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
. 15, No. 1, Febr
ua
r
y 2026
, pp.
655
~
671
I
S
S
N
:
2252
-
8938
,
D
O
I
:
10.11591/
ij
a
i.
v
15
.i
1
.pp
655
-
671
655
Jou
r
n
al
h
om
e
page
:
ht
tp
:
//
ij
ai
.
ia
e
s
c
or
e
.c
om
S
t
r
ok
e
p
r
e
d
i
c
t
i
on
u
si
n
g d
at
a b
a
l
an
c
i
n
g m
e
t
h
od
a
n
d
e
xt
r
e
m
e
gr
ad
i
e
n
t
b
oost
i
n
g
A
b
d
M
iz
w
ar
A
. R
ah
im
1
, A
n
n
a B
ai
t
a
1
, F
ir
m
an
A
s
h
ar
u
d
in
1
,
Wah
id
M
if
t
ah
u
l
A
s
h
ar
i
1
,
Wal
id
y R
ah
m
an
H
ak
im
2
,
A
n
d
r
iy
an
D
w
i
P
u
t
r
a
1
,
S
u
p
r
ia
t
in
1
,
E
k
o P
r
am
on
o
1
1
I
nf
or
m
a
t
i
c
s
S
t
udy P
r
og
r
a
m
, F
a
c
ul
t
y of
C
om
put
e
r
S
c
i
e
nc
e
, A
m
i
kom
U
ni
ve
r
s
i
t
y
of
Y
ogya
ka
r
t
a
, Y
ogya
ka
r
t
a
, I
ndone
s
i
a
2
D
e
pa
r
t
m
e
nt
of
P
ha
r
m
a
c
e
ut
i
c
a
l
S
c
i
e
nc
e
s
a
nd T
e
c
hnol
ogy
,
Al
-
I
r
s
ya
d U
ni
ve
r
s
i
t
y
,
C
i
l
a
c
a
p
, I
ndone
s
i
a
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
A
ug 13, 2024
R
e
vi
s
e
d
J
a
n 3, 2026
A
c
c
e
pt
e
d
J
a
n 22, 2026
Stroke
is
one
of
the
leading
causes
of
death
worldwide,
creating
an
urgent
need
for
effective
early
detection
systems,
particularly
because
conve
ntional
methods ofte
n struggle w
ith
class imbala
nce
and produc
e biased
evalu
ations.
Previou
s
studies
have
primar
ily
focuse
d
on
accu
racy
while
overl
ooking
model
consistency,
data
pre
-
processing
quality,
and
probability
-
based
evaluatio
n.
This
study
evaluates
model
performance
under
three
cond
itions
:
original
data
using
extreme
gradient
boosting
(XGBoost)
with
scale_p
os_weight,
original
data
using
the
easy
ensemble
classifie
r,
and
class
-
balanced
data
generated
using
random
oversampling
(ROS)
,
a
daptive
synthetic
sampling
(ADASYN),
and
synthetic
minority
over
-
sa
mpling
technique
(SMOTE).
Each
model
underwen
t
missing
value
ha
ndling,
normalization,
feature
preparation,
and
hyperparameter
optimization
using
grid
search.
Performance
was
assessed
using
area
under
the
r
eceiver
operating
characteristic
curve
(AUROC),
area
under
the
precision
-
recall
curve
(AUPRC),
confidence
inter
vals,
calibration
curves,
Shapley
a
dditive
explanati
ons
(SHAP),
decision
curve
analysis
(DCA)
,
and
e
xternal
validation.
The
results
demonstrate
that
data
resampling
signif
icantly
improves
performa
nce,
with
the
XGBoost
-
SMOTE
combination
ach
ieving
the
best
results,
including
an
accuracy
of
0.99,
AUROC
of
0.998,
and
AUPRC
of
0.986,
outperforming
the
other
approaches.
This
method
provides
more
consistent
and
balanced
predictions,
supportin
g
the
applicati
on of art
ificial
intell
igence for
early str
oke risk
identi
fication.
K
e
y
w
o
r
d
s
:
D
a
ta
ba
la
nc
in
g
D
a
ta
pr
e
pr
oc
e
s
s
in
g
E
xt
r
e
m
e
gr
a
di
e
nt
boos
ti
ng
F
e
a
tu
r
e
s
e
le
c
ti
on
S
tr
oke
pr
e
di
c
ti
on
This is an
open
acce
ss artic
le unde
r the
CC BY
-
SA
license.
C
or
r
e
s
pon
di
n
g A
u
th
or
:
A
bd M
iz
w
a
r
A
. R
a
hi
m
D
e
pa
r
tm
e
nt
of
C
om
put
e
r
S
c
ie
nc
e
, A
m
ik
om
U
ni
ve
r
s
it
y of
Y
ogya
ka
r
ta
D
e
pok, S
le
m
a
n, Y
ogya
ka
r
ta
55281, I
ndone
s
ia
E
m
a
il
:
a
bdul
m
iz
w
a
r
@
a
m
ik
om
.a
c
.i
d
1.
I
N
T
R
O
D
U
C
T
I
O
N
E
ve
r
y
ye
a
r
13.7
m
il
li
on
pe
opl
e
e
xpe
r
ie
nc
e
s
tr
oke
,
a
nd
m
or
e
th
a
n
5.8
m
il
li
on
of
th
e
m
di
e
f
r
om
th
is
di
s
e
a
s
e
[
1]
.
A
c
c
or
di
ng
to
th
e
W
or
ld
H
e
a
lt
h
O
r
ga
ni
z
a
ti
on
(
W
H
O
)
,
s
tr
oke
i
s
th
e
s
e
c
ond
c
a
u
s
e
of
de
a
th
gl
oba
ll
y,
c
ont
r
ib
ut
in
g
to
a
r
ound
11%
of
to
ta
l
de
a
th
s
[
2]
.
D
is
a
bi
li
ti
e
s
th
a
t
of
te
n
oc
c
ur
a
f
te
r
a
pe
r
s
on
e
xpe
r
ie
nc
e
s
a
s
tr
oke
in
c
lu
de
s
p
e
e
c
h
pr
obl
e
m
s
,
phy
s
ic
a
l
li
m
it
a
ti
ons
,
w
e
a
kne
s
s
or
pa
r
a
ly
s
i
s
on
one
s
id
e
of
th
e
body, dif
f
ic
ul
ty
i
n gr
a
s
pi
ng or
hol
di
ng obje
c
ts
, a
nd de
c
r
e
a
s
e
d c
om
m
uni
c
a
ti
on a
bi
li
ti
e
s
[
3]
.
R
e
s
e
a
r
c
h
on
s
tr
oke
s
how
s
th
a
t
th
is
c
ondi
ti
on
r
e
qui
r
e
s
s
e
r
io
us
a
tt
e
nt
io
n
be
c
a
u
s
e
it
c
a
n
ha
ve
a
s
ig
ni
f
ic
a
nt
im
pa
c
t
on
a
c
ount
r
y
’
s
e
c
onomi
c
gr
ow
th
.
I
f
not
tr
e
a
te
d
qui
c
kl
y
a
nd
a
ppr
opr
ia
te
ly
,
s
tr
oke
c
a
n
c
a
u
s
e
s
e
r
io
us
c
om
pl
ic
a
ti
on
s
s
u
c
h
a
s
de
m
e
nt
ia
[
4]
.
D
e
m
e
nt
ia
i
s
a
m
e
d
ic
a
l
t
e
r
m
th
a
t
r
e
f
e
r
s
t
o
a
n
um
b
e
r
of
s
y
m
pt
om
s
a
s
s
oc
ia
te
d
w
it
h
a
s
i
gn
if
i
c
a
nt
d
e
c
l
in
e
in
c
ogn
it
i
ve
f
u
nc
ti
o
n,
c
a
u
s
i
ng
di
s
r
u
pt
i
on
in
a
p
e
r
s
o
n
’
s
da
il
y
a
c
t
iv
i
ti
e
s
[
5
]
.
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
. 15, No. 1, Febr
ua
r
y 2026
:
655
-
671
656
S
ym
p
to
m
s
t
h
a
t
of
te
n
o
c
c
ur
in
de
m
e
nt
i
a
i
nc
lu
d
e
di
s
tu
r
ba
n
c
e
s
i
n
m
e
m
or
y,
ju
d
gm
e
nt
s
ki
l
ls
,
pr
ob
le
m
-
s
ol
v
in
g,
la
n
gu
a
g
e
,
a
nd
in
de
pe
nd
e
n
c
e
i
n
d
a
i
ly
a
c
ti
vi
ti
e
s
[
6]
.
T
he
r
e
a
r
e
va
r
io
us
w
a
y
s
t
ha
t
c
a
n he
lp
m
e
di
c
a
l
s
ta
f
f
qui
c
kl
y i
de
nt
if
y w
he
th
e
r
s
om
e
one
i
s
e
xpe
r
ie
nc
in
g
s
tr
oke
s
ym
pt
om
s
,
one
of
w
hi
c
h
is
us
in
g
m
a
c
hi
ne
le
a
r
ni
ng
te
c
hnol
ogy.
T
he
us
e
of
th
is
te
c
hnol
ogy
ha
s
pr
ove
n
e
f
f
e
c
ti
ve
in
c
la
s
s
if
yi
ng
a
nd
opt
im
iz
in
g
th
e
de
ve
lo
pm
e
nt
of
th
e
he
a
lt
h
s
e
r
vi
c
e
s
ys
te
m
[
7]
−
[
9]
.
F
or
e
xa
m
pl
e
,
tr
e
a
ti
ng
pa
ti
e
nt
s
in
f
e
c
te
d
w
it
h
h
e
a
r
t
di
s
e
a
s
e
c
a
n
be
pr
e
di
c
te
d
f
r
om
da
ta
ge
ne
r
a
te
d
by
th
e
he
a
lt
h
in
du
s
tr
y
s
o
th
a
t
it
c
a
n he
lp
a
nd s
a
ve
s
om
e
one
’
s
l
if
e
i
n t
he
l
ong te
r
m
, a
t
le
a
s
t
it
c
a
n s
hor
te
n t
he
t
im
e
i
t
ta
ke
s
t
o f
in
d out i
f
a
pa
ti
e
nt
i
s
di
a
gnos
e
d w
it
h t
he
di
s
e
a
s
e
be
c
a
us
e
i
t
is
h
e
lp
e
d by the
m
a
c
hi
ne
l
e
a
r
ni
ng me
th
od us
e
d
[
10]
, [
11]
.
T
he
r
e
ha
ve
be
e
n
s
e
v
e
r
a
l
pr
e
vi
ous
s
tu
di
e
s
w
it
h
th
e
s
a
m
e
c
a
s
e
,
na
m
e
ly
th
e
pr
e
di
c
ti
on
of
s
tr
oke
.
E
xi
s
ti
ng
r
e
s
e
a
r
c
h
ha
s
a
ppl
ie
d
s
e
v
e
r
a
l
m
a
c
hi
ne
le
a
r
ni
ng
m
e
th
ods
f
or
c
la
s
s
if
ic
a
ti
on,
in
c
lu
di
ng
th
e
r
a
ndom
f
or
e
s
t
(
R
F
)
c
la
s
s
if
ie
r
m
e
th
od,
a
r
ti
f
ic
ia
l
ne
ur
a
l
ne
twor
k
(
ANN
)
,
s
uppor
t
ve
c
to
r
m
a
c
hi
ne
(
S
V
M
)
,
C
4.5,
a
nd
na
ïv
e
B
a
ye
s
(
N
B
)
. T
he
be
s
t
r
e
s
ul
t
f
r
om
pr
e
vi
ous
r
e
s
e
a
r
c
h w
a
s
9
8%
a
c
c
ur
a
c
y.
