lo
w
be
gi
ns
w
it
h
a
c
qui
r
in
g
a
nd
c
le
a
ni
ng
f
our
di
s
ti
nc
t
da
ta
s
e
t
s
to
r
e
m
ove
in
c
o
ns
is
te
nc
ie
s
a
nd
ir
r
e
le
va
nt
in
f
or
m
a
ti
on.
T
he
s
e
c
le
a
ne
d da
ta
s
e
ts
a
r
e
t
he
n m
e
r
ge
d i
nt
o a
s
in
gl
e
da
ta
s
e
t,
w
hi
c
h un
de
r
goe
s
f
ur
th
e
r
c
le
a
ni
ng t
o e
ns
ur
e
qua
li
ty
a
nd
c
ons
is
te
nc
y.
T
he
out
put
c
ol
um
n
is
tr
a
ns
f
or
m
e
d
us
in
g
a
la
be
l
e
nc
ode
r
,
c
onve
r
ti
ng
c
a
te
gor
ic
a
l
da
ta
in
to
num
e
r
ic
a
l
f
or
m
a
t
a
nd
s
im
pl
if
yi
ng
th
e
m
ul
ti
-
c
la
s
s
c
la
s
s
if
ic
a
ti
on
pr
obl
e
m
in
to
a
bi
na
r
y
c
la
s
s
if
ic
a
ti
on.
F
e
a
tu
r
e
va
r
ia
bl
e
s
a
r
e
s
ta
nd
a
r
di
z
e
d
to
e
ns
ur
e
e
qua
l
c
ont
r
ib
ut
io
n
to
th
e
m
ode
l.
T
he
da
ta
s
e
t
is
s
pl
it
in
to
tr
a
in
in
g
a
nd
te
s
t
s
e
ts
,
ty
pi
c
a
ll
y
in
a
n
80:
20
r
a
ti
o,
to
e
va
lu
a
te
th
e
m
ode
l'
s
pe
r
f
or
m
a
nc
e
r
e
a
li
s
ti
c
a
ll
y. T
o
a
ddr
e
s
s
c
l
a
s
s
im
b
a
la
nc
e
in
th
e
tr
a
in
in
g
s
e
t,
th
e
s
ynt
he
ti
c
S
M
O
T
E
is
a
ppl
ie
d,
ge
ne
r
a
ti
ng
s
ynt
he
ti
c
e
xa
m
pl
e
s
f
or
th
e
m
in
or
it
y
c
la
s
s
.
V
a
r
io
us
M
L
m
ode
ls
a
r
e
th
e
n
t
r
a
in
e
d
on
th
is
ba
la
nc
e
d
a
nd
pr
e
pr
oc
e
s
s
e
d
da
ta
.
F
in
a
ll
y,
th
e
m
ode
ls
a
r
e
e
va
lu
a
te
d
us
in
g
m
e
tr
ic
s
s
uc
h
a
s
a
c
c
ur
a
c
y,
pr
e
c
i
s
io
n,
r
e
c
a
ll
,
a
nd
F
1
-
s
c
or
e
,
pr
ovi
di
ng
a
c
om
pr
e
he
ns
iv
e
a
s
s
e
s
s
m
e
nt
of
th
e
ir
pe
r
f
o
r
m
a
nc
e
.
T
hi
s
s
ys
te
m
a
ti
c
a
ppr
oa
c
h
e
n
s
ur
e
s
th
e
de
ve
lo
pm
e
nt
of
r
obus
t
a
nd
r
e
li
a
bl
e
M
L
m
ode
ls
,
pa
r
ti
c
ul
a
r
ly
va
lu
a
bl
e
in
r
e
s
e
a
r
c
h
s
e
tt
in
gs
w
he
r
e
da
ta
qua
li
ty
a
nd
m
ode
l
a
c
c
ur
a
c
y
a
r
e
pa
r
a
m
ount
.
T
he
a
lg
or
it
hm
f
or
pr
e
pa
r
in
g da
ta
a
nd t
r
a
in
in
g our
pr
opos
e
d m
ode
l
is
s
how
n i
n A
lg
or
it
hm
1
.
A
lg
or
it
hm
1. D
a
ta
pr
e
pa
r
a
ti
on a
nd mode
l
tr
a
in
in
g
I
nput
:
D
at
as
e
t_
1, D
at
as
e
t_
2, D
at
as
e
t_
3, D
at
a
s
e
t_
4
O
ut
put
:
M
ode
l
e
va
lu
a
ti
on r
e
s
ul
ts
S
te
p 1:
L
oa
d da
ta
s
e
ts
D
at
as
e
t_
1
← l
oa
d_da
ta
(
"
pa
th
/t
o/
da
ta
s
e
t1
"
)
D
at
as
e
t_
2 ←
lo
a
d_da
ta
(
"
pa
th
/t
o/
da
ta
s
e
t2
"
)
D
at
as
e
t_
3 ←
lo
a
d_da
ta
(
"
pa
th
/t
o/
da
ta
s
e
t3
"
)
D
at
as
e
t_
4 ←
lo
a
d_da
ta
(
"
pa
th
/t
o/
da
ta
s
e
t4
"
)
S
te
p 2:
C
le
a
n i
ndi
vi
dua
l
da
ta
s
e
t
s
D
at
as
e
t1
_c
le
an ←
c
le
a
n_da
ta
(
D
a
ta
s
e
t_
1)
D
at
as
e
t2
_c
le
an ←
c
le
a
n_da
ta
(
D
a
ta
s
e
t_
2)
D
at
as
e
t3
_c
le
an ←
c
le
a
n_da
ta
(
D
a
ta
s
e
t_
3)
D
at
as
e
t4
_c
le
an ←
c
le
a
n_da
ta
(
D
a
ta
s
e
t_
4)
S
te
p 3:
M
e
r
ge
c
le
a
ne
d da
t
a
s
e
t
s
M
e
r
ge
d_D
at
as
e
t
s
← me
r
ge
_da
ta
s
e
ts
(
[
D
at
as
e
t1
_c
le
an
,
D
at
a
s
e
t
2_c
le
an
,
D
at
as
e
t3
_c
le
an,
D
at
as
e
t4
_c
le
an
])
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
. 14, No. 3, J
une
2025
:
2012
-
2025
2016
S
te
p 4:
U
s
e
L
a
be
l
E
nc
ode
r
on output
c
ol
um
n
M
e
r
ge
d_D
at
as
e
t
s
[
'
ty
pe
'
]
← l
a
be
l_
e
nc
ode
(
M
e
r
ge
d_D
at
as
e
ts
[
'
ty
pe
'
]
)
S
te
p 5:
C
onve
r
t
m
ul
ti
-
c
la
s
s
if
ic
a
ti
on t
o bi
na
r
y c
la
s
s
if
ic
a
ti
on
M
e
r
ge
d_D
at
as
e
t
s
[
'
bi
na
r
y_c
la
s
s
'
]
← c
onve
r
t_
to
_bi
na
r
y(
M
e
r
ge
d
_D
at
as
e
ts
[
'
ty
pe
'
]
)
S
te
p 6:
S
pl
it
f
e
a
tu
r
e
s
a
nd l
a
be
ls
X
←
M
e
r
ge
d_D
at
as
e
ts
.dr
op(
c
ol
um
ns
=
[
'
bi
na
r
y_c
la
s
s
'
]
)
y
←
M
e
r
ge
d_D
at
as
e
ts
[
'
bi
na
r
y_c
la
s
s
'
]
S
te
p 7:
S
ta
nda
r
di
z
e
f
e
a
tu
r
e
va
r
ia
bl
e
s
X
_s
ta
ndar
di
z
e
d
← s
ta
nd
a
r
di
z
e
(
X
)
S
te
p 8:
S
pl
it
t
he
da
ta
s
e
t
in
to
t
r
a
in
in
g a
nd t
e
s
t
s
e
ts
(
X
_t
r
ai
n,
X
_t
e
s
t,
y
_t
r
ai
n,
y
_t
e
s
t
)
←
tr
a
in
_t
e
s
t_
s
pl
it
(
X
_s
ta
ndar
di
z
e
d,
y
,
te
s
t_
s
i
z
e
=
0.2,
r
andom_s
ta
te
=
42)
S
te
p 9:
A
ppl
y S
M
O
T
E
on t
r
a
in
in
g da
ta
(
X
_t
r
ai
n_r
e
s
am
pl
e
d, y
_t
r
ai
n_
r
e
s
am
pl
e
d
)
← a
ppl
y_s
m
ot
e
(
X
_t
r
ai
n, y
_t
r
ai
n
)
S
te
p 10: T
r
a
in
m
ode
ls
m
ode
ls
← t
r
a
in
_m
ode
ls
(
X
_t
r
ai
n_r
e
s
a
m
pl
e
d, y
_t
r
ai
n_
r
e
s
am
pl
e
d
)
S
te
p 11: E
va
lu
a
te
m
ode
ls
on t
e
s
t
da
ta
v
al
uat
io
n_r
e
s
ul
ts
← e
v
a
lu
a
te
_m
ode
ls
(
m
ode
l
s
,
X
_t
e
s
t,
y
_t
e
s
t
)
S
te
p 12: R
e
tu
r
n e
va
lu
a
ti
on r
e
s
ul
ts
F
ig
ur
e
1.