P
r
e
vi
ous
r
e
s
e
a
r
c
h
r
e
la
te
d
to
s
tr
oke
pr
e
di
c
ti
on,
th
is
r
e
s
e
a
r
c
h
us
e
d
th
e
e
xt
r
e
m
e
gr
a
d
ie
nt
bo
os
ti
ng
(
X
G
B
oos
t
)
m
e
th
o
d
a
nd
a
ls
o
im
pl
e
m
e
n
te
d
da
ta
p
r
e
-
p
r
oc
e
s
s
in
g
te
c
hn
iq
ue
s
in
c
lu
di
ng
la
be
l
E
nc
ode
r
,
a
nd
de
a
li
ng
w
i
th
e
m
p
ty
va
lu
e
s
us
in
g
th
e
te
c
h
ni
que
o
f
c
ha
ng
in
g
e
m
pt
y
va
lu
e
s
w
i
th
a
ve
r
a
ge
va
lu
e
s
.
T
hi
s
r
e
s
e
a
r
c
h
a
c
hi
e
ve
d
a
n
a
c
c
u
r
a
c
y
r
a
te
of
96%
[
12
]
.
O
th
e
r
s
tu
di
e
s
a
ls
o
pr
e
di
c
t
s
tr
oke
bu
t
us
e
di
f
f
e
r
e
nt
m
e
th
o
ds
,
na
m
e
l
y
X
GB
oos
t,
k
-
ne
a
r
e
s
t
ne
ig
h
bor
(
K
N
N
)
,
NB
,
RF
,
S
V
M
,
a
nd
lo
g
is
ti
c
r
e
gr
e
s
s
io
n
(
L
R
)
.
T
he
im
p
le
m
e
nt
a
ti
on
o
f
th
e
p
r
e
-
p
r
oc
e
s
s
in
g
te
c
hn
iq
ue
of
th
is
r
e
s
e
a
r
c
h
is
to
ove
r
c
om
e
m
is
s
in
g
va
lu
e
s
a
n
d
no
r
m
a
li
z
e
da
ta
.
T
he
r
e
s
ul
ts
of
th
is
r
e
s
e
a
r
c
h
r
e
a
c
he
d
91%
a
c
c
u
r
a
c
y
[
13
]
.
T
he
ne
x
t
r
e
s
e
a
r
c
h
ha
s
th
e
s
a
m
e
to
pi
c
,
na
m
e
ly
s
tr
o
ke
c
la
s
s
if
ic
a
ti
on
us
in
g
m
a
c
hi
ne
le
a
r
ni
n
g
m
e
th
ods
,
na
m
e
l
y
A
N
N
,
S
V
M
,
de
c
is
io
n
t
r
e
e
(
D
T
)
,
L
R
,
a
nd
ba
gg
in
g
a
nd boos
t
in
g.
i
m
pl
e
m
e
n
ta
ti
o
n o
f
t
e
c
h
ni
que
s
be
f
o
r
e
e
n
te
r
i
ng
th
e
c
la
s
s
if
ic
a
t
io
n
pr
oc
e
s
s
,
na
m
e
ly
c
le
a
ni
ng da
ta
,
in
c
lu
de
s
de
a
l
in
g
w
i
th
m
is
s
in
g
va
lu
e
s
a
nd
de
le
ti
ng
dup
li
c
a
te
d
a
ta
.
T
he
r
e
s
ul
ts
o
f
th
is
r
e
s
e
a
r
c
h
s
ta
te
t
ha
t
th
e
be
s
t
le
ve
l
of
a
c
c
ur
a
c
y
is
9
5%
[
14
]
.
N
e
xt
,
r
e
s
e
a
r
c
h
on
th
e
s
a
m
e
to
p
ic
a
ls
o
us
e
s
s
e
ve
r
a
l
m
a
c
hi
ne
le
a
r
ni
n
g
m
e
th
ods
in
c
l
udi
ng
NB
,
RF
,
LR
,
KNN
,
s
t
oc
ha
s
ti
c
gr
a
d
ie
nt
de
s
c
e
nt
(
S
G
D
)
,
DT
,
a
nd
m
ul
ti
la
ye
r
pe
r
c
e
p
tr
o
n
(
M
L
P
)
.
T
h
is
r
e
s
e
a
r
c
h
a
ls
o
a
p
pl
ie
s
s
e
ve
r
a
l
p
r
e
-
pr
oc
e
s
s
in
g
te
c
h
ni
que
s
to
f
ir
s
t
ove
r
c
om
e
m
is
s
in
g
va
lu
e
s
i
n
th
e
pr
oc
e
s
s
in
g
da
ta
s
e
t,
a
nd
ove
r
c
om
e
da
ta
im
ba
la
nc
e
i
n
t
he
da
t
a
s
e
t
us
in
g
s
ynt
he
ti
c
m
in
o
r
i
ty
ove
r
-
s
a
m
pl
in
g
te
c
hni
que
(
S
M
O
T
E
)
.
T
he
r
e
s
ul
ts
o
f
th
is
r
e
s
e
a
r
c
h
obt
a
in
e
d
a
n
a
c
c
ur
a
c
y
r
a
te
o
f
98
%
[
15
]
.
T
he
la
te
s
t
r
e
s
e
a
r
c
h
is
on
s
tr
oke
p
r
e
di
c
ti
on
m
a
c
hi
ne
le
a
r
ni
n
g
a
lg
o
r
it
hm
s
,
de
ve
l
opm
e
nt
a
nd
e
va
lu
a
ti
on
o
f
pr
e
di
c
ti
on
m
o
de
ls
.
T
hi
s
r
e
s
e
a
r
c
h
c
a
r
r
ie
s
o
ut
a
c
o
m
pa
r
a
ti
ve
a
na
ly
s
is
o
f
m
a
c
h
in
e
le
a
r
ni
ng
m
e
th
ods
us
in
g
da
ta
s
e
ts
w
i
th
ba
la
nc
e
d
a
nd
unba
la
nc
e
d
da
ta
c
ond
it
i
ons
. T
he
r
e
s
u
lt
s
of
t
h
is
r
e
s
e
a
r
c
h
ha
ve
t
he
be
s
t
a
c
c
ur
a
c
y
o
f
96%
us
i
ng
th
e
RF
m
e
th
od
us
in
g ba
la
nc
e
d da
ta
[
16
]
.
T
he
r
e
a
r
e
s
hor
tc
om
in
gs
in
th
a
t
pr
e
vi
ous
r
e
s
e
a
r
c
h
f
ir
s
tl
y
h
a
s
not
a
ddr
e
s
s
e
d
th
e
c
ondi
ti
on
of
unba
la
nc
e
d
da
ta
,
th
is
c
a
n
c
a
us
e
th
e
m
ode
l
to
be
bi
a
s
e
d
to
w
a
r
ds
th
e
m
a
jo
r
it
y
c
la
s
s
,
c
a
us
in
g
in
a
c
c
ur
a
te
e
va
lu
a
ti
ons
s
uc
h
a
s
m
is
le
a
di
ng
a
c
c
ur
a
c
y,
a
nd
th
e
pot
e
nt
ia
l
f
or
ove
r
f
it
ti
ng
on
m
a
jo
r
it
y
da
ta
.
A
s
a
r
e
s
ul
t,
th
e
m
ode
l
m
a
y
f
a
il
to
r
e
c
ogni
z
e
or
pr
e
di
c
t
oc
c
ur
r
e
nc
e
s
of
m
in
o
r
it
y
c
la
s
s
e
s
e
f
f
e
c
ti
ve
ly
,
r
e
duc
in
g
th
e
ge
ne
r
a
l
a
bi
li
ty
of
th
e
m
ode
l
to
a
da
pt
,
a
nd
pr
oduc
in
g
s
ubopti
m
a
l
s
ol
ut
io
ns
in
th
e
r
e
le
va
nt
a
ppl
ic
a
ti
on
c
ont
e
xt
[
17]
.
A
pa
r
t
f
r
om
th
a
t,
da
ta
nor
m
a
li
z
a
ti
on
te
c
hni
que
s
ha
ve
not
be
e
n
i
m
pl
e
m
e
nt
e
d,
a
nd
di
f
f
e
r
e
nc
e
s
in
s
c
a
le
be
twe
e
n
f
e
a
tu
r
e
s
c
a
n
s
ig
ni
f
ic
a
nt
ly
a
f
f
e
c
t
th
e
pe
r
f
or
m
a
nc
e
a
nd
s
ta
bi
li
ty
of
th
e
m
ode
l.
F
e
a
tu
r
e
s
w
it
h
a
la
r
ge
r
r
a
nge
o
f
va
lu
e
s
te
nd
to
h
a
ve
a
m
or
e
dom
in
a
nt
in
f
lu
e
nc
e
in
th
e
le
a
r
ni
ng
pr
oc
e
s
s
,
w
hi
le
f
e
a
tu
r
e
s
w
it
h
a
s
m
a
ll
e
r
r
a
nge
of
va
lu
e
s
m
a
y
pl
a
y
le
s
s
of
a
r
ol
e
or
be
ig
nor
e
d
in
de
te
r
m
in
in
g
m
o
de
l
pr
e
di
c
ti
ons
[
18]
.
T
he
la
s
t
te
c
hni
que
th
a
t
is
not
a
ppl
ie
d
is
k
-
f
ol
d
c
r
os
s
-
va
li
da
ti
on
in
e
va
lu
a
ti
ng
m
a
c
hi
n
e
le
a
r
ni
ng
m
ode
ls
,
th
e
r
e
is
a
r
is
k
th
a
t
th
e
e
va
lu
a
ti
on
of
m
ode
l
pe
r
f
or
m
a
nc
e
w
il
l
be
in
c
ons
is
te
nt
a
nd
s
ubj
e
c
ti
ve
.
B
y
onl
y
doi
ng
one
di
vi
s
io
n
of
tr
a
in
in
g
da
ta
a
nd
va
li
da
ti
on
da
ta
,
e
va
lu
a
ti
on
r
e
s
ul
ts
c
a
n
be
to
o
opt
im
is
ti
c
or
pe
s
s
im
is
ti
c
de
pe
ndi
ng
on
how
th
e
da
ta
is
r
a
ndoml
y di
vi
de
d, t
hus
not
pr
ovi
di
ng a
n a
c
c
ur
a
te
pi
c
tu
r
e
of
how
w
e
ll
t
he
m
ode
l
c
a
n pr
e
di
c
t
uns
e
e
n da
ta
[
19]
.
T
o
ge
t
g
oo
d
a
c
c
u
r
a
c
y
,
r
e
l
y
o
n
one
o
f
th
e
da
t
a
p
r
e
-
p
r
o
c
e
s
s
i
ng
te
c
h
ni
que
s
,
na
m
e
ly
f
e
a
tu
r
e
s
e
le
c
ti
on
w
he
n
us
in
g
t
hi
s
m
e
th
od
in
t
he
c
la
s
s
i
f
ic
a
t
io
n
pr
oc
e
s
s
.
R
e
c
e
nt
r
e
s
e
a
r
c
h
o
n
th
e
e
f
f
e
c
t
of
f
e
a
t
u
r
e
s
e
le
c
ti
o
n
on
th
e
a
c
c
u
r
a
c
y
of
m
a
c
h
in
e
le
a
r
n
in
g
m
ode
ls
ha
s
m
a
de
a
m
a
j
or
c
ont
r
ib
u
ti
on
to
t
he
i
de
n
ti
f
ic
a
t
io
n
p
r
oc
e
s
s
[
20
]
.
A
pa
r
t
f
r
om
th
a
t,
th
e
a
c
c
u
r
a
c
y
o
f
t
he
m
ode
l
a
c
hi
e
ve
s
g
ood
r
e
s
ul
ts
by
a
pp
ly
in
g
th
e
da
t
a
ba
la
nc
in
g
m
e
th
od,
w
hi
c
h
ha
s
be
e
n
pr
ove
n
in
r
e
s
e
a
r
c
h
r
e
ga
r
di
ng
th
e
im
pa
c
t
of
t
he
da
ta
b
a
la
nc
i
ng
a
p
pr
oa
c
h
w
i
th
a
c
a
s
e
s
tu
dy
[
2
1]
.
A
no
th
e
r
te
c
h
ni
que
t
ha
t
c
a
n
be
a
p
pl
ie
d
to
a
c
h
ie
ve
go
od
e
va
l
ua
t
io
n
s
c
o
r
e
s
is
da
ta
no
r
m
a
l
iz
a
ti
on.