T
he
w
or
kf
lo
w
of
th
e
pr
opos
e
d
w
or
k
2
.1.
D
at
as
e
t
s
c
ol
le
c
t
io
n
In
th
is
s
tu
d
y,
th
e
M
I
T
-
B
I
H
s
u
pr
a
ve
nt
r
i
c
ul
a
r
a
r
r
hy
th
m
i
a
da
ta
ba
s
e
[
1
5]
is
u
s
e
d
as
t
he
m
a
in
d
a
t
a
s
e
t
or
f
oun
da
ti
o
n
d
a
t
a
s
e
t
.
S
o
m
e
of
th
e
o
th
e
r
d
a
t
a
s
e
ts
u
s
e
d
in
c
l
ud
e
th
e
I
N
C
A
R
T
2
-
l
e
a
d
a
r
r
hy
th
m
ia
da
ta
ba
s
e
,
t
h
e
S
C
D
H
da
t
a
b
a
s
e
,
a
n
d
t
he
M
I
T
-
B
I
H
a
r
r
h
yt
hm
i
a
da
t
a
ba
s
e
.
A
ll
of
th
e
da
t
a
in
t
he
c
ol
l
e
c
ti
on
c
a
m
e
f
r
o
m
K
a
ggl
e
.
T
h
e
78
E
C
G
r
e
c
or
d
in
g
s
th
a
t
m
a
k
e
up
th
e
M
I
T
-
B
I
H
s
u
pr
a
ve
nt
r
i
c
u
la
r
a
r
r
hyt
hm
i
a
d
a
t
a
s
e
t
ha
ve
a
d
ur
a
ti
on
of
a
r
o
un
d
30
m
in
u
te
s
a
pi
e
c
e
.
A
s
in
gl
e
pul
s
e
is
r
e
pr
e
s
e
nt
e
d
by
e
a
c
h
of
t
he
1
84,
428
o
c
c
ur
r
e
n
c
e
s
in
t
he
c
ol
l
e
c
ti
on.
A
c
r
os
s
a
ll
of
th
e
s
e
d
a
t
a
s
e
t
s
,
t
he
r
e
a
r
e
a
to
ta
l
of
34
c
h
a
r
a
c
t
e
r
i
s
ti
c
s
,
w
hi
c
h
e
n
c
om
p
a
s
s
a
pa
ti
e
nt
'
s
r
e
c
or
d
a
nd
th
e
c
la
s
s
if
ic
a
t
io
n
ty
p
e
of
th
e
ir
h
e
a
r
tb
e
a
t
(
l
a
b
e
l)
.
T
h
e
r
e
m
a
in
in
g
32
c
h
a
r
a
c
te
r
is
ti
c
s
a
r
e
pa
r
ti
t
io
n
e
d
in
to
t
w
o
gr
ou
p
s
,
e
a
c
h
in
c
lu
di
ng
16
f
e
a
tu
r
e
s
.
O
n
e
s
e
t
c
or
r
e
s
pon
d
s
to
th
e
le
a
d
II
s
ig
na
l
,
w
h
il
e
t
h
e
ot
he
r
s
e
t
c
or
r
e
s
p
on
ds
to
th
e
l
e
a
d
V5
f
e
a
tu
r
e
s
[
14]
.
F
i
gur
e
2
pr
ovi
d
e
s
c
le
a
r
e
vi
d
e
n
c
e
of
a
s
ig
ni
f
i
c
a
nt
im
ba
la
nc
e
in
t
he
c
la
s
s
e
s
of
t
he
d
a
t
a
s
e
t
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
E
ns
e
m
bl
e
m
ode
l
-
bas
e
d a
r
r
hy
th
m
ia
c
la
s
s
if
ic
at
io
n w
it
h l
oc
al
i
nt
e
r
pr
e
ta
bl
e
…
(
M
d. R
abi
ul
I
s
la
m
)
2017
F
ig
ur
e
2.
C
la
s
s
di
s
tr
ib
ut
io
n
of
th
e
ba
s
e
da
t
a
s
e
t
2
.
2
.
D
at
a
p
r
e
p
r
oc
e
s
s
in
g
A
ppl
yi
ng
th
e
m
ode
l
to
an
im
ba
la
nc
e
d
da
ta
s
e
t
can
le
a
d
to
m
is
le
a
di
ng
a
c
c
ur
a
c
y
a
nd
ul
ti
m
a
te
ly
r
e
s
ul
t
in
s
ubpa
r
pe
r
f
or
m
a
nc
e
.
T
he
S
C
D
H
da
ta
s
e
t
ha
s
num
e
r
ous
oc
c
ur
r
e
nc
e
s
th
a
t
ha
ve
m
is
s
in
g
va
lu
e
s
.
C
ons
e
que
nt
ly
,
it
is
ne
c
e
s
s
a
r
y
f
or
us
to
pur
if
y
th
is
da
ta
s
e
t.
In
th
is
s
tu
dy,
we
a
ddr
e
s
s
th
e
i
s
s
ue
of
m
is
s
in
g
va
lu
e
s
by
im
put
in
g
th
e
m
w
it
h
th
e
c
ol
um
n
-
w
is
e
m
e
a
n
va
lu
e
s
.
In
a
d
di
ti
on,
we
e
xc
lu
de
a
ny
r
e
c
or
ds
th
a
t
c
ont
a
in
m
is
s
in
g
va
lu
e
s
e
xpl
ic
it
ly
in
th
e
ty
pe
c
ol
um
n.
A
f
te
r
w
a
r
ds
,
da
ta
r
e
ga
r
di
ng
th
e
S
V
E
B
,
V
E
B
,
a
nd
F
c
a
te
gor
ie
s
w
a
s
obt
a
in
e
d
f
r
om
ot
he
r
da
ta
ba
s
e
s
a
nd
a
dde
d
to
th
e
or
ig
in
a
l
d
a
ta
s
e
t.
T
he
'
r
e
c
or
d'
p
a
r
t
w
a
s
r
e
m
ove
d
as
it
ju
s
t
pe
r
ta
in
s
to
th
e
pa
ti
e
nt
nu
m
be
r
a
nd
doe
s
not
c
ont
r
ib
ut
e
to
th
e
pr
e
di
c
ti
on
of
th
e
he
a
r
tb
e
a
t
ty
pe
.
T
he
'
ty
pe
'
c
ol
um
n
in
th
e
da
ta
s
e
t
is
c
l
a
s
s
if
ie
d
as
th
e
obj
e
c
t
da
ta
ty
p
e
.
H
e
n
c
e
,
it
is
ne
c
e
s
s
a
r
y
to
tr
a
ns
f
or
m
th
e
obj
e
c
t
ty
pe
in
to
a
num
e
r
ic
a
l
ty
pe
us
in
g
th
e
la
be
l
e
n
c
ode
r
te
c
hni
que
.