T
h
is
te
c
h
ni
que
ha
s
a
ls
o
be
e
n
p
r
ove
n
to
be
a
b
le
to
in
c
r
e
a
s
e
a
c
c
u
r
a
c
y
in
t
he
c
la
s
s
i
f
i
c
a
t
io
n
pr
oc
e
s
s
.
T
hi
s
ha
s
be
e
n
done
i
n
r
e
s
e
a
r
c
h
in
v
e
s
t
ig
a
ti
ng
th
e
i
m
p
a
c
t
o
f
da
ta
n
or
m
a
li
z
a
t
io
n
o
n
c
la
s
s
i
f
ic
a
t
io
n
pe
r
f
or
m
a
nc
e
[
2
2]
.
T
h
e
la
s
t
one
is
t
he
i
m
p
le
m
e
nt
a
t
io
n
o
f
th
e
k
-
f
ol
d
c
r
os
s
-
va
li
da
t
io
n
te
c
h
ni
que
.
T
hi
s
te
c
h
ni
que
is
n
ot
a
f
unc
ti
o
n
th
a
t
i
nc
r
e
a
s
e
s
a
c
c
u
r
a
c
y
d
i
r
e
c
t
ly
,
b
ut
r
a
th
e
r
a
n
e
va
lu
a
ti
on
te
c
hni
que
t
ha
t
he
lp
s
in
va
l
id
a
t
in
g
m
ode
l
pe
r
f
o
r
m
a
nc
e
be
tt
e
r
[
23
]
.
T
hi
s
s
tu
dy
c
la
s
s
if
ie
d
s
tr
oke
r
is
k
th
r
ough
a
s
e
r
ie
s
of
s
ta
ge
s
c
ons
i
s
ti
ng
of
da
ta
pr
e
-
pr
oc
e
s
s
in
g
(
im
put
a
ti
on
of
m
is
s
in
g
body
m
a
s
s
in
de
x
(
B
M
I
)
va
lu
e
s
,
ha
ndl
in
g
out
li
e
r
s
us
in
g
th
e
in
te
r
qua
r
ti
le
r
a
nge
(
I
Q
R
)
m
e
th
od,
e
nc
odi
ng
c
a
te
gor
ic
a
l
va
r
ia
bl
e
s
us
in
g
L
a
be
lE
nc
ode
r
,
a
nd
nor
m
a
li
z
in
g
num
e
r
ic
a
l
f
e
a
tu
r
e
s
w
it
h
min
-
m
a
x
s
c
a
li
ng
)
,
di
vi
di
ng
th
e
da
t
a
in
to
tr
a
in
in
g
a
nd
te
s
t
da
ta
,
a
nd
a
ppl
yi
ng
va
r
io
us
ba
la
n
c
in
g
te
c
hni
que
s
to
th
e
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
St
r
ok
e
pr
e
di
c
ti
on us
in
g data bal
anc
in
g m
e
th
od and e
x
t
r
e
m
e
g
r
adi
e
nt
boos
ti
ng
(
A
bd M
iz
w
ar
A
. R
ahi
m
)
657
tr
a
in
in
g
da
ta
us
in
g
S
M
O
T
E
,
a
da
pt
iv
e
s
ynt
he
ti
c
s
a
m
pl
in
g
(
A
D
A
S
Y
N
)
,
a
nd
r
a
ndom
ove
r
s
a
m
pl
in
g
(
R
O
S
)
.
T
he
m
ode
l
w
a
s
th
e
n
de
ve
lo
p
e
d
us
in
g
th
e
X
G
B
oos
t
a
lg
or
i
th
m
w
it
h
hype
r
pa
r
a
m
e
te
r
tu
ni
ng,
in
c
lu
di
ng
va
r
ia
ti
ons
of
th
e
s
c
a
le
_pos
_w
e
ig
ht
pa
r
a
m
e
te
r
,
a
nd
th
e
E
a
s
yE
n
s
e
m
bl
e
m
e
th
od
a
s
a
c
om
pa
r
is
on.
P
e
r
f
or
m
a
nc
e
e
va
lu
a
ti
on
w
a
s
c
onduc
te
d
us
in
g
th
e
a
r
e
a
unde
r
th
e
r
e
c
e
iv
e
r
ope
r
a
ti
ng
c
ha
r
a
c
te
r
is
ti
c
c
ur
ve
(
A
U
R
O
C
)
a
nd
a
r
e
a
unde
r
th
e
pr
e
c
is
io
n
-
r
e
c
a
ll
c
ur
ve
(
A
U
P
R
C
)
m
e
tr
ic
s
,
c
a
li
br
a
ti
on
pl
ot
s
,
S
ha
pl
e
y
a
ddi
ti
ve
e
xpl
a
na
ti
ons
(
S
H
A
P
)
in
te
r
pr
e
ta
bi
li
ty
a
na
ly
s
is
,
boot
s
tr
a
p
c
onf
id
e
nc
e
in
te
r
va
ls
,
a
nd
e
xt
e
r
na
l
va
li
da
ti
on
to
de
te
r
m
in
e
th
e
opt
im
a
l
m
ode
l
f
or
pr
e
di
c
ti
ng s
tr
oke
r
is
k.
2.
M
E
T
H
O
D
T
h
is
s
e
c
t
io
n
e
xp
la
in
s
th
e
r
e
s
e
a
r
c
h
f
lo
w
r
e
ga
r
di
n
g
di
s
e
a
s
e
c
l
a
s
s
if
ic
a
ti
on,
s
t
a
r
ti
ng
f
r
om
d
a
t
a
a
c
qu
i
s
it
i
on
,
da
t
a
pr
e
-
pr
o
c
e
s
s
i
ng
,
u
s
e
of
th
e
X
G
B
oo
s
t
m
o
de
l,
a
n
d
e
va
lu
a
ti
on
of
m
o
de
l
p
e
r
f
or
m
a
nc
e
a
s
in
F
ig
ur
e
1
.
F
ig
ur
e
1
c
l
e
a
r
ly
il
l
us
tr
a
te
s
t
h
e
ov
e
r
a
ll
pr
o
c
e
s
s
i
n
a
v
is
ua
l
m
a
nn
e
r
.
T
hu
s
,
r
e
a
d
e
r
s
c
a
n
e
a
s
il
y
f
ol
lo
w
t
he
m
e
th
od
ol
o
gy.
F
ig
ur
e
1.
R
e
s
e
a
r
c
h f
lo
w
2
.1.
D
at
as
e
t
T
hi
s
r
e
s
e
a
r
c
h
d
a
ta
s
e
t
w
a
s
ta
ke
n
f
r
om
th
e
K
a
ggl
e
da
t
a
s
e
t
[
24]
.
T
hi
s
da
ta
s
e
t
ha
s
5
,
110
d
a
ta
poi
nt
s
,
c
ons
is
ts
of
12
a
tt
r
ib
ut
e
s
,
10
in
de
p
e
nde
nt
va
r
ia
bl
e
s
a
s
f
e
a
tu
r
e
s
,
a
nd
one
de
pe
nde
nt
va
r
ia
bl
e
a
s
a
c
la
s
s
la
be
l.
T
he
10
in
de
p
e
nde
nt
va
r
ia
bl
e
s
in
que
s
ti
on
a
r
e
ge
nd
e
r
,
a
ge
,
hype
r
te
ns
io
n,
he
a
r
t_
di
s
e
a
s
e
,
e
ve
r
_m
a
r
r
ie
d,
w
or
k_t
ype
,
r
e
s
id
e
nc
e
_t
ype
,
a
vg_gluc
os
e
_l
e
ve
l,
bm
i,
a
nd
s
m
ok
in
g_s
ta
tu
s
.
T
he
la
b
e
l
is
th
e
s
tr
oke
a
tt
r
ib
ut
e
in
th
is
da
ta
s
e
t.
T
he
c
la
s
s
ha
s
two
va
lu
e
s
:
0,
w
hi
c
h
m
e
a
n
s
th
e
r
e
i
s
no
in
di
c
a
ti
on
of
s
tr
oke
,
a
nd
1,
w
hi
c
h
m
e
a
n
s
th
e
r
e
i
s
a
n i
ndi
c
a
ti
on of
s
tr
oke
. T
a
bl
e
1
da
ta
s
e
t
de
s
c
r
ib
e
s
t
he
da
ta
s
e
t
f
e
a
tu
r
e
i
nf
or
m
a
ti
on.
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
. 15, No. 1, Febr
ua
r
y 2026
:
655
-
671
658
T
a
bl
e
1.
D
a
ta
s
e
t
No
F
e
a
t
ur
e
I
nf
or
m
a
t
i
on
1
G
e
nde
r
F
e
m
a
l
e
M
a
l
e
2
A
ge
A
ge
3
H
ype
r
t
e
ns
i
1 hype
r
t
e
ns
i
on
0 doe
s
not
ha
ve
hype
r
t
e
ns
i
4
he
a
r
t
_di
s
e
a
s
e
1 H
a
ve
he
a
r
t
di
s
e
a
s
e
0 doe
s
not
ha
ve
he
a
r
t
di
s
e
a
s
e
5
e
ve
r
_m
a
r
r
i
e
d
1 m
e
a
ns
m
a
r
r
i
e
d
0 m
e
a
ns
not
m
a
r
r
i
e
d
6
w
or
k_t
ype
C
hi
l
dr
e
n
P
e
r
s
ona
l
N
e
ve
r
w
or
k
G
ove
r
nm
e
nt
w
or
k
E
nt
r
e
pr
e
ne
ur
7
R
e
s
i
de
nc
e
_t
ype
R
ur
a
l
U
r
ba
n
8
a
vg_gl
uc
os
e
_l
e
ve
l
A
ve
r
a
ge
gl
uc
os
e
l
e
ve
l
9
bm
i
body m
a
s
s
i
nde
x
10
s
m
oki
ng_s
t
a
t
us
N
e
ve
r
s
m
oke
d
U
s
e
d t
o s
m
oke
11
S
t
r
oke
0 (
C
l
a
s
s
doe
s
not
i
ndi
c
a
t
e
s
t
r
oke
)
1 (
C
l
a
s
s
i
ndi
c
a
t
e
d
s
t
r
oke
)
2.2
.
P
r
e
-
p
r
oc
e
s
s
in
g d
at
a
I
n
th
is
s
tu
dy,
th
e
da
ta
pr
e
pr
oc
e
s
s
in
g
s
ta
g
e
c
on
s
is
ts
of
f
our
m
a
in
s
te
ps
.
T
h
e
s
e
s
te
p
s
in
c
lu
de
da
ta
im
put
a
ti
on,
c
a
te
gor
ic
a
l
e
nc
odi
ng,
out
li
e
r
ha
ndl
in
g,
a
nd
f
e
a
tu
r
e
s
c
a
li
ng.
T
he
s
e
s
te
ps
a
im
to
e
ns
ur
e
th
a
t
th
e
da
ta
us
e
d i
n t
he
t
r
a
in
in
g m
ode
l
is
c
le
a
n, c
ons
i
s
te
nt
, a
nd r
e
a
dy f
or
pr
oc
e
s
s
in
g by ma
c
hi
ne
l
e
a
r
ni
ng a
lg
or
it
hm
s
.
2.2.1.
D
at
a
im
p
u
t
at
io
n
D
a
t
a
s
e
t
f
e
a
tu
r
e
s
th
a
t
ha
ve
e
m
pt
y
va
l
ue
s
,
na
m
e
l
y
t
he
B
M
I
f
e
a
tu
r
e
,
w
e
c
ho
s
e
a
te
c
h
ni
q
ue
to
o
ve
r
c
o
m
e
th
i
s
by
c
h
a
n
gi
n
g
t
he
e
m
pt
y
va
l
u
e
s
to
th
e
a
v
e
r
a
g
e
va
l
u
e
of
B
M
I
ha
vi
n
g
a
s
tr
ok
e
a
n
d
B
M
I
n
ot
ha
vi
n
g
a
s
tr
ok
e
.
T
h
e
e
m
pt
y
v
a
lu
e
s
of
th
e
e
n
ti
r
e
d
a
t
a
s
e
t
f
e
a
tu
r
e
s
c
a
n
b
e
s
e
e
n
in
T
a
bl
e
2.
T
a
b
le
2
s
h
ow
s
th
a
t
t
he
e
m
pt
y
v
a
l
ue
i
s
201
d
a
t
a
,
i
f
t
h
e
e
m
pt
y
va
lu
e
i
n
th
e
B
M
I
f
e
a
tu
r
e
h
a
s
c
l
a
s
s
0,
t
he
n
t
he
e
m
p
ty
v
a
l
ue
i
s
c
h
a
ng
e
d t
o
th
e
a
v
e
r
a
ge
B
M
I
va
l
ue
o
f
c
l
a
s
s
0,
a
nd
v
ic
e
v
e
r
s
a
if
t
he
e
m
p
ty
v
a
lu
e
in
t
he
B
M
I
f
e
a
t
ur
e
ha
s
c
la
s
s
1
,
th
e
n
th
e
e
m
pt
y
v
a
lu
e
i
s
c
h
a
ng
e
d t
o t
he
a
ve
r
a
ge
B
M
I
v
a
lu
e
of
c
l
a
s
s
1
.
T
he
a
ve
r
a
g
e
B
M
I
v
a
lu
e
s
of
c
l
a
s
s
e
s
0
a
nd
1
c
a
n
b
e
s
e
e
n i
n
T
a
b
le
3.
T
a
bl
e
2. F
e
a
tu
r
e
s
t
h
a
t
ha
ve
a
n e
m
pt
y va
lu
e
c
ondi
ti
on
No
.
F
e
a
t
ur
e
N
um
be
r
of
e
m
pt
y va
l
ue
s
1
G
e
nde
r
0
2
A
ge
0
3
H
ype
r
t
e
ns
i
0
4
he
a
r
t
_di
s
e
a
s
e
0
5
e
ve
r
_m
a
r
r
i
e
d
0
6
w
or
k_t
ype
0
7
R
e
s
i
de
nc
e
_t
ype
0
8
a
vg_gl
uc
os
e
_l
e
ve
l
0
9
bm
i
201
10
s
m
oki
ng_s
t
a
t
us
0
11
S
t
r
oke
0
T
a
bl
e
3. A
ve
r
a
ge
B
M
I
c
la
s
s
1 a
nd
B
M
I
c
la
s
s
0
No
.
F
e
a
t
ur
e
A
ve
r
a
ge
va
l
ue
1
B
M
I
i
ndi
c
a
t
e
s
s
t
r
oke
(
1)
30.47
2
B
M
I
doe
s
not
i
ndi
c
a
t
e
s
t
r
oke
(
0)
28.82
T
a
bl
e
3
s
how
s
a
c
om
pa
r
is
on
of
th
e
a
ve
r
a
ge
B
M
I
va
lu
e
s
be
t
w
e
e
n
two
c
la
s
s
e
s
:
c
la
s
s
1
(
in
di
c
a
ti
ng
s
tr
oke
)
a
nd
c
la
s
s
0
(
not
in
di
c
a
ti
ng
s
tr
oke
)
.
B
a
s
e
d
on
th
e
c
a
lc
ul
a
ti
on
r
e
s
ul
ts
,
in
di
vi
dua
ls
w
ho
e
xpe
r
ie
nc
e
d
a
s
tr
oke
ha
d
a
n
a
ve
r
a
g
e
B
M
I
of
30.47,
w
hi
le
in
di
vi
dua
ls
w
ho
di
d
not
e
xpe
r
ie
nc
e
a
s
tr
oke
ha
d
a
n
a
ve
r
a
ge
B
M
I
of
28.82.
A
ll
s
ta
ge
s
of
t
hi
s
pr
oc
e
s
s
c
a
n be
s
e
e
n i
n F
ig
ur
e
2.
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
St
r
ok
e
pr
e
di
c
ti
on us
in
g data bal
anc
in
g m
e
th
od and e
x
t
r
e
m
e
g
r
adi
e
nt
boos
ti
ng
(
A
bd M
iz
w
ar
A
. R
ahi
m
)
659
F
ig
ur
e
2. V
is
ua
li
z
a
ti
on of
t
he
e
xpe
r
im
e
nt
s
e
tu
p f
or
f
il
li
ng i
n bl
a
nk va
lu
e
s
i
n
th
e
B
M
I
f
e
a
tu
r
e
2.2.2.
C
at
e
gor
ic
al
e
n
c
od
in
g
A
ppl
yi
ng
th
e
L
a
be
lE
nc
ode
r
te
c
hni
que
to
c
onve
r
t
c
a
te
gor
ic
a
l
f
e
a
tu
r
e
s
in
to
num
e
r
ic
da
ta
,
onl
y
a
f
e
w
a
tt
r
ib
ut
e
s
a
r
e
c
ha
ng
e
d
in
th
is
l
a
be
li
ng,
in
c
lu
di
ng
ge
nd
e
r
,
e
ve
r
_m
a
r
r
ie
d,
w
or
k_t
ype
,
r
e
s
id
e
nc
e
_t
ype
,
a
nd
s
m
oki
ng_s
ta
tu
s
.
T
he
r
e
s
ul
ts
of
th
is
pr
oc
e
s
s
c
a
n
be
s
e
e
n
in
T
a
b
le
4.
F
ig
ur
e
3
s
how
s
th
e
c
om
pl
e
te
pr
oc
e
s
s
of
c
onve
r
ti
ng
c
a
te
gor
ic
a
l
va
lu
e
s
to
num
e
r
ic
va
lu
e
s
us
in
g
L
a
be
lE
n
c
ode
r
.
S
in
c
e
m
a
c
hi
ne
le
a
r
ni
ng
a
lg
or
it
hm
s
li
ke
X
G
B
oos
t
c
a
n only pr
oc
e
s
s
nume
r
ic
da
t
a
, l
a
be
l
e
nc
odi
ng t
e
c
hni
que
s
a
r
e
us
e
d t
o t
r
a
ns
f
or
m
t
hos
e
f
e
a
tu
r
e
s
.
T
a
bl
e
4. L
a
be
lE
nc
ode
r
r
e
s
ul
ts
G
e
nde
r
A
ge
H
ype
r
t
e
ns
i
H
e
a
r
t
_
di
s
e
a
s
e
E
ve
r
_
m
a
r
r
i
e
d
W
or
k
_
t
ype
R
e
s
i
de
nc
e
_
t
ype
A
vg
_
G
l
uc
os
e
_
l
e
ve
l
bm
i
S
m
oki
ng
_
s
t
a
t
us
S
t
r
oke
1
67
0
1
1
2
1
228.69
36.6
1
1
0
61
0
0
1
3
0
202.21
30.47
2
1
1
80
0
1
1
2
0
105.92
32.5
2
1
0
9
0
0
1
2
1
171.23
34.4
3
1
0
79
1
0
1
3
0
174.12
24
2
1
0
82
0
1
0
2
1
215.6
24.9
2
0
0
54
0
0
1
0
0
91.61
25.2
2
0
0
49
0
0
1
2
0
138.16
19.4
2
0
0
24
0
0
1
2
1
75.23
29
2
0
0
37
0
0
1
2
0
75.18
48.2
1
0
1
34
0
1
1
2
1
106.23
28.82
1
0
F
ig
ur
e
3. V
is
ua
li
z
a
ti
on of
th
e
e
xpe
r
im
e
nt
s
e
tu
p f
or
c
onve
r
ti
ng
c
a
te
gor
ic
a
l
va
lu
e
s
t
o nume
r
ic
va
lu
e
s
w
it
h L
a
be
lE
nc
ode
r
2.2.3.
O
u
t
li
e
r
h
an
d
li
n
g
T
o
a
voi
d
th
e
e
xt
r
e
m
e
in
f
lu
e
nc
e
of
unr
e
pr
e
s
e
nt
a
ti
ve
va
lu
e
s
[
25]
,
out
li
e
r
de
te
c
ti
on
a
nd
ha
ndl
in
g
a
r
e
pe
r
f
or
m
e
d
on
num
e
r
ic
f
e
a
tu
r
e
s
s
uc
h
a
s
a
vg_gluc
os
e
_l
e
v
e
l
a
nd
B
M
I
.
T
hi
s
pr
oc
e
s
s
us
e
s
th
e
I
Q
R
m
e
th
od,
de
te
r
m
in
in
g t
he
l
ow
e
r
a
nd uppe
r
bounds
ba
s
e
d on the
(
1)
.
=
3
−
1
(
1)
I
nf
or
m
a
ti
on:
Q1
is
f
ir
s
t
qua
r
ti
le
(
25
th
pe
r
c
e
nt
il
e
)
→
lo
w
e
r
li
m
i
t
of
th
e
f
ir
s
t
25%
of
da
ta
.
3
is
t
hi
r
d
qua
r
ti
le
(
75
th
pe
r
c
e
nt
il
e
)
→
uppe
r
li
m
it
of
th
e
la
s
t
25%
of
da
ta
.
I
Q
R
is
i
nt
e
r
qua
r
ti
le
r
a
nge
,
s
how
s
th
e
s
pr
e
a
d
of
th
e
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
. 15, No. 1, Febr
ua
r
y 2026
:
655
-
671
660
m
id
dl
e
da
ta
(
50%
of
th
e
da
ta
)
.
A
f
te
r
th
e
Q1
a
nd
Q3
va
lu
e
s
a
r
e
obt
a
in
e
d,
th
e
lo
w
e
r
a
nd
uppe
r
li
m
it
s
f
o
r
de
te
c
ti
ng outl
ie
r
s
a
r
e
c
a
lc
ul
a
te
d
a
s
(
2)
a
nd (
3)
.
=
1
−
1
.
5
×
(
2)
=
3
+
1
.
5
×
(
3)
I
nf
or
m
a
t
io
n
:
l
ow
e
r
bou
nd
i
s
m
i
ni
m
um
v
a
lu
e
t
ha
t
i
s
s
ti
l
l
c
on
s
i
de
r
e
d nor
m
a
l
,
u
pp
e
r
bou
nd
i
s
t
he
m
a
xi
m
um
v
a
lu
e
th
a
t
i
s
s
ti
l
l
c
on
s
i
d
e
r
e
d n
or
m
a
l
,
a
nd
v
a
l
ue
s
ou
ts
id
e
t
he
r
a
ng
e
lo
w
e
r
bo
und
,
upp
e
r
bo
un
d
a
r
e
c
on
s
i
de
r
e
d
o
ut
l
ie
r
s
.
2.2.4.
F
e
at
u
r
e
s
c
al
in
g u
s
e
s
t
h
e
m
in
-
m
ax n
or
m
al
iz
at
io
n
t
e
c
h
n
iq
u
e
T
o
e
n
s
ur
e
th
e
va
lu
e
s
of
w
id
e
ly
di
f
f
e
r
e
nt
da
t
a
s
e
t
f
e
a
tu
r
e
s
h
a
ve
a
uni
f
or
m
s
c
a
le
w
he
n
bui
ld
in
g
m
a
c
hi
ne
le
a
r
ni
ng
m
ode
ls
,
w
e
ne
e
d
to
s
t
a
nda
r
di
z
e
.
O
ne
of
th
e
s
ta
nda
r
di
z
a
ti
on
te
c
hni
que
s
u
s
e
d
is
min
-
m
a
x
nor
m
a
li
z
a
ti
on,
w
hi
c
h
c
ha
nge
s
th
e
or
ig
in
a
l
da
ta
li
ne
a
r
ly
s
o
th
a
t
th
e
va
lu
e
s
be
twe
e
n
a
tt
r
ib
ut
e
s
a
r
e
b
a
la
nc
e
d.
W
it
h
th
is
nor
m
a
li
z
a
ti
on,
a
tt
r
ib
ut
e
va
lu
e
s
a
r
e
tr
a
ns
f
or
m
e
d
s
uc
h
th
a
t
th
e
y
r
a
nge
be
twe
e
n
0
a
nd
1,
a
ll
ow
in
g
f
or
c
ons
is
te
nt
c
om
p
a
r
is
ons
a
c
r
os
s
a
tt
r
ib
ut
e
s
[
26]
.