N
e
xt
,
us
in
g
th
e
S
ta
nda
r
dS
c
a
le
r
te
c
hni
que
to
s
ta
nda
r
di
z
e
th
e
f
e
a
tu
r
e
va
r
ia
bl
e
,
e
ns
ur
in
g
th
a
t
th
e
y
ha
ve
a
s
im
il
a
r
r
a
nge
a
nd
di
s
tr
ib
ut
io
n.
T
hi
s
ha
s
th
e
pot
e
nt
ia
l
to
im
pr
ove
bot
h
p
e
r
f
or
m
a
nc
e
a
nd
a
c
c
ur
a
c
y.
U
s
in
g
S
ta
nda
r
dS
c
a
le
r
is
be
n
e
f
ic
ia
l
w
he
n
th
e
d
a
ta
s
e
t
de
vi
a
te
s
f
r
om
a
nor
m
a
l
di
s
tr
ib
ut
io
n.
T
he
f
or
m
ul
a
f
or
s
ta
nda
r
di
z
a
ti
on
is
as
(
1)
.
′
=
−
µ
,
(
1)
To
a
ddr
e
s
s
th
e
im
ba
la
nc
e
in
th
is
da
t
a
s
e
t,
we
e
m
pl
oye
d
th
e
S
M
O
T
E
to
c
or
r
e
c
t
th
is
une
ve
nne
s
s
.
T
h
e
obj
e
c
ti
ve
of
S
M
O
T
E
is
to
a
c
hi
e
ve
c
la
s
s
di
s
tr
ib
ut
io
n
pa
r
it
y
by
s
ynt
he
s
iz
in
g
f
a
ls
e
s
a
m
pl
e
s
f
or
th
e
m
in
or
it
y
c
la
s
s
.
A
f
te
r
w
a
r
d,
th
e
d
a
ta
s
e
t
is
pa
r
ti
ti
one
d
in
to
a
tr
a
in
in
g
da
ta
s
e
t
a
nd
a
t
e
s
t
d
a
ta
s
e
t.
T
he
m
ode
ls
a
r
e
tr
a
in
e
d
us
in
g
a
da
ta
s
e
t
s
pe
c
if
ic
a
ll
y
de
s
ig
na
te
d
f
or
tr
a
in
in
g
pur
pos
e
s
,
a
nd
th
e
ML
a
lg
or
it
hm
s
go
th
r
ough
m
ul
t
ip
le
r
ounds
to
opt
im
iz
e
th
e
hype
r
pa
r
a
m
e
te
r
s
.
E
va
lu
a
ti
on
of
th
e
m
ode
ls
is
c
onduc
t
e
d
us
in
g
th
e
te
s
t
da
ta
s
e
t.
2
.
3
.
A
p
p
li
e
d
m
ac
h
in
e
l
e
ar
n
in
g
al
gor
it
h
m
s
an
d
e
n
s
e
m
b
le
m
o
d
e
ls
T
hi
s
s
tu
dy
in
ve
s
ti
ga
te
s
m
ul
ti
pl
e
ML
a
lg
or
it
hm
s
a
nd
e
ns
e
m
bl
e
m
ode
ls
to
a
s
s
e
s
s
th
e
ir
e
f
f
e
c
ti
ve
ne
s
s
in
c
la
s
s
if
yi
ng
a
r
r
hyt
hm
ia
s
.
T
he
ML
a
lg
or
it
hm
s
us
e
d
a
r
e
m
ul
ti
la
ye
r
pe
r
c
e
pt
r
on
(
M
L
P
)
,
A
da
B
oos
t
,
L
R
,
DT
,
KNN,
na
ïv
e
B
a
ye
s
(
N
B
)
,
a
nd
X
G
B
.
C
r
e
a
ti
ng
a
ba
ggi
ng
e
ns
e
m
b
le
c
ons
is
ti
ng
of
L
R
,
D
T
,
KNN,
N
B
,
a
nd
X
G
B
m
ode
ls
.
E
ns
e
m
bl
e
c
om
bi
ni
ng
X
G
B
a
nd
B
a
ggi
ng
X
G
B
.
2.3.1.
L
ogi
s
t
ic
r
e
gr
e
s
s
io
n
T
he
pr
im
a
r
y
obj
e
c
ti
ve
of
th
e
s
upe
r
vi
s
e
d
ML
te
c
hni
que
c
a
ll
e
d
LR
is
to
f
or
e
c
a
s
t
th
e
pr
oba
bi
li
ty
of
di
f
f
e
r
e
nt
c
la
s
s
e
s
ba
s
e
d
on
pa
r
ti
c
ul
a
r
in
de
pe
nde
nt
va
r
ia
bl
e
s
.
L
R
di
f
f
e
r
s
f
r
om
li
ne
a
r
r
e
gr
e
s
s
io
n
in
th
a
t
it
us
e
th
e
s
ig
m
oi
d
f
unc
ti
on
to
c
a
lc
ul
a
te
th
e
pr
oba
bi
li
ty
of
an
in
s
ta
nc
e
be
lo
ngi
ng
to
a
s
p
e
c
if
ic
c
la
s
s
,
r
a
th
e
r
th
a
n
pr
oduc
in
g
c
ont
in
uous
out
put
va
lu
e
s
.
T
hi
s
a
lg
or
it
hm
de
m
ons
tr
a
t
e
s
pr
of
ic
ie
nc
y
in
bi
na
r
y
c
la
s
s
if
ic
a
ti
on
ta
s
k
s
by
a
ppl
yi
ng
a
s
ig
m
oi
d
f
unc
ti
on
to
th
e
out
put
of
th
e
li
ne
a
r
r
e
gr
e
s
s
io
n
f
unc
ti
on,
ge
ne
r
a
ti
ng
pr
oba
bi
li
ti
e
s
[
16]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
. 14, No. 3, J
une
2025
:
2012
-
2025
2018
2.3.2.
B
aggi
n
g
lo
gi
s
t
ic
r
e
gr
e
s
s
io
n
In
th
e
pr
o
c
e
s
s
of
ba
g
gi
ng
u
s
in
g
LR
,
a
te
c
hni
q
ue
c
a
ll
e
d
r
a
ndom
s
a
m
pl
in
g
w
it
h
r
e
pl
a
c
e
m
e
nt
is
e
m
pl
oy
e
d.
T
hi
s
in
vo
lv
e
s
tr
a
in
in
g
m
a
ny
in
s
t
a
n
c
e
s
of
LR
s
e
pa
r
a
te
ly
on
di
f
f
e
r
e
nt
s
ub
s
e
ts
of
t
he
da
t
a
s
e
t.
E
a
c
h
LR
or
ba
s
ic
m
o
de
l
in
th
is
s
e
t
le
a
r
ns
to
id
e
nt
if
y
di
s
ti
nc
t
p
a
tt
e
r
n
s
in
t
he
d
a
ta
as
a
r
e
s
ul
t
of
th
e
va
r
i
a
bi
li
ty
in
tr
o
duc
e
d
by
th
e
s
u
bs
e
ts
.
W
h
e
n
c
r
e
a
ti
ng
pr
e
di
c
ti
on
s
,
e
a
c
h
LR
m
ode
l
pr
oduc
e
s
i
ts
out
put
.
T
he
f
i
na
l
pr
e
di
c
ti
on
s
f
or
a
c
la
s
s
if
i
c
a
ti
on
jo
b
a
r
e
de
t
e
r
m
in
e
d
by
ta
ki
ng
a
m
a
jo
r
it
y
vot
e
a
m
on
g
th
e
out
put
s
of
a
ll
t
he
LR
m
od
e
ls
.
2.3.3.
D
e
c
is
io
n
t
ree
DT
is
a
popula
r
s
upe
r
vi
s
e
d
ML
a
ppr
oa
c
h.
Y
ou
m
a
y
us
e
th
is
to
ol
f
or
bot
h
r
e
gr
e
s
s
io
n
a
nd
c
la
s
s
if
ic
a
ti
on
pr
obl
e
m
s
.
I
t'
s
qui
te
f
le
xi
bl
e
.