N
um
e
r
ic
f
e
a
tu
r
e
s
w
it
h
w
id
e
ly
va
r
yi
ng
va
lu
e
r
a
ng
e
s
,
s
uc
h
a
s
a
ge
,
a
ve
r
a
g
e
gl
uc
o
s
e
le
v
e
l,
a
nd
body
w
e
ig
ht
,
w
e
r
e
nor
m
a
li
z
e
d
us
in
g
th
e
min
-
m
a
x
nor
m
a
li
z
a
ti
on
m
e
th
od
to
ke
e
p
e
a
c
h
a
tt
r
ib
ut
e
’
s
va
lu
e
w
it
hi
n
th
e
r
a
nge
of
0
to
1,
pr
e
ve
nt
in
g
a
ny
s
in
gl
e
f
e
a
tu
r
e
f
r
om
dom
in
a
ti
ng
th
e
m
ode
l
’
s
le
a
r
ni
ng
pr
oc
e
s
s
due
to
di
f
f
e
r
e
nc
e
s
in
s
c
a
le
.
O
nl
y
th
e
s
e
f
e
a
tu
r
e
s
r
e
qui
r
e
nor
m
a
li
z
a
ti
on
be
c
a
us
e
th
e
y
a
r
e
c
ont
in
uous
num
e
r
ic
va
lu
e
s
w
it
h
s
ig
ni
f
ic
a
nt
va
r
ia
ti
ons
in
va
lu
e
s
.
O
th
e
r
f
e
a
tu
r
e
s
,
s
u
c
h
a
s
ge
nd
e
r
,
s
m
oki
ng
s
ta
tu
s
,
oc
c
upa
ti
on,
r
e
s
id
e
n
c
e
ty
pe
,
a
nd
s
e
x,
w
e
r
e
c
a
te
gor
iz
e
d
u
s
in
g
L
a
be
lE
nc
ode
r
,
a
nd
th
e
hyp
e
r
te
ns
io
n
a
nd
he
a
r
t
di
s
e
a
s
e
f
e
a
tu
r
e
s
w
e
r
e
a
lr
e
a
dy
bi
na
r
y
da
ta
s
e
ts
of
0
a
n
d
1
a
nd
th
e
r
e
f
or
e
di
d
not
r
e
qui
r
e
a
ddi
ti
ona
l
nor
m
a
li
z
a
ti
on
[
27]
.
T
hi
s
m
e
th
od c
a
n us
e
t
he
(
4
).
=
(
−
)
(
−
)
−
(
4
)
W
he
r
e
N
is
th
e
nor
m
a
li
z
e
d
va
lu
e
(
th
e
ne
w
va
lu
e
a
f
te
r
tr
a
ns
f
or
m
a
ti
on)
;
x
is
th
e
or
ig
in
a
l
va
lu
e
of
a
n
a
tt
r
ib
ut
e
be
f
or
e
nor
m
a
li
z
a
ti
on
;
m
in
V
al
ue
is
th
e
m
a
xi
m
um
va
lu
e
of
th
e
a
t
tr
ib
ut
e
,
a
nd
m
ax
V
al
ue
is
th
e
nor
m
a
li
z
e
d
va
lu
e
in
th
e
r
a
nge
0,
1.
m
in
R
ange
is
th
e
lo
w
e
r
bound
of
th
e
de
s
ir
e
d
ne
w
s
c
a
le
r
a
ng
e
(
us
ua
ll
y
0)
;
a
nd
m
ax
R
ange
is
th
e
uppe
r
bound
of
th
e
de
s
ir
e
d
ne
w
s
c
a
le
r
a
nge
(
us
u
a
ll
y
1)
.
T
he
e
nt
ir
e
nor
m
a
li
z
a
ti
on
pr
oc
e
s
s
c
a
n
be
s
e
e
n
in
F
ig
ur
e
4.
F
ig
ur
e
4. V
is
ua
li
z
a
ti
on of
t
he
e
xpe
r
im
e
nt
s
e
tu
p f
or
nume
r
ic
f
e
a
tu
r
e
s
t
ha
t
ha
ve
ve
r
y di
f
f
e
r
e
nt
va
lu
e
r
a
nge
s
us
in
g m
in
-
m
a
x nor
m
a
li
z
a
ti
on
2.3.
S
p
li
t
d
a
t
a
T
r
a
in
in
g/
te
s
t
da
ta
s
pl
it
di
vi
de
s
th
e
da
ta
s
e
t
in
to
tr
a
in
in
g
da
ta
a
nd
te
s
t
da
ta
.
T
he
s
pl
it
in
th
is
s
tu
dy
di
vi
de
s
th
e
tr
a
in
in
g
da
ta
a
nd
te
s
t
d
a
ta
in
to
70/
30
us
in
g
a
s
tr
a
ti
f
ie
d
s
pl
it
to
m
a
in
ta
in
c
la
s
s
pr
opor
ti
on
ba
la
nc
e
.
T
he
s
tr
a
ti
f
ie
d
5
-
f
ol
d
c
r
os
s
-
va
li
da
ti
on
pr
oc
e
s
s
is
a
ppl
ie
d
onl
y
to
th
e
tr
a
in
in
g
da
ta
to
e
va
lu
a
te
a
nd
opt
im
is
e
m
ode
l
pe
r
f
or
m
a
nc
e
.
I
n
c
ont
r
a
s
t,
th
e
te
s
t
da
ta
is
us
e
d
onc
e
in
th
e
f
in
a
l
s
ta
ge
to
m
e
a
s
ur
e
th
e
m
ode
l'
s
ge
ne
r
a
li
s
a
ti
on
a
bi
li
ty
w
it
hout
th
e
r
is
k
of
da
ta
le
a
ka
ge
.
W
it
h
th
is
di
vi
s
io
n,
th
e
pr
im
a
r
y
obj
e
c
ti
ve
is
to
e
va
lu
a
te
th
e
m
ode
l
’
s
pe
r
f
or
m
a
nc
e
w
he
n
m
a
ki
ng
pr
e
di
c
ti
ons
us
in
g
a
to
ta
l
of
1
,
536
te
s
t
da
ta
poi
nt
s
.
I
n
ge
ne
r
a
l,
m
a
c
hi
ne
le
a
r
ni
ng
m
ode
ls
c
a
n
pr
oduc
e
good
a
c
c
ur
a
c
y
if
th
e
y
ha
ve
a
s
m
a
ll
a
m
ount
of
te
s
t
da
ta
[
28]
.
T
he
r
e
f
or
e
,
in
th
is
s
tu
dy,
w
e
in
c
r
e
a
s
e
d
th
e
a
m
ount
of
t
e
s
t
d
a
ta
a
nd
e
va
lu
a
te
d
w
h
e
th
e
r
th
e
m
ode
l
pr
oduc
e
d
s
a
ti
s
f
a
c
to
r
y
r
e
s
ul
ts
.
T
a
bl
e
5
il
lu
s
tr
a
te
s
t
h
e
da
ta
di
s
tr
ib
ut
io
n.
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
St
r
ok
e
pr
e
di
c
ti
on us
in
g data bal
anc
in
g m
e
th
od and e
x
t
r
e
m
e
g
r
adi
e
nt
boos
ti
ng
(
A
bd M
iz
w
ar
A
. R
ahi
m
)
661
T
a
bl
e
5. R
e
s
ul
ts
of
f
e
a
tu
r
e
s
e
l
e
c
ti
on us
in
g t
he
ba
c
kw
a
r
d e
li
m
in
a
ti
on me
th
od
F
ol
d=
k
1
–
k
5
A
m
ount
of
t
r
a
i
ni
ng da
ta
T
e
s
t
da
t
a
a
m
ount
A
m
ount
of
da
t
a
3
,
582
1
,
536
5
,
118
2.3.1.
S
am
p
li
n
g
T
hi
s
r
e
s
e
a
r
c
h
da
ta
s
e
t
e
xpe
r
ie
nc
e
d
unb
a
la
nc
e
d
d
a
ta
c
ondi
ti
ons
,
w
e
a
ppl
ie
d
s
e
ve
r
a
l
m
e
th
od
s
,
one
of
w
hi
c
h
w
a
s
th
e
u
s
e
of
th
e
S
M
O
T
E
m
e
th
od
to
ove
r
c
om
e
th
is
. T
h
is
m
e
th
od
c
a
n
ha
ndl
e
a
c
la
s
s
im
ba
la
nc
e
in
th
e
da
ta
s
e
t
by
b
a
la
nc
in
g
th
e
num
be
r
of
m
in
or
it
y
c
la
s
s
e
s
s
o
th
a
t
it
is
c
om
pa
r
a
bl
e
to
th
e
m
a
jo
r
it
y
c
la
s
s
.
T
h
e
r
e
s
ul
t
of
t
hi
s
t
e
c
hni
que
i
s
s
ynt
he
ti
c
d
a
ta
c
r
e
a
te
d ba
s
e
d on the
K
N
N
[
29]
a
s
de
f
in
e
d i
n (
5
).
=
+
(
−
)
−
(
5
)
N
e
xt
,
a
ppl
yi
ng
th
e
R
O
S
m
e
th
od,
th
is
m
e
th
od
c
a
n
a
ls
o
pe
r
f
or
m
da
ta
ba
la
nc
in
g.
T
hi
s
m
e
th
od
w
or
ks
by
r
a
ndoml
y
c
opyi
ng
th
e
m
in
or
it
y
c
la
s
s
(
f
e
w
e
r
c
la
s
s
e
s
)
unt
il
th
e
num
be
r
is
c
om
pa
r
a
bl
e
to
th
e
m
a
jo
r
it
y
(
m
or
e
c
la
s
s
e
s
)
[
30]
.
F
ur
th
e
r
m
o
r
e
,
th
e
im
pl
e
m
e
nt
a
ti
on
of
th
e
r
a
nd
om
unde
r
s
a
m
pl
in
g
(
R
U
S
)
m
e
th
od
w
or
ks
by
r
a
ndoml
y r
e
duc
in
g t
he
numbe
r
of
m
a
jo
r
it
y
c
la
s
s
e
s
s
o t
ha
t
it
i
s
pr
opor
ti
ona
l
to
t
he
numbe
r
of
m
in
o
r
it
y c
la
s
s
e
s
.
T
hi
s
a
ppr
oa
c
h r
e
duc
e
s
t
he
m
a
jo
r
it
y r
e
pr
e
s
e
nt
a
ti
on i
n t
he
da
t
a
s
e
t
[
31]
.
T
he
la
tt
e
r
im
pl
e
m
e
nt
s
th
e
A
D
A
S
Y
N
m
e
th
od,
th
is
m
e
th
od
ope
r
a
te
s
by
id
e
nt
if
yi
ng
th
e
r
e
la
ti
ve
di
f
f
ic
ul
ty
le
ve
l
of
e
a
c
h
m
in
or
it
y
e
xa
m
pl
e
in
th
e
d
a
ta
s
e
t,
th
is
is
done
by
c
a
lc
ul
a
ti
ng
th
e
r
a
ti
o
b
e
twe
e
n
th
e
num
be
r
of
m
a
jo
r
it
y
ne
ig
hbor
s
a
nd
th
e
to
ta
l
num
be
r
of
ne
ig
hb
or
s
(
m
a
jo
r
it
y
a
nd
m
in
o
r
it
y)
f
or
e
a
c
h
m
in
or
it
y
e
xa
m
pl
e
.
M
in
or
it
y
e
xa
m
pl
e
s
th
a
t
ha
ve
lo
w
e
r
r
a
ti
os
a
r
e
c
ons
id
e
r
e
d
m
or
e
di
f
f
ic
ul
t
a
nd
m
o
r
e
im
por
ta
nt
to
e
xpa
nd.
A
D
A
S
Y
N
th
e
n
c
r
e
a
te
s
s
ynt
he
ti
c
s
a
m
pl
e
s
f
or
th
e
s
e
e
xa
m
pl
e
s
by
e
xt
e
ndi
ng
th
e
li
ne
be
twe
e
n
th
e
m
in
or
it
y
e
xa
m
pl
e
a
nd
it
s
ne
ig
hbor
s
in
f
e
a
tu
r
e
s
pa
c
e
,
f
oc
us
in
g
on
th
e
e
xa
m
pl
e
s
th
a
t
a
r
e
m
os
t
di
f
f
ic
ul
t
f
or
th
e
m
ode
l
to
id
e
nt
if
y.