T
he
da
ta
s
e
t'
s
pr
ope
r
ti
e
s
s
e
r
ve
as
th
e
in
ne
r
node
s
,
th
e
out
c
om
e
s
as
th
e
le
a
f
node
s
,
a
nd
th
e
de
c
i
s
io
n
r
ul
e
s
as
th
e
br
a
nc
h
e
s
in
a
tr
e
e
s
tr
uc
tu
r
e
.
T
he
two
m
a
in
ki
nd
s
of
node
s
in
DT
a
r
e
de
c
is
io
n
node
s
a
nd
le
a
f
node
s
.
D
e
c
i
s
io
n
node
s
a
r
e
ut
il
iz
e
d
f
or
m
a
ki
ng
de
te
r
m
in
a
ti
ons
a
nd
can
pos
s
e
s
s
s
e
ve
r
a
l
br
a
nc
he
s
,
w
hi
le
le
a
f
node
s
a
r
e
e
m
pl
oye
d
to
s
ym
bol
iz
e
th
e
ul
ti
m
a
te
out
c
om
e
s
of
s
uc
h
de
te
r
m
in
a
ti
ons
a
nd
do
not
pos
s
e
s
s
a
ny
e
xt
r
a
br
a
nc
he
s
.
T
he
te
s
t
or
de
c
is
io
ns
a
r
e
de
te
r
m
in
e
d
ba
s
e
d
on
th
e
c
ha
r
a
c
te
r
is
ti
c
s
of
th
e
gi
ve
n
da
ta
s
e
t,
a
nd
a
vi
s
ua
l
de
pi
c
ti
on
of
pot
e
nt
ia
l
s
ol
ut
io
n
s
is
of
f
e
r
e
d
de
pe
ndi
ng
on
s
ta
te
d
c
r
it
e
r
ia
[
17
]
.
T
he
pr
oc
e
s
s
in
vol
ve
s
th
e
de
ve
lo
pm
e
nt
of
a
hi
e
r
a
r
c
hi
c
a
l
s
tr
uc
t
ur
e
,
s
im
il
a
r
to
th
a
t
of
a
tr
e
e
,
s
ta
r
ti
ng
w
it
h
th
e
r
oot
node
a
nd
e
xpa
ndi
ng
w
it
h
a
ddi
ti
ona
l
br
a
nc
he
s
.
2.3.4.
B
aggi
n
g
d
e
c
is
io
n
t
r
e
e
B
a
ggi
ng
is
an
e
f
f
e
c
ti
ve
e
ns
e
m
bl
e
ML
te
c
hni
que
th
a
t
pa
ir
s
w
e
ll
w
it
h
DT
.
T
he
ba
ggi
ng
te
c
hni
qu
e
in
vol
ve
s
tr
a
in
in
g
m
ul
ti
pl
e
DT
s
e
pa
r
a
te
ly
on
di
f
f
e
r
e
nt
s
ub
s
e
ts
of
th
e
da
ta
s
e
t
u
s
in
g
r
a
ndom
s
a
m
pl
in
g
w
it
h
r
e
pl
a
c
e
m
e
nt
[
18]
.
T
hi
s
va
r
ia
bi
li
ty
e
na
bl
e
s
e
a
c
h
tr
e
e
to
c
a
pt
ur
e
di
s
ti
nc
t
pa
tt
e
r
ns
in
th
e
da
t
a
.
E
a
c
h
de
c
i
s
io
n
tr
e
e
in
ba
ggi
ng
yi
e
ld
s
its
ow
n
out
put
.
T
he
f
in
a
l
pr
e
di
c
ti
ons
f
or
a
c
l
a
s
s
if
ic
a
ti
on
ta
s
k
a
r
e
de
te
r
m
in
e
d
by
a
m
a
jo
r
it
y
vot
e
a
m
ong
a
ll
of
th
e
DT
out
put
s
.
2.3.5.
B
aggi
n
g
e
xt
r
e
m
e
g
r
ad
ie
n
t
b
oos
t
in
g
W
he
n
e
m
pl
oyi
ng
X
G
B
w
it
h
ba
ggi
ng,
r
a
ndom
s
a
m
pl
in
g
w
it
h
r
e
pl
a
c
e
m
e
nt
is
ut
il
iz
e
d
to
tr
a
in
m
ul
ti
pl
e
X
G
B
m
ode
ls
s
e
pa
r
a
te
ly
on
di
f
f
e
r
e
nt
s
ubs
e
ts
of
th
e
da
ta
s
e
t.
E
a
c
h
of
th
e
s
e
X
G
B
or
ba
s
ic
m
ode
ls
is
tr
a
in
e
d
to
de
te
c
t
di
s
ti
nc
t
pa
tt
e
r
ns
in
th
e
da
ta
as
a
r
e
s
ul
t
of
th
e
va
r
ie
t
y
in
tr
oduc
e
d
by
th
e
s
ubs
e
ts
.
W
he
n
c
r
e
a
ti
ng
pr
e
di
c
ti
ons
,
each
in
di
vi
dua
l
X
G
B
m
ode
l
pr
oduc
e
s
its
ow
n
out
put
.
T
he
f
in
a
l
pr
e
di
c
ti
ons
f
or
a
c
la
s
s
if
ic
a
ti
on
jo
b
a
r
e
de
te
r
m
in
e
d
by
ta
ki
ng
a
m
a
jo
r
it
y
vot
e
a
m
ong
th
e
out
put
s
of
a
ll
th
e
X
G
B
m
ode
ls
[
19]
.
2.3.6.
B
aggi
n
g
k
-
n
e
ar
e
s
t
n
e
ig
h
b
or
In
th
e
ba
ggi
ng
te
c
hni
que
w
it
h
KNN,
r
a
ndom
s
a
m
pl
in
g
w
it
h
r
e
pl
a
c
e
m
e
nt
is
us
e
d
to
tr
a
in
s
e
ve
r
a
l
KNN
m
ode
ls
in
de
pe
nde
nt
ly
on
di
f
f
e
r
e
nt
s
ubs
e
ts
of
th
e
da
ta
s
e
t.
E
a
c
h
of
th
e
s
e
KNN
or
ba
s
e
m
ode
ls
le
a
r
ns
to
c
a
pt
ur
e
uni
que
pa
tt
e
r
ns
in
th
e
da
ta
c
a
us
e
d
by
v
a
r
ia
bi
li
ty
.
D
ur
in
g
th
e
pr
e
di
c
ti
on
pr
oc
e
s
s
,
e
a
c
h
KNN
a
lg
or
it
hm
pr
oduc
e
s
its
out
put
.
In
c
la
s
s
if
ic
a
ti
on
ta
s
ks
,
th
e
f
in
a
l
pr
e
di
c
ti
on
is
de
te
r
m
in
e
d
by
a
m
a
jo
r
it
y
vot
e
[
20]
.
2.3.7.
B
aggi
n
g
n
aï
ve
B
aye
s
W
h
e
n
e
m
pl
o
yi
n
g
NB
f
or
ba
ggi
ng
,
m
ul
ti
pl
e
NB
c
l
a
s
s
if
i
e
r
s
a
r
e
tr
a
i
ne
d
i
nd
e
p
e
nd
e
n
tl
y
on
di
f
f
e
r
e
n
t
s
u
bs
e
t
s
of
th
e
da
ta
s
e
t
,
w
hi
c
h
a
r
e
g
e
n
e
r
a
te
d
th
r
oug
h
r
a
ndo
m
s
a
m
p
li
n
g
w
it
h
r
e
pl
a
c
e
m
e
n
t.
D
ue
to
th
e
he
t
e
r
o
g
e
n
e
it
y
c
a
u
s
e
d
by
th
e
s
ub
s
e
t
s
,
e
a
c
h
NB
or
b
a
s
e
m
o
de
l
a
c
qui
r
e
s
th
e
c
a
p
a
c
it
y
to
i
d
e
nt
i
f
y
di
s
ti
nc
t
p
a
tt
e
r
n
s
a
nd
c
or
r
e
l
a
ti
on
s
w
it
hi
n
t
he
d
a
t
a
.
D
ur
in
g
t
he
pr
e
di
c
t
io
n
pr
oc
e
s
s
,
e
a
c
h
u
ni
qu
e
NB
m
o
de
l
g
e
ne
r
a
t
e
s
i
ts
o
ut
p
ut
[
21]
.
2.3.8.
S
t
ac
k
in
g
e
n
s
e
m
b
le
of
e
xt
r
e
m
e
g
r
ad
ie
n
t
b
oos
t
in
g
an
d
b
aggi
n
g
e
xt
r
e
m
e
gr
ad
ie
n
t
b
oos
t
in
g
S
ta
c
ki
ng
is
an
e
f
f
e
c
ti
ve
e
ns
e
m
bl
e
l
e
a
r
ni
ng
te
c
hni
que
in
M
L
,
w
he
r
e
th
e
pr
e
di
c
ti
ons
of
m
ul
ti
pl
e
ba
s
e
m
ode
ls
a
r
e
c
om
bi
ne
d
to
a
c
hi
e
ve
a
f
in
a
l
pr
e
di
c
ti
on
th
a
t
de
m
ons
t
r
a
te
s
im
pr
ove
d
pe
r
f
or
m
a
nc
e
.
It
is
a
lt
e
r
na
ti
ve
ly
r
e
f
e
r
r
e
d
to
as
a
s
ta
c
k
e
d
e
n
s
e
m
bl
e
or
s
ta
c
ke
d
ge
ne
r
a
li
z
a
ti
on
.
A
s
ta
c
ki
ng
e
ns
e
m
bl
e
c
a
n
be
li
ke
n
e
d
to
a
c
ol
le
c
ti
on
of
e
xpe
r
ts
le
d
by
a
le
a
d
e
r
.
T
he
l
e
a
de
r
ta
ke
s
in
to
a
c
c
ount
th
e
out
put
s
of
each
e
xpe
r
t
be
f
or
e
m
a
ki
ng
th
e
f
in
a
l
de
c
is
io
n.
A
ppl
yi
ng
a
s
ta
c
ki
ng
e
ns
e
m
bl
e
to
a
bi
g
a
nd
di
ve
r
s
e
da
ta
s
e
t
is
a
dva
nt
a
ge
ou
s
.
T
hi
s
di
ve
r
s
it
y
a
ll
ow
s
th
e
m
od
e
l
to
e
f
f
ic
ie
nt
ly
le
a
r
n
th
e
c
or
r
e
la
ti
on
b
e
twe
e
n
th
e
pr
e
di
c
ti
ons
of
th
e
ba
s
e
m
ode
l
s
a
nd
th
e
ta
r
ge
t
va
r
ia
bl
e
.
T
he
s
tu
dy
ut
il
iz
e
s
X
G
B
a
nd
ba
ggi
ng
X
G
B
as
ba
s
e
e
s
ti
m
a
to
r
s
,
w
it
h
LR
s
e
r
vi
ng
as
th
e
m
e
ta
-
m
ode
l
[
22]
.
To
e
nha
nc
e
p
e
r
f
or
m
a
nc
e
of
th
e
c
la
s
s
if
ie
r
s
,
th
e
opt
im
iz
a
ti
on
of
s
e
ve
r
a
l
hype
r
pa
r
a
m
e
te
r
s
ha
s
b
e
e
n
pe
r
f
or
m
e
d.
T
a
bl
e
1
di
s
pl
a
ys
th
e
opt
im
iz
e
d
hyp
e
r
pa
r
a
m
e
te
r
s
.
2
.
4
.
P
e
r
f
or
m
an
c
e
e
val
u
at
io
n
m
e
t
r
ic
s
T
o
a
s
s
e
s
s
t
he
p
e
r
f
or
m
a
n
c
e
of
a
r
r
h
yt
h
m
i
a
c
l
a
s
s
if
i
c
a
ti
o
n
f
r
om
E
E
G
,
m
a
n
y
e
v
a
l
ua
ti
o
n
m
e
tr
i
c
s
a
r
e
ut
il
iz
e
d
.
E
a
c
h
ta
r
g
e
ti
ng
a
s
p
e
c
if
i
c
a
s
pe
c
t
of
i
ts
p
e
r
f
or
m
a
nc
e
u
s
i
ng
p
r
e
c
i
s
i
on,
r
e
c
a
ll
,
F
1
-
s
c
or
e
,
a
n
d
a
c
c
ur
a
c
y
m
e
tr
ic
s
.
T
h
e
s
e
m
e
tr
i
c
s
of
f
e
r
a
n
um
e
r
i
c
a
l
e
v
a
lu
a
t
io
n
of
t
h
e
m
o
de
l
'
s
a
b
il
it
y
to
pr
e
c
i
s
e
l
y
c
l
a
s
s
i
f
y
a
r
r
hyt
hm
i
a
[
23]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
E
ns
e
m
bl
e
m
ode
l
-
bas
e
d a
r
r
hy
th
m
ia
c
la
s
s
if
ic
at
io
n w
it
h l
oc
al
i
nt
e
r
pr
e
ta
bl
e
…
(
M
d. R
abi
ul
I
s
la
m
)
2019
T
a
bl
e
1.
O
pt
im
iz
e
d
m
ode
l
pe
r
f
or
m
a
nc
e
us
e
d
f
in
e
-
tu
ne
d
M
ode
l
H
ype
r
pa
r
a
m
e
t
e
r
s
T
une
d
LR
C
, pe
na
l
t
y, s
ol
ve
r
, m
a
x
-
i
t
e
r
XGB
n
-
e
s
t
i
m
a
t
or
s
, m
a
x
-
de
pt
h, l
e
a
r
ni
ng r
a
t
e
, ga
m
m
a
DT
c
r
i
t
e
r
i
on, m
a
x
-
de
pt
h, m
i
n
-
s
a
m
pl
e
s
-
s
pl
i
t
KNN
n_ne
i
ghbor
s
NB
va
r
_s
m
oot
hi
ng
A
da
B
oos
t
ba
s
e
-
e
s
t
i
m
a
t
or
, n_e
s
t
i
m
a
t
or
s
, l
e
a
r
ni
ng_r
a
t
e
M
L
P
hi
dde
n_l
a
ye
r
_s
i
z
e
s
2
.
5
.
M
od
e
l
e
xp
la
n
at
io
n
u
s
in
g
L
I
M
E
L
I
M
E
is
a
ve
r
s
a
t
i
le
t
oo
l
t
ha
t
e
n
ha
nc
e
s
ou
r
c
om
p
r
e
he
ns
i
on
of
th
e
de
c
is
i
on
-
m
a
k
in
g
p
r
oc
e
s
s
of
in
t
r
ic
a
t
e
m
o
de
ls
.
It
a
c
hi
e
ve
s
t
hi
s
by
c
o
ns
t
r
u
c
t
in
g
a
c
le
a
r
a
nd
c
o
m
p
r
e
he
ns
ib
le
f
r
a
m
e
w
or
k
ba
s
e
d
on
a
s
pe
c
i
f
i
c
s
c
e
na
r
io
,
of
f
e
r
i
ng
in
s
ig
ht
s
in
to
t
he
be
ha
vi
or
of
th
e
bl
a
c
k
-
bo
x
m
ode
l
in
t
ha
t
pa
r
t
ic
u
la
r
c
on
te
x
t.
An
a
dva
nt
a
ge
o
us
a
s
pe
c
t
of
L
I
M
E
is
its
c
om
pa
t
ib
il
it
y
w
i
th
s
e
ve
r
a
l
ML
m
o
de
ls
.
T
he
r
e
f
o
r
e
,
L
I
M
E
s
e
r
ve
s
as
a
va
lu
a
b
le
to
ol
f
o
r
im
pr
ov
in
g
ou
r
un
de
r
s
ta
ndi
ng
of
m
ode
ls
in
ot
he
r
f
ie
l
ds
.