T
hi
s
a
ppr
oa
c
h
e
ns
ur
e
s
th
a
t
th
e
r
e
s
ul
ti
ng
d
a
ta
s
e
t
h
a
s
a
be
tt
e
r
r
e
pr
e
s
e
nt
a
ti
on
of
m
in
or
it
y
c
la
s
s
e
s
,
im
pr
ovi
ng
m
ode
l
pe
r
f
or
m
a
nc
e
in
c
a
s
e
s
w
it
h
s
ig
ni
f
ic
a
nt
c
la
s
s
im
ba
la
nc
e
[
32]
.
A
n
ove
r
vi
e
w
of
th
e
e
nt
ir
e
da
ta
ba
la
nc
in
g pr
oc
e
s
s
w
it
h a
ll
t
he
m
e
th
ods
u
s
e
d c
a
n be
s
e
e
n i
n F
ig
ur
e
5.
F
ig
ur
e
5. D
a
ta
ba
la
nc
in
g a
nd non
-
ba
la
nc
in
g r
e
s
ul
t
s
2.4
.
C
la
s
s
if
i
c
at
io
n
w
it
h
X
G
B
oos
t
al
gor
it
h
m
, E
as
yE
n
s
e
m
b
le
c
la
s
s
i
f
ie
r
,
an
d
h
yp
e
r
p
ar
am
e
t
e
r
t
u
n
n
in
g
T
hi
s
s
tu
dy
e
m
pl
oys
th
r
e
e
m
ode
ll
in
g
m
e
th
ods
to
e
v
a
lu
a
te
a
lg
or
it
hm
pe
r
f
or
m
a
nc
e
unde
r
va
r
io
us
da
ta
c
ondi
ti
ons
,
in
c
lu
di
ng
ove
r
s
a
m
pl
e
d
da
ta
a
nd
or
ig
in
a
l
unba
la
n
c
e
d
da
ta
,
dur
in
g
bot
h
th
e
c
la
s
s
if
ic
a
ti
on
a
nd
hype
r
pa
r
a
m
e
te
r
tu
ni
ng
s
ta
ge
s
.
I
n
th
e
f
i
r
s
t
m
e
th
od,
th
e
X
G
B
oo
s
t
a
lg
or
it
hm
w
a
s
tr
a
in
e
d
on
t
r
a
in
in
g
da
ta
th
a
t
ha
d unde
r
gone
ove
r
s
a
m
pl
in
g w
it
h S
M
O
T
E
, A
D
A
S
Y
N
, a
nd
R
O
S
. S
in
c
e
t
he
c
la
s
s
di
s
tr
ib
ut
io
n i
n
t
hi
s
da
ta
w
a
s
ba
la
nc
e
d,
th
e
s
e
tt
in
gs
w
e
r
e
m
a
d
e
w
it
hout
th
e
s
c
a
l
e
_pos
_w
e
ig
ht
pa
r
a
m
e
te
r
.
I
n
th
e
s
e
c
ond
m
e
th
od,
th
e
c
ha
r
a
c
te
r
is
ti
c
s
of
c
la
s
s
im
ba
la
nc
e
w
e
r
e
m
a
in
ta
in
e
d
by
us
in
g
X
G
B
oos
t
on
th
e
or
ig
in
a
l
d
a
ta
w
it
hout
s
a
m
pl
in
g.
T
hi
s
m
e
th
od
a
ll
ow
s
th
e
s
c
a
le
_pos
_w
e
ig
ht
pa
r
a
m
e
te
r
to
be
i
nc
lu
de
d
in
th
e
hype
r
pa
r
a
m
e
te
r
s
e
a
r
c
h
s
pa
c
e
be
c
a
us
e
it
s
e
r
ve
s
to
im
pos
e
a
gr
e
a
te
r
pe
na
lt
y
on
pr
e
di
c
ti
on
e
r
r
or
s
in
m
in
or
it
y
c
la
s
s
e
s
.
T
hi
s
e
n
a
bl
e
s
th
e
m
ode
l
to
l
e
a
r
n f
r
om
i
m
ba
la
nc
e
d c
la
s
s
di
s
tr
ib
ut
io
ns
m
or
e
pr
opor
ti
ona
ll
y.
X
G
B
oos
t
c
om
bi
ne
s
boo
s
ti
ng
a
nd
gr
a
di
e
nt
boos
ti
ng
m
e
th
od
s
.
I
n
boos
ti
ng,
X
G
B
oos
t
is
us
e
d
to
c
la
s
s
if
y
e
r
r
or
s
f
r
om
pr
e
vi
ous
m
ode
ls
,
a
nd
it
i
s
us
e
of
gr
a
di
e
nt
de
s
c
e
nt
he
lp
s
m
in
im
iz
e
e
r
r
or
s
dur
in
g
th
e
c
r
e
a
ti
on
or
de
ve
lo
pm
e
nt
of
ne
w
m
ode
ls
[
33]
.
X
G
B
oos
t
r
e
qui
r
e
s
s
e
ve
r
a
l
pa
r
a
m
e
t
e
r
s
to
obt
a
in
a
n
opt
im
a
l
m
ode
l
c
a
ll
e
d
hype
r
pa
r
a
m
e
te
r
s
w
hi
c
h
a
r
e
us
e
d
to
a
dj
us
t
va
r
io
us
a
s
pe
c
t
s
of
m
a
c
hi
ne
le
a
r
ni
ng
s
o
th
a
t
th
e
y
c
a
n
in
f
lu
e
nc
e
th
e
pe
r
f
or
m
a
nc
e
of
th
e
m
e
th
od
in
pr
oc
e
s
s
in
g
da
ta
s
e
ts
,
s
e
v
e
r
a
l
pa
r
a
m
e
te
r
s
a
r
e
us
e
d
to
im
pr
ove
c
la
s
s
if
ic
a
ti
on us
in
g t
he
X
G
B
oos
t
m
e
th
od
[
34]
, c
a
n be
s
e
e
n i
n
T
a
bl
e
6
.
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
. 15, No. 1, Febr
ua
r
y 2026
:
655
-
671
662
T
a
bl
e
6
. P
a
r
a
m
e
te
r
s
i
n X
G
B
oos
t
m
e
th
od
P
a
r
a
m
e
t
e
r
I
nf
or
m
a
t
i
on
m
a
x_de
pt
h
M
a
xi
m
um
de
pt
h of
t
he
t
r
e
e
.
e
t
a
(
l
e
a
r
ni
ng_r
a
t
e
)
P
r
e
ve
nt
s
ove
r
f
i
t
t
i
ng by r
e
duc
i
ng s
i
z
e
m
i
n_c
hi
l
d_w
e
i
ght
M
i
ni
m
um
w
e
i
ght
of
c
hi
l
d_node
n_e
s
t
i
m
a
t
or
s
N
um
be
r
of
t
r
e
e
s
s
ubs
a
m
pl
e
R
a
ndom
l
y
s
a
m
pl
i
ng f
r
om
t
r
a
i
ni
ng da
t
a
be
f
or
e
c
ons
t
r
uc
t
i
ng t
he
t
r
e
e
.
r
a
ndom
_s
t
a
t
e
i
nt
e
r
na
l
r
a
ndom
num
be
r
ge
ne
r
a
t
or
i
ni
t
i
a
l
i
z
a
t
i
on
H
ype
r
pa
r
a
m
e
te
r
tu
ni
ng
w
a
s
pe
r
f
or
m
e
d
us
in
g
th
e
g
r
id
s
e
a
r
c
h
m
e
th
od,
w
hi
c
h
te
s
ts
a
ll
pa
r
a
m
e
te
r
c
om
bi
na
ti
ons
in
a
pr
e
de
te
r
m
in
e
d
s
e
a
r
c
h
s
pa
c
e
.
T
he
r
a
nge
o
f
va
lu
e
s
us
e
d
in
c
lu
de
d
m
a
x_d
e
pt
h
w
it
h
f
iv
e
va
r
ia
ti
ons
(
8,
10,
11,
13,
15
)
,
le
a
r
ni
ng_r
a
te
w
it
h
f
iv
e
va
lu
e
s
(
0.01,
0.02,
0.05,
0.07,
0.1)
,
m
in
_c
hi
ld
_w
e
ig
ht
w
it
h
two
va
lu
e
s
(
0.5
a
nd
1.0)
,
a
nd
n_e
s
ti
m
a
to
r
s
w
it
h
two
va
r
ia
ti
ons
(
150
a
nd
300)
.
T
he
s
ubs
a
m
pl
e
pa
r
a
m
e
te
r
is
lo
c
ke
d
a
t
a
v
a
lu
e
of
0.5
to
m
a
in
ta
in
c
ons
i
s
te
nc
y
in
th
e
pr
opor
ti
on
of
s
a
m
pl
e
s
us
e
d
in
e
a
c
h
tr
e
e
,
w
hi
le
r
a
ndom_s
ta
te
is
s
e
t
to
42
to
e
ns
ur
e
r
e
pr
oduc
ib
il
it
y.
W
it
h
th
is
c
onf
ig
ur
a
ti
on,
th
e
to
ta
l
num
be
r
of
hype
r
pa
r
a
m
e
te
r
c
om
bi
na
ti
ons
te
s
te
d
is
100
(
5×
5
×
2×
2
×
1)
.
E
a
c
h
c
om
bi
na
ti
on
is
e
va
lu
a
te
d
us
in
g
k
-
f
ol
d
c
r
os
s
-
va
li
da
ti
on
(
k=
5)
,
r
e
s
ul
ti
ng
in
a
to
ta
l
of
500
m
ode
l
tr
a
in
in
gs
.
T
hi
s
a
ppr
oa
c
h
e
ns
ur
e
s
th
a
t
hyp
e
r
pa
r
a
m
e
te
r
s
e
le
c
ti
on
i
s
s
ta
bl
e
,
c
on
s
is
te
nt
, a
nd
c
a
p
a
bl
e
of
r
e
pr
e
s
e
nt
in
g
th
e
m
ode
l'
s
pe
r
f
or
m
a
nc
e
in
a
ge
ne
r
a
li
z
a
bl
e
m
a
nne
r
on uns
e
e
n da
ta
.
T
he
t
hi
r
d m
e
th
od us
e
s
a
n E
a
s
yE
ns
e
m
bl
e
c
la
s
s
if
ie
r
on t
he
i
ni
ti
a
l
da
ta
t
o a
ddr
e
s
s
da
ta
i
m
ba
la
nc
e
. T
hi
s
m
e
th
od w
or
ks
by unde
r
s
a
m
pl
in
g t
he
m
a
jo
r
it
y
c
la
s
s
t
o f
or
m
s
e
v
e
r
a
l
ba
la
nc
e
d s
ubs
e
t
s
. T
he
n, i
n e
a
c
h s
ubgr
oup,
s
e
ve
r
a
l
w
e
a
k
m
ode
l
s
a
r
e
tr
a
in
e
d,
a
nd
a
pool
in
g
m
e
c
ha
ni
s
m
is
u
s
e
d
to
c
om
bi
ne
th
e
ir
pr
e
di
c
ti
ons
.
T
h
e
r
e
s
ul
t
is
a
m
ode
l
th
a
t
is
m
or
e
r
e
s
i
s
ta
nt
to
c
la
s
s
im
ba
la
nc
e
a
nd
m
or
e
s
ta
b
le
.
A
ddi
ti
ona
ll
y,
w
e
pe
r
f
or
m
pa
r
a
m
e
te
r
tu
ni
ng
on
th
e
m
ode
l
us
in
g
E
a
s
yE
ns
e
m
bl
e
.
T
h
e
pa
r
a
m
e
te
r
s
w
e
us
e
a
r
e
n_e
s
ti
m
a
to
r
s
,
ba
s
e
_
e
s
ti
m
a
to
r
(
a
DT
w
it
h
two
m
a
x_de
pt
h)
,
a
nd
s
ubs
ti
tu
ti
on;
th
e
be
s
t
v
a
lu
e
s
f
or
e
a
c
h
c
a
n
be
f
ound
by
te
s
ti
ng
th
e
m
e
th
od,
s
pe
c
if
ic
a
ll
y
th
r
ough a
gr
id
s
e
a
r
c
h vi
e
w
.
2.5
.
E
val
u
at
io
n
an
d
i
n
t
e
r
p
r
e
t
at
io
n
M
ode
l
pe
r
f
or
m
a
nc
e
e
va
lu
a
ti
on
is
c
onduc
te
d
c
om
pr
e
he
ns
iv
e
ly
by
c
om
bi
ni
ng
s
e
ve
r
a
l
ke
y
m
e
tr
ic
s
.
I
t
a
ls
o
in
c
lu
de
s
unc
e
r
ta
in
ty
a
n
a
ly
s
is
,
c
a
li
br
a
ti
on
m
e
a
s
ur
e
m
e
nt
s
,
m
ode
l
in
te
r
pr
e
ta
bi
li
ty
,
a
nd
e
xt
e
r
na
l
va
li
da
ti
on
to
e
ns
ur
e
g
e
ne
r
a
li
s
a
ti
on
c
a
pa
bi
li
ti
e
s
.
T
hi
s
e
va
lu
a
ti
on
a
ppr
oa
c
h
is
de
s
ig
ne
d
in
a
c
c
or
da
n
c
e
w
it
h
b
e
s
t
pr
a
c
ti
c
e
s
in
m
a
c
hi
ne
l
e
a
r
ni
ng
-
ba
s
e
d pr
e
di
c
ti
ve
m
ode
ll
in
g i
n t
he
he
a
lt
h do
m
a
in
a
nd f
or
ha
ndl
in
g i
m
ba
la
nc
e
d da
ta
.
2.5.1.
A
U
R
O
C
an
d
A
U
P
R
C
A
U
R
O
C
a
nd
A
U
P
R
C
is
us
e
d
to
a
s
s
e
s
s
th
e
di
s
c
r
im
in
a
to
r
y
pe
r
f
or
m
a
nc
e
of
a
m
ode
l.
A
U
R
O
C
a
s
s
e
s
s
e
s
th
e
m
ode
l
’
s
a
bi
li
ty
to
di
s
ti
ngui
s
h
be
twe
e
n
pos
it
iv
e
a
nd
ne
ga
ti
v
e
c
la
s
s
e
s
a
t
v
a
r
io
us
de
c
i
s
io
n
th
r
e
s
hol
ds
.
I
n
c
ont
r
a
s
t,
A
U
P
R
C
a
s
s
e
s
s
e
s
unba
la
nc
e
d
da
t
a
s
e
t
s
m
or
e
a
c
c
ur
a
te
ly
be
c
a
u
s
e
it
f
oc
us
e
s
on
th
e
r
e
la
ti
ons
hi
p
be
twe
e
n
pr
e
c
is
io
n
a
nd
r
e
c
a
ll
f
or
m
in
or
it
y
c
la
s
s
e
s
.
U
s
in
g
th
e
s
e
two
m
e
tr
ic
s
e
ns
ur
e
s
a
n
unbi
a
s
e
d
a
nd ba
la
nc
e
d e
v
a
lu
a
ti
on, pa
r
ti
c
ul
a
r
ly
w
he
n pr
e
di
c
ti
ng t
he
r
is
k of
r
a
r
e
e
ve
nt
s
[
35]
.
2.5.2.
C
on
f
id
e
n
c
e
in
t
e
r
val
s
T
o
il
lu
s
tr
a
te
th
e
s
ta
ti
s
ti
c
a
l
unc
e
r
ta
in
ty
of
th
e
e
va
lu
a
ti
on
r
e
s
ul
ts
,
e
a
c
h
pe
r
f
or
m
a
nc
e
m
e
tr
ic
is
a
c
c
om
pa
ni
e
d
by
a
c
onf
id
e
nc
e
in
te
r
va
l.
C
onf
id
e
n
c
e
in
te
r
va
ls
a
r
e
c
a
lc
ul
a
te
d
th
r
ough
r
e
pe
a
te
d
boot
s
tr
a
ppi
ng
on
th
e
te
s
t
da
ta
,
a
nd
th
e
e
s
ti
m
a
te
s
obt
a
in
e
d
r
e
f
le
c
t
th
e
va
r
ia
b
il
it
y
of
m
ode
l
pe
r
f
o
r
m
a
nc
e
a
c
r
os
s
di
f
f
e
r
e
nt
s
a
m
pl
e
s
[
36]
.
C
om
bi
ni
ng
C
I
s
im
pr
ove
s
th
e
r
e
li
a
bi
li
ty
of
in
te
r
pr
e
ta
ti
on
a
nd
a
ll
ow
s
f
or
be
tt
e
r
c
om
pa
r
is
ons
be
twe
e
n m
ode
ls
.
2.5.3.
C
al
ib
r
at
io
n
p
lo
t
s
T
o
a
s
s
e
s
s
c
a
li
br
a
ti
on,
c
a
li
br
a
ti
on
pl
ot
s
a
nd
a
ddi
ti
ona
l
c
a
li
br
a
ti
on
s
c
or
e
s
,
s
u
c
h
a
s
th
e
B
r
ie
r
s
c
or
e
,
a
r
e
us
e
d.
C
a
li
br
a
ti
on
de
te
r
m
in
e
s
th
e
le
ve
l
of
li
ke
li
hood
of
th
e
m
ode
l'
s
pr
e
di
c
ti
ons
c
om
pa
r
e
d
to
th
e
a
c
tu
a
l
pr
oba
bi
li
ty
of
e
ve
nt
s
.
C
a
li
br
a
ti
on
pl
ot
s
a
r
e
us
e
d
to
a
s
s
e
s
s
w
h
e
th
e
r
th
e
m
ode
l
te
nds
to
be
ove
r
c
onf
id
e
nt
in
it
s
pr
e
di
c
ti
ons
.
C
li
ni
c
a
l
a
ppl
ic
a
ti
ons
a
nd
d
e
c
is
io
n
s
uppor
t
s
ys
t
e
m
s
r
e
qui
r
e
good
m
ode
ls
be
c
a
u
s
e
th
e
y
c
a
n
di
s
ti
ngui
s
h c
la
s
s
e
s
a
nd ge
ne
r
a
te
w
e
ll
-
c
a
li
br
a
te
d pr
oba
bi
li
ti
e
s
[
3
7]
.
2.5.4.
E
xp
la
in
ab
il
it
y
U
s
in
g
th
e
e
xpl
a
in
a
bl
e
m
e
th
od,
S
H
A
P
,
th
e
a
s
pe
c
t
of
in
te
r
pr
e
ta
bi
li
ty
w
a
s
e
x
a
m
in
e
d.
T
hi
s
s
tu
dy
pr
ovi
de
s
a
n
unde
r
s
ta
ndi
ng
of
th
e
c
ont
r
ib
ut
io
n
of
e
a
c
h
f
e
a
tu
r
e
to
m
ode
l
pr
e
di
c
ti
ons
a
t
bot
h
th
e
gl
oba
l
le
ve
l
(
a
c
r
os
s
th
e
e
nt
ir
e
da
ta
s
e
t)
a
nd
th
e
lo
c
a
l
le
ve
l
(
f
or
in
di
vi
dua
l
pr
e
di
c
ti
ons
)
.
T
h
e
e
x
pl
a
in
a
bi
li
ty
a
ppr
oa
c
h
m
a
ke
s
th
e
m
o
de
l
c
l
e
a
r
a
nd
f
a
c
i
li
t
a
t
e
s
s
t
a
k
e
h
ol
d
e
r
s
,
e
s
pe
c
i
a
ll
y t
ho
s
e
w
or
ki
ng
i
n t
he
m
e
di
c
a
l
or
pub
li
c
p
ol
i
c
y
f
i
e
l
ds
[
3
8]
.
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
St
r
ok
e
pr
e
di
c
ti
on us
in
g data bal
anc
in
g m
e
th
od and e
x
t
r
e
m
e
g
r
adi
e
nt
boos
ti
ng
(
A
bd M
iz
w
ar
A
. R
ahi
m
)
663
2.5.5.
D
e
c
is
io
n
an
al
ys
is
I
n
a
ddi
ti
on
to
s
ta
nda
r
d
m
e
tr
ic
s
,
a
de
c
is
io
n
a
na
ly
s
i
s
w
a
s
pe
r
f
or
m
e
d
to
e
va
lu
a
te
th
e
m
ode
l'
s
va
lu
e
in
r
e
a
l
-
w
or
ld
de
c
is
io
n
-
m
a
ki
ng.
D
e
c
is
io
n
c
ur
ve
a
na
ly
s
is
(
D
C
A
)
,
w
hi
c
h
a
s
s
e
s
s
e
s
th
e
ne
t
be
ne
f
it
of
th
e
m
ode
l
a
t
va
r
io
us
r
is
k
th
r
e
s
hol
ds
,
is
in
c
lu
de
d
in
th
is
a
na
ly
s
i
s
.
D
C
A
s
e
r
ve
s
to
e
va
lu
a
te
w
he
th
e
r
th
e
m
ode
l
tr
ul
y
ha
s
c
li
ni
c
a
l
or
ope
r
a
ti
ona
l
a
dva
nt
a
ge
s
ove
r
ba
s
ic
m
e
th
ods
s
uc
h a
s
t
r
e
a
t
-
a
ll
or
t
r
e
a
t
-
none
.
T
he
r
e
f
or
e
, t
he
e
va
lu
a
ti
on
not
onl
y c
ons
id
e
r
s
s
ta
ti
s
ti
c
a
l
pe
r
f
or
m
a
nc
e
but
a
ls
o t
he
va
lu
e
of
t
he
m
ode
l
in
r
e
a
l
-
li
f
e
s
it
ua
ti
ons
[
39]
.
2.5.6.
E
xt
e
r
n
al
val
id
at
io
n
T
he
tr
a
in
e
d
a
nd
e
va
lu
a
te
d
m
ode
l
i
s
th
e
n
te
s
te
d
th
r
ough
e
xt
e
r
n
a
l
va
li
da
ti
on
w
it
h
da
ta
f
r
om
va
r
io
us
s
our
c
e
s
or
ti
m
e
pe
r
io
ds
.
E
xt
e
r
na
l
va
li
da
ti
on
s
how
s
th
e
m
ode
l
’
s
pe
r
f
or
m
a
nc
e
in
s
it
ua
ti
ons
out
s
id
e
th
e
in
it
ia
l
tr
a
in
in
g
da
ta
di
s
tr
ib
ut
io
n.
T
hi
s
s
te
p
i
s
c
r
uc
ia
l
f
or
a
s
s
e
s
s
in
g
th
e
ge
ne
r
a
li
s
a
ti
on
a
nd
s
tr
e
ngt
h
of
th
e
m
ode
l
a
nd
i
s
e
s
s
e
nt
ia
l
in
pr
e
di
c
ti
ve
r
e
s
e
a
r
c
h a
im
e
d
a
t
w
id
e
r
a
ppl
ic
a
ti
on
[
40]
.
3.
R
E
S
U
L
T
S
A
N
D
D
I
S
C
U
S
S
I
O
N
T
hi
s
s
e
c
ti
on
pr
e
s
e
nt
s
th
e
r
e
s
ul
ts
of
th
e
m
ode
ll
in
g
a
nd
e
va
lu
a
ti
on
pr
oc
e
s
s
of
th
is
s
tu
dy.
T
o
f
ul
f
il
th
e
r
e
s
e
a
r
c
h
obj
e
c
ti
ve
s
,
a
c
om
pr
e
he
ns
iv
e
a
na
ly
s
is
w
a
s
c
onduc
te
d
to
e
va
lu
a
te
th
e
pe
r
f
or
m
a
nc
e
o
f
th
e
X
G
B
oos
t
m
ode
l
unde
r
th
r
e
e
di
f
f
e
r
e
nt
c
ondi
ti
ons
:
ove
r
s
a
m
pl
in
g
da
ta
,
or
ig
in
a
l
da
ta
w
it
h
w
e
ig
ht
a
dj
us
tm
e
nt
us
in
g
s
c
a
le
_pos
_
w
e
ig
ht
,
a
nd
or
ig
in
a
l
da
ta
pr
oc
e
s
s
e
d
u
s
in
g
th
e
E
a
s
y
E
ns
e
m
bl
e
c
la
s
s
if
ie
r
.