F
o
r
th
is
s
tu
dy,
t
he
a
u
th
or
s
e
m
p
lo
y
in
g
L
I
M
E
to
a
s
c
e
r
ta
i
n
th
e
in
f
lu
e
nc
e
of
va
r
i
ous
v
a
r
i
a
bl
e
s
on
th
e
ou
tc
o
m
e
s
a
nd
un
c
ov
e
r
th
e
unde
r
ly
in
g
r
a
t
io
na
le
b
e
hi
nd
t
he
m
o
de
l'
s
de
c
is
i
on
-
m
a
k
in
g
p
r
o
c
e
s
s
.
L
I
M
E
is
a
r
ob
us
t
t
e
c
hn
iq
ue
de
v
e
l
ope
d
to
im
p
r
ove
th
e
c
om
pr
e
he
ns
ib
il
it
y
of
i
nt
r
ic
a
te
ML
m
ode
ls
at
a
s
pe
c
if
ic
le
ve
l
.
T
h
e
a
im
is
to
e
n
ha
n
c
e
th
e
lu
c
id
i
ty
a
nd
c
o
m
p
r
e
he
ns
i
on
of
f
o
r
e
c
a
s
ts
by
in
c
o
r
po
r
a
t
in
g
th
e
no
ti
o
n
of
lo
c
a
li
z
e
d
e
xp
la
na
ti
ons
.
T
he
e
m
ph
a
s
is
li
e
s
on
th
e
i
nt
e
r
p
r
e
ta
b
il
it
y
of
i
nd
iv
i
dua
l
da
t
a
i
ns
t
a
nc
e
s
r
a
th
e
r
th
a
n
th
e
c
o
m
p
le
te
m
ode
l.
T
hi
s
a
pp
r
oa
c
h
f
unc
ti
ons
by
c
r
e
a
ti
ng
m
od
if
ie
d
s
a
m
p
le
s
in
t
he
v
ic
i
ni
t
y
of
th
e
s
p
e
c
i
f
i
c
e
ve
n
t
of
in
te
r
e
s
t
,
c
a
us
in
g
r
a
n
do
m
f
l
uc
t
ua
ti
o
ns
in
t
he
c
ha
r
a
c
te
r
is
ti
c
va
lu
e
s
.
L
I
M
E
is
a
te
c
h
ni
que
t
ha
t
e
f
f
e
c
t
iv
e
ly
e
s
ti
m
a
te
s
t
he
in
t
r
ic
a
t
e
de
c
is
io
n
b
o
r
de
r
of
a
m
o
de
l
n
e
a
r
a
s
pe
c
if
ic
in
s
ta
nc
e
,
w
it
ho
ut
be
in
g
c
o
ns
t
r
a
in
e
d
to
a
s
in
g
le
m
o
de
l
.
T
hi
s
is
a
c
hi
e
ve
d
by
d
e
ve
lo
p
in
g
a
l
oc
a
ll
y
in
te
r
p
r
e
ta
b
le
m
ode
l,
us
ua
l
ly
in
th
e
f
o
r
m
of
a
l
in
e
a
r
m
ode
l.
L
I
M
E
ut
i
li
z
e
s
k
e
r
ne
l
iz
e
d
w
e
ig
ht
s
to
gu
a
r
a
nt
e
e
th
a
t
t
he
pe
r
tu
r
be
d
s
a
m
pl
e
s
ha
ve
a
s
i
gn
if
ic
a
n
t
im
pa
c
t
on
t
he
lo
c
a
l
m
ode
l
[
24
]
.
T
he
w
e
ig
ht
s
a
s
s
i
gn
h
ig
h
e
r
p
r
io
r
i
ty
to
s
a
m
p
le
s
th
a
t
a
r
e
c
l
os
e
r
to
th
e
o
r
ig
i
na
l
in
s
ta
nc
e
.
T
he
s
ig
n
i
f
ic
a
nc
e
of
th
e
f
e
a
tu
r
e
is
e
va
lu
a
te
d
by
a
na
ly
z
in
g
th
e
c
oe
f
f
ic
ie
nt
s
of
th
is
s
pe
c
if
ic
m
o
de
l,
w
h
ic
h
m
e
a
s
u
r
e
s
t
he
im
p
a
c
t
of
e
a
c
h
c
ha
r
a
c
t
e
r
is
t
ic
on
t
he
de
c
is
i
on
-
m
a
ki
n
g
p
r
oc
e
s
s
.
T
he
m
os
t
i
nf
lu
e
nt
ia
l
t
r
a
i
ts
,
as
a
s
s
e
s
s
e
d
by
t
he
i
r
hi
ghe
s
t
r
e
le
va
nc
e
s
c
o
r
e
s
,
p
r
ovi
de
a
lo
c
a
li
z
e
d
e
xp
la
na
t
io
n
th
a
t
id
e
nt
if
ie
s
t
he
va
r
ia
bl
e
s
w
it
h
th
e
g
r
e
a
te
s
t
i
m
pa
c
t
on
t
he
s
pe
c
if
ic
p
r
og
nos
is
.
L
I
M
E
is
va
lu
a
b
l
e
in
s
e
ve
r
a
l
i
ndus
t
r
ie
s
,
e
s
pe
c
ia
l
ly
in
s
e
ns
i
ti
v
e
s
e
c
to
r
s
li
ke
he
a
lt
hc
a
r
e
or
f
in
a
nc
e
,
w
he
r
e
c
om
p
r
e
he
nd
in
g
th
e
f
u
nda
m
e
n
ta
l
r
a
t
io
na
le
be
hi
nd
s
pe
c
i
f
ic
pr
e
di
c
ti
ons
is
v
it
a
l
[
25
]
.
L
I
M
E
p
la
ys
a
c
r
uc
ia
l
r
o
le
in
bu
il
d
i
ng
t
r
us
t
a
nd
e
n
a
bl
in
g
th
e
im
pl
e
m
e
n
ta
ti
o
n
of
ML
m
ode
ls
in
r
e
a
l
-
li
f
e
s
it
u
a
t
io
ns
by
o
f
f
e
r
in
g
c
le
a
r
a
nd
unde
r
s
ta
n
da
b
le
e
xp
la
na
t
io
ns
of
th
e
m
od
e
l
'
s
de
c
is
io
n
-
m
a
k
in
g
p
r
oc
e
s
s
f
o
r
each
pa
r
t
ic
ul
a
r
c
a
s
e
[
2
6]
.
T
h
e
f
o
r
m
ul
a
f
o
r
L
I
M
E
is
as
(
2)
.
(
,
,
)
=
∑
=
1
(
)
(
)
+
|
|
1
(
2)
T
he
va
r
ia
bl
e
s
in
th
e
(
2)
a
r
e
de
f
in
e
d
as
f
ol
lo
w
s
:
x
r
e
pr
e
s
e
nt
s
th
e
in
s
ta
nc
e
,
f
r
e
pr
e
s
e
nt
s
th
e
a
ppr
oxi
m
a
ti
on
m
ode
l,
r
e
pr
e
s
e
nt
s
th
e
f
e
a
tu
r
e
im
por
ta
nc
e
w
e
ig
ht
s
,
d
r
e
pr
e
s
e
nt
s
th
e
f
e
a
tu
r
e
c
ount
,
a
nd
C
r
e
pr
e
s
e
nt
s
th
e
r
e
gul
a
r
iz
a
ti
on
pa
r
a
m
e
t
e
r
.
T
he
e
qu
a
ti
on
c
ons
i
s
ts
of
a
w
e
ig
ht
e
d
c
om
bi
na
ti
on
of
f
e
a
tu
r
e
s
a
nd
a
r
e
gul
a
r
iz
a
ti
on
f
a
c
to
r
th
a
t
e
nc
our
a
ge
s
s
pa
r
s
it
y
in
th
e
w
e
ig
ht
s
a
s
s
ig
ne
d
to
each
f
e
a
tu
r
e
.
T
h
e
opt
im
iz
a
ti
on
is
s
u
e
e
nt
a
il
s
f
in
di
ng
w
e
ig
ht
s
Π
th
a
t
m
in
im
iz
e
th
e
di
s
c
r
e
pa
nc
y
b
e
twe
e
n
pr
e
di
c
ti
ons
m
a
de
by
a
bl
a
c
k
-
box
m
ode
l
a
nd
f
or
e
c
a
s
ts
m
a
de
by
an
a
ppr
oxi
m
a
ti
on
m
ode
l
f
or
a
gi
ve
n
in
s
ta
nc
e
x.
3.
R
E
S
U
L
T
S
AND
D
I
S
C
U
S
S
I
O
N
T
a
bl
e
2
il
lu
s
tr
a
te
s
t
he
pe
r
f
o
r
m
a
nc
e
c
om
pa
r
is
o
n
of
ML
m
od
e
ls
a
nd
e
ns
e
m
bl
e
te
c
hni
que
m
o
de
ls
w
he
n
ut
il
iz
i
ng
a
s
in
gl
e
ba
s
e
d
a
ta
ba
s
e
v
e
r
s
us
w
h
e
n
us
in
g
c
o
m
bi
ne
d
da
t
a
ba
s
e
s
.
C
o
m
b
in
in
g
da
t
a
s
e
ts
of
te
n
le
a
ds
to
an
in
c
r
e
a
s
e
in
a
c
c
ur
a
c
y
in
m
os
t
c
a
s
e
s
.
T
a
b
le
2
de
m
ons
t
r
a
te
s
th
a
t
in
th
e
c
o
m
b
in
e
d
da
ta
s
e
t,
th
e
s
ta
c
k
in
g
E
B
X
G
B
a
n
d
B
X
G
B
out
pe
r
f
o
r
m
s
a
ll
o
th
e
r
m
ode
ls
u
t
il
iz
e
d
in
t
hi
s
s
tu
d
y
in
te
r
m
s
of
a
c
c
u
r
a
c
y
.
T
he
B
X
G
B
m
ode
l
a
c
hi
e
ve
s
an
a
c
c
u
r
a
c
y
of
9
8.
59
%,
a
r
e
c
a
l
l
of
9
7.
43
%,
a
nd
a
p
r
e
c
is
i
on
of
97
.
95
%.
T
he
E
B
X
G
B
m
ode
l
a
c
h
ie
ve
s
an
a
c
c
u
r
a
c
y
of
9
8.
61
%,
a
r
e
c
a
ll
of
97.
66
%,
a
n
d
a
p
r
e
c
is
io
n
of
9
7.
77
%
.
T
he
c
on
f
us
io
n
m
a
t
r
ix
o
f
t
he
be
s
t
m
o
de
l
(
E
B
X
G
B
)
i
s
s
how
n
in
F
ig
ur
e
3.
F
i
gu
r
e
3
(
a
)
r
e
pr
e
s
e
nt
s
th
e
c
on
f
us
io
n
m
a
t
r
i
x
f
o
r
th
e
E
B
X
G
B
m
o
de
l
us
in
g
a
s
in
gl
e
da
t
a
ba
s
e
.
F
ig
ur
e
3
(
b
)
r
e
pr
e
s
e
n
ts
t
he
c
on
f
us
io
n
m
a
t
r
i
x
f
o
r
th
e
E
B
X
G
B
m
ode
l
us
in
g
t
he
c
om
bi
ne
d
da
ta
ba
s
e
.
I
n
F
ig
u
r
e
4
,
T
h
e
r
e
c
e
iv
e
r
ope
r
a
t
in
g
c
ha
r
a
c
te
r
is
t
ic
(
R
O
C
)
c
u
r
ve
o
f
th
e
E
B
X
G
B
m
ode
l
h
a
s
b
e
e
n
i
ll
us
t
r
a
te
d.
I
t
c
a
n
be
s
e
e
n
th
a
t
th
e
A
U
C
i
s
1
.00
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
. 14, No. 3, J
une
2025
:
2012
-
2025
2020
T
a
bl
e
2.
T
he
pe
r
f
or
m
a
nc
e
of
a
ll
m
ode
l
s
in
gl
e
a
nd c
om
bi
ne
d d
a
ta
ba
s
e
s
M
ode
l
S
i
ngl
e
da
t
a
ba
s
e
C
om
bi
ne
d da
t
a
ba
s
e
P
r
e
c
i
s
i
on (
%
)
R
e
c
a
l
l
(
%
)
A
c
c
ur
a
c
y (
%
)
P
r
e
c
i
s
i
on (
%
)
R
e
c
a
l
l
(
%
)
A
c
c
ur
a
c
y (
%
)
W
i
t
hout
B
a
ggi
ng
LR
57.8
7
85.94
90.79
90.22
87.50
93.28
XGB
91.65
91.88
98.02
97.75
97.45
98.53
DT
82.24
90.02
96.46
94.86
95.89
97.16
KNN
84.47
93.99
97.20
96.29
97.56
98.11
NB
36.09
55.16
82.87
83.55
78.87
88.79
A
da
B
oos
t
88.02
87.90
97.11
96.93
96.15
97.89
M
L
P
86.86
92.76
97.44
97.39
97.19
98.34
W
i
t
h B
a
ggi
ng
B
a
ggi
ng L
R
52.69
86.12
89.04
90.51
85.83
92.91
B
X
G
B
91.84
92.51
98.11
97.95
97.43
98.59
B
a
ggi
ng D
T
88.86
93.95
97.86
97.52
97.52
98.48
B
a
ggi
ng K
N
N
85.69
94.67
97.62
97.09
97.60
98.37
B
a
ggi
ng N
B
36.63
54.28
83.22
84.07
78.63
88.91
P
r
opos
e
d M
ode
l
(
E
B
X
G
B
)
90.51
93.03
97.99
97.77
97.66
98.61
(
a
)
(
b)
F
ig
ur
e
3.
C
onf
us
io
n
m
a
tr
i
x
of
t
he
b
e
s
t
m
o
de
l
(
E
B
X
G
B
)
of
u
s
i
ng:
(
a
)
s
in
gl
e
d
a
t
a
b
a
s
e
a
nd
(
b)
c
om
b
in
e
d
da
ta
ba
s
e
F
ig
ur
e
4. R
O
C
c
ur
ve
of
t
he
E
B
X
G
B
m
ode
l
L
I
M
E
is
e
m
p
lo
y
e
d
t
o
c
o
m
pr
e
h
e
n
d
t
he
d
e
c
i
s
io
n
-
m
a
k
in
g
pr
oc
e
s
s
of
a
s
ta
c
ki
n
g
m
od
e
l
.
F
ig
ur
e
s
5
a
n
d
6
de
p
ic
t
a
n
a
pp
li
c
a
ti
o
n
of
th
e
L
I
M
E
-
b
a
s
e
d
X
A
I
m
e
th
od
t
o
a
n
a
l
yz
e
th
e
s
t
a
c
ki
ng
m
od
e
l.
I
n
F
ig
ur
e
5
,
t
h
e
s
t
a
c
ki
ng
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
E
ns
e
m
bl
e
m
ode
l
-
bas
e
d a
r
r
hy
th
m
ia
c
la
s
s
if
ic
at
io
n w
it
h l
oc
al
i
nt
e
r
pr
e
ta
bl
e
…
(
M
d. R
abi
ul
I
s
la
m
)
2021
m
od
e
l
pr
e
di
c
t
s
a
r
r
hy
th
m
ia
b
a
s
e
d
on
th
e
f
ol
lo
w
in
g
c
r
i
te
r
ia
:
a
Q
R
S
in
te
r
va
l
gr
e
a
t
e
r
th
a
n
0
.4
8,
a
p
os
t
-
R
R
i
nt
e
r
v
a
l
gr
e
a
t
e
r
t
ha
n
0
.9
5,
a
Q
T
i
nt
e
r
v
a
l
gr
e
a
t
e
r
th
a
n
0.5
0,
a
Q
T
i
nt
e
r
v
a
l
gr
e
a
te
r
th
a
n
0.
47
,
a
n
S
T
in
t
e
r
va
l
gr
e
a
te
r
t
ha
n
0.4
9,
a
qP
e
a
k
gr
e
a
te
r
t
h
a
n
0.
47
,
a
po
s
t
-
R
R
in
te
r
va
l
gr
e
a
t
e
r
th
a
n
0
.9
0,
a
qr
s
_m
or
ph
1
gr
e
a
t
e
r
th
a
n
0
.4
0,
a
nd
a
qr
s
_m
or
ph
0
gr
e
a
te
r
th
a
n
0
.4
7.
I
n
F
ig
ur
e
6,
t
he
s
t
a
c
ki
ng
m
od
e
l
p
r
e
di
c
t
s
non
-
a
r
r
hyt
hm
i
a
b
a
s
e
d
o
n
th
e
f
o
ll
o
w
in
g
c
r
it
e
r
i
a
:
a
Q
T
i
nt
e
r
va
l
of
l
e
s
s
th
a
n
-
0.5
3,
a
Q
T
i
nt
e
r
v
a
l
o
f
l
e
s
s
t
h
a
n
-
0.4
7,
a
pr
e
_R
R
i
nt
e
r
v
a
l
of
m
or
e
t
ha
n 0.6
5,
a
Q
R
S
i
n
te
r
v
a
l
of
l
e
s
s
t
ha
n
-
0.
12,
a
Q
R
S
m
or
ph
3 v
a
l
ue
of
m
or
e
th
a
n 0
.62
,
a
n
d
a
s
P
e
a
k
va
lu
e
of
l
e
s
s
t
ha
n 0
.1
0.
F
ig
ur
e
5. L
I
M
E
e
xpl
a
in
a
bl
e
pr
e
di
c
ti
on i
nt
e
r
pr
e
ta
ti
on
(
pr
e
di
c
t
a
r
r
hyt
hm
ia
)
F
ig
ur
e
6. L
I
M
E
e
xpl
a
in
a
bl
e
pr
e
di
c
ti
on i
nt
e
r
pr
e
ta
ti
on
(
pr
e
di
c
t
n
on
-
a
r
r
hyt
hm
ia
)
W
hi
le
pr
e
vi
ous
s
tu
di
e
s
ha
v
e
e
xpl
or
e
d
a
r
r
hyt
hm
ia
de
te
c
ti
on
us
i
ng
s
in
gl
e
da
ta
s
e
ts
a
nd
in
di
vi
dua
l
M
L
m
ode
ls
,
th
e
y
ha
ve
not
e
xpl
ic
it
ly
a
ddr
e
s
s
e
d
th
e
pot
e
nt
ia
l
im
pr
ove
m
e
nt
s
in
de
te
c
ti
on
a
c
c
ur
a
c
y
a
nd
m
od
e
l
in
te
r
pr
e
ta
bi
li
ty
th
a
t
c
oul
d
be
a
c
hi
e
ve
d
by
c
om
bi
ni
ng
m
ul
ti
pl
e
da
ta
s
e
ts
a
nd
e
m
pl
oyi
ng
e
ns
e
m
bl
e
le
a
r
ni
ng
te
c
hni
que
s
.
T
h
e
li
m
it
e
d
num
be
r
of
e
xa
m
pl
e
s
f
or
c
e
r
ta
in
a
r
r
hyt
hm
ia
c
la
s
s
e
s
in
e
a
r
li
e
r
r
e
s
e
a
r
c
h
ha
s
a
l
s
o
im
pe
de
d
th
e
m
ode
ls
'
a
bi
li
ty
to
ge
ne
r
a
li
z
e
e
f
f
e
c
ti
ve
ly
.
T
hi
s
s
tu
dy
s
e
e
ks
to
f
il
l
th
e
s
e
ga
ps
by
in
te
gr
a
ti
ng
da
ta
f
r
om
m
ul
ti
pl
e
s
our
c
e
s
a
nd
us
in
g
e
ns
e
m
bl
e
m
e
th
ods
to
e
nha
nc
e
m
ode
l
r
obus
tn
e
s
s
a
nd
pe
r
f
or
m
a
nc
e
,
th
e
r
e
by
pr
ovi
di
ng
a
m
o
r
e
c
om
pr
e
he
ns
iv
e
a
nd
a
c
c
ur
a
te
a
ppr
oa
c
h
to
a
r
r
hyt
hm
ia
de
te
c
ti
on.
T
hi
s
a
ppr
oa
c
h
he
lp
s
in
be
tt
e
r
ge
ne
r
a
li
z
a
ti
on
a
nd
m
or
e
a
c
c
ur
a
te
c
la
s
s
if
ic
a
ti
on
of
a
r
r
hyt
hm
ia
s
.
A
ddi
ti
ona
ll
y,
w
e
e
m
pl
oye
d
e
n
s
e
m
bl
e
le
a
r
ni
ng
te
c
hni
que
s
,
s
pe
c
if
ic
a
ll
y
a
s
ta
c
ki
ng
E
B
X
G
B
,
w
hi
c
h
s
i
gni
f
ic
a
nt
ly
im
pr
ove
d
m
ode
l
pe
r
f
or
m
a
nc
e
.
B
y
us
in
g
th
e
S
M
O
T
E
to
ba
la
n
c
e
th
e
da
ta
s
e
t,
w
e
f
ur
th
e
r
e
nha
nc
e
d
th
e
m
ode
l'
s
a
c
c
ur
a
c
y,
r
e
c
a
ll
,
a
nd
pr
e
c
is
io
n.
T
hi
s
c
om
pr
e
he
n
s
iv
e
s
tr
a
te
gy
of
c
om
bi
ni
ng
da
t
a
s
e
t
s
a
nd
le
ve
r
a
gi
ng
a
dva
nc
e
d
e
ns
e
m
bl
e
m
e
th
od
s
pr
ovi
de
s
a
m
or
e
r
obus
t
a
nd i
nt
e
r
pr
e
ta
bl
e
s
ol
ut
io
n f
or
a
r
r
hyt
hm
ia
de
te
c
ti
on.
T
he
E
B
X
G
B
m
ode
l
yi
e
ld
s
s
upe
r
io
r
r
e
s
ul
ts
in
te
r
m
s
of
pr
e
c
is
i
on,
r
e
c
a
ll
,
a
nd
a
c
c
ur
a
c
y.
U
ti
li
z
e
th
e
E
B
X
G
B
m
ode
l,
th
e
a
c
c
ur
a
c
y
s
c
or
e
f
or
th
e
s
in
gl
e
da
ta
ba
s
e
w
a
s
97.99%
,
w
it
h
pr
e
c
is
io
n
a
nd
r
e
c
a
ll
va
lu
e
s
of
90.51
a
nd
93.03%
r
e
s
pe
c
ti
ve
ly
.
T
he
c
om
bi
ne
d
da
ta
ba
s
e
e
a
r
n
e
d
a
c
c
ur
a
c
y,
pr
e
c
i
s
io
n,
a
nd
r
e
c
a
ll
r
a
ti
ngs
of
98.61,
97.77,
a
nd
97.66%
r
e
s
pe
c
ti
ve
ly
.
W
e
ut
il
iz
e
L
I
M
E
a
na
ly
s
is
to
id
e
nt
if
y
e
s
s
e
nt
ia
l
f
e
a
tu
r
e
s
,
a
s
w
e
ll
a
s
to
c
la
s
s
if
y
da
ta
ba
s
e
d
on
di
f
f
e
r
e
nt
c
la
s
s
e
s
.
T
a
bl
e
3
pr
e
s
e
nt
s
a
c
om
pa
r
is
on
be
twe
e
n
our
s
ugge
s
te
d
m
ode
l
a
nd
th
e
m
ode
ls
us
e
d i
n e
a
r
li
e
r
s
tu
di
e
s
. T
he
pr
opos
e
d m
ode
l,
E
B
X
G
B
o
ut
pe
r
f
or
m
s
e
xi
s
ti
ng s
ta
te
-
of
-
th
e
-
a
r
t
a
r
r
hyt
h
m
ia
de
te
c
ti
on
a
lg
or
it
hm
s
w
it
h
a
n
a
c
c
ur
a
c
y
of
98.61%
.
T
he
E
B
X
G
B
m
ode
l
not
onl
y
out
pe
r
f
or
m
s
but
a
ls
o
a
dds
Evaluation Warning : The document was created with Spire.PDF for Python.