T
o
e
va
lu
a
te
th
e
pr
a
c
ti
c
a
l
be
ne
f
it
s
of
th
e
m
ode
l
in
de
c
is
io
n
m
a
ki
ng,
de
c
is
io
n
a
na
ly
s
is
in
c
lu
de
d
pe
r
f
or
m
a
nc
e
e
va
lu
a
ti
on
us
in
g
A
U
R
O
C
a
nd
A
U
P
R
C
m
e
tr
ic
s
,
pr
e
di
c
ti
on
c
a
li
br
a
ti
on,
unc
e
r
ta
in
ty
l
e
ve
ls
th
r
ough
c
onf
id
e
nc
e
in
te
r
va
ls
,
m
ode
l
in
te
r
pr
e
ta
bi
li
ty
us
in
g
S
H
A
P
,
a
nd
de
c
is
io
n
a
na
ly
s
is
.
T
o
e
ns
ur
e
th
e
m
ode
l
’
s
ge
ne
r
a
li
s
a
ti
on
a
bi
li
ty
to
da
ta
out
s
id
e
th
e
tr
a
in
in
g
di
s
tr
ib
ut
io
n,
e
xt
e
r
na
l
v
a
li
da
ti
on
w
a
s
a
l
s
o
pe
r
f
or
m
e
d.
T
he
f
ol
lo
w
in
g
a
r
e
th
e
te
s
t
r
e
s
ul
t
s
f
or
th
e
or
ig
in
a
l
da
ta
w
it
h w
e
ig
ht
a
dj
us
tm
e
nt
us
in
g s
c
a
le
_pos
_
w
e
ig
ht
, a
s
s
how
n i
n F
ig
ur
e
6.
F
ig
ur
e
6. E
va
lu
a
ti
on
r
e
s
ul
ts
of
A
U
R
O
C
a
nd
A
U
P
R
C
,
c
onf
id
e
nc
e
i
nt
e
r
va
ls
, c
a
li
br
a
ti
on plot
s
, e
xpl
a
in
a
bi
li
ty
,
de
c
is
io
n a
na
ly
s
is
, a
nd e
xt
e
r
na
l
v
a
li
da
ti
on
in
t
e
s
ti
ng or
ig
in
a
l
da
ta
w
it
h t
he
X
G
B
oos
t
m
e
th
od us
in
g t
he
s
c
a
le
_pos
_
w
e
ig
ht
pa
r
a
m
e
te
r
F
ig
ur
e
6
s
how
s
th
a
t
th
e
X
G
B
oos
t
m
ode
l
c
a
li
br
a
te
d
us
in
g
s
c
a
le
_pos
_w
e
ig
ht
e
xhi
bi
ts
s
ol
id
pr
e
di
c
ti
ve
pe
r
f
or
m
a
nc
e
on
th
e
te
s
t
da
ta
,
w
it
h
th
e
b
e
s
t
c
om
bi
n
a
ti
on
o
f
hype
r
pa
r
a
m
e
te
r
s
be
in
g
le
a
r
ni
ng_r
a
te
=
0.01,
m
a
x_de
pt
h=
8,
m
in
_c
hi
ld
_w
e
ig
ht
=
1.0,
a
nd
n_e
s
ti
m
a
to
r
s
=
150.
T
he
m
od
e
l
a
c
hi
e
v
e
d
a
n
A
U
R
O
C
va
lu
e
of
0.836,
in
di
c
a
ti
ng
a
s
tr
ong
a
bi
li
ty
to
di
s
ti
ngui
s
h
be
twe
e
n
s
tr
oke
a
nd
non
-
s
tr
oke
c
la
s
s
e
s
.
T
he
A
U
P
R
C
v
a
lu
e
of
0.314
r
e
m
a
in
e
d
a
bove
th
e
b
a
s
e
li
ne
in
c
ondi
ti
ons
of
hi
gh
c
la
s
s
i
m
ba
la
nc
e
.
T
h
e
c
a
li
br
a
ti
on
gr
a
ph
in
di
c
a
te
s
th
a
t
th
e
m
ode
l
te
nds
to
e
s
ti
m
a
te
pr
oba
bi
li
ti
e
s
c
on
s
e
r
va
ti
ve
ly
in
th
e
lo
w
r
a
nge
but
is
m
or
e
a
c
c
ur
a
te
a
t
hi
gh
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
. 15, No. 1, Febr
ua
r
y 2026
:
655
-
671
664
pr
oba
bi
li
ti
e
s
.
S
H
A
P
a
na
ly
s
is
c
onf
ir
m
e
d
th
a
t
a
ge
,
B
M
I
,
a
nd
gl
u
c
os
e
le
ve
ls
a
r
e
th
e
m
os
t
in
f
lu
e
nt
ia
l
pr
e
di
c
to
r
s
,
in
li
ne
w
it
h
c
li
ni
c
a
l
li
te
r
a
tu
r
e
.
I
n
a
ddi
ti
on,
D
C
A
s
how
s
th
a
t
th
e
m
ode
l
pr
ovi
de
s
be
tt
e
r
de
c
is
io
n
be
ne
f
it
s
th
a
n
tr
e
a
t
-
a
ll
or
t
r
e
a
t
-
none
s
tr
a
te
gi
e
s
, pr
ovi
ng t
ha
t
th
is
c
la
s
s
-
w
e
ig
ht
in
g a
ppr
oa
c
h i
s
pr
a
c
ti
c
a
l
a
nd r
e
le
va
nt
f
or
s
tr
oke
r
is
k
pr
e
di
c
ti
on
s
c
e
na
r
io
s
.
F
ur
th
e
r
m
or
e
,
th
e
r
e
s
ul
ts
of
te
s
ti
ng
on
th
e
or
ig
in
a
l
da
ta
us
in
g
th
e
E
a
s
yE
ns
e
m
bl
e
c
la
s
s
if
i
e
r
m
e
th
od a
r
e
s
how
n i
n F
ig
ur
e
7.
F
ig
ur
e
7. E
va
lu
a
ti
on
r
e
s
ul
ts
of
A
U
R
O
C
a
nd
A
U
P
R
C
,
c
onf
id
e
nc
e
i
nt
e
r
va
ls
, c
a
li
br
a
ti
on plot
s
, e
xpl
a
in
a
bi
li
ty
,
de
c
is
io
n a
na
ly
s
is
, a
nd e
xt
e
r
na
l
v
a
li
da
ti
on i
n t
e
s
ti
ng or
ig
in
a
l
da
ta
us
in
g
th
e
E
a
s
yE
ns
e
m
bl
e
C
la
s
s
if
ie
r
m
e
th
od
F
ig
ur
e
7
s
how
s
th
e
be
s
t
pe
r
f
or
m
a
nc
e
r
e
s
ul
ts
f
or
th
e
E
a
s
yE
ns
e
m
bl
e
c
la
s
s
if
i
e
r
m
e
th
od
us
in
g
th
e
or
ig
in
a
l
da
ta
w
it
h t
he
DT
c
onf
ig
ur
a
ti
on de
pt
h=
6, n_e
s
ti
m
a
to
r
s
=
30, a
nd r
e
pl
a
c
e
m
e
nt
=
fa
ls
e
. I
n t
he
e
xt
e
r
na
l
te
s
t
s
e
t,
th
e
m
ode
l
a
c
hi
e
ve
d
a
n
A
U
R
O
C
of
0.839,
in
di
c
a
ti
ng
s
tr
ong
di
s
c
r
im
in
a
to
r
y
pow
e
r
;
how
e
ve
r
,
th
e
A
U
P
R
C
va
lu
e
of
0.239
s
ugge
s
t
s
c
h
a
ll
e
nge
s
in
pr
e
di
c
ti
ng
m
in
or
it
y
c
l
a
s
s
e
s
.
T
he
c
a
li
br
a
ti
on
c
ur
ve
s
how
s
th
a
t
th
e
pr
e
di
c
ti
on
pr
oba
bi
li
ti
e
s
a
r
e
not
f
ul
ly
a
li
gne
d
w
it
h
th
e
a
c
tu
a
l
pr
oba
bi
li
ti
e
s
,
e
s
p
e
c
ia
ll
y
in
th
e
m
id
dl
e
pr
oba
bi
li
ty
r
a
nge
.
F
e
a
tu
r
e
im
por
ta
nc
e
a
na
ly
s
is
id
e
nt
if
ie
s
a
ge
a
s
th
e
m
os
t
in
f
lu
e
nt
ia
l
pr
e
di
c
to
r
,
f
ol
lo
w
e
d
by
B
M
I
a
nd
a
ve
r
a
ge
gl
uc
os
e
le
ve
l,
w
hi
le
ot
he
r
f
e
a
tu
r
e
s
c
ont
r
ib
ut
e
m
in
im
a
ll
y.
M
e
a
nw
hi
le
,
D
C
A
s
how
s
th
a
t
th
e
m
ode
l
pr
ovi
de
s
s
m
a
ll
but
s
ti
ll
pos
it
iv
e
de
c
is
io
n
be
n
e
f
it
s
a
t
m
o
s
t
th
r
e
s
h
ol
ds
.
O
ve
r
a
ll
,
th
is
m
od
e
l
is
s
ui
ta
bl
e
f
or
us
e
a
s
a
s
ol
id
ba
s
e
li
ne
,
but
im
pr
ove
m
e
nt
s
in
a
c
c
ur
a
c
y
f
or
m
in
o
r
it
y
c
l
a
s
s
e
s
a
r
e
s
ti
ll
ne
e
de
d.
N
e
xt
a
r
e
th
e
r
e
s
ul
ts
of
te
s
ti
ng
on
s
a
m
pl
e
d
da
ta
us
in
g
th
e
A
D
A
S
Y
N
te
c
hni
que
,
w
h
ic
h
w
e
r
e
th
e
n
c
la
s
s
if
ie
d
us
in
g
th
e
X
G
B
oos
t
m
e
th
od, a
s
s
how
n i
n F
ig
ur
e
8.
F
ig
ur
e
8
s
how
s
th
e
r
e
s
ul
ts
of
te
s
ti
ng
th
e
s
a
m
pl
in
g
da
t
a
(
a
da
s
yn)
us
in
g
th
e
X
G
B
oos
t
m
e
th
od,
in
di
c
a
ti
ng
th
a
t
th
e
be
s
t
c
onf
ig
ur
a
ti
on,
w
it
h
m
a
x_de
pt
h
=
15,
le
a
r
ni
ng_r
a
te
=
0.05,
m
in
_c
hi
ld
_w
e
ig
ht
=
1.0,
a
nd
n_e
s
ti
m
a
to
r
s
=
300,
e
xhi
bi
ts
f
a
ir
ly
good
c
la
s
s
if
ic
a
ti
on
c
a
pa
bi
li
ti
e
s
.
T
he
A
U
R
O
C
va
lu
e
of
0.768
in
di
c
a
te
s
th
a
t
th
e
m
ode
l
c
a
n di
s
ti
ngui
s
h be
twe
e
n n
e
ga
ti
ve
a
nd pos
it
iv
e
c
l
a
s
s
e
s
m
ode
r
a
te
ly
. I
n c
ont
r
a
s
t,
t
he
A
U
P
R
C
va
lu
e
of
0.107
s
ugge
s
t
s
th
a
t
p
e
r
f
or
m
a
nc
e
on
m
in
or
it
y
c
la
s
s
e
s
r
e
m
a
i
ns
li
m
it
e
d,
w
hi
c
h
m
a
y
be
a
tt
r
ib
ut
e
d
to
d
a
ta
im
ba
la
nc
e
.
A
lt
hough
th
e
pr
oba
bi
li
ty
p
r
e
di
c
ti
ons
a
r
e
not
ye
t
f
ul
ly
a
li
gne
d
w
it
h
th
e
a
c
tu
a
l
di
s
tr
ib
u
ti
on,
th
e
c
a
li
br
a
ti
on
c
ur
ve
gi
ve
s
a
B
r
ie
r
s
c
or
e
of
0.0921.
T
hi
s
in
di
c
a
te
s
th
a
t
th
e
pr
oba
bi
li
ty
pr
e
di
c
ti
ons
a
r
e
r
e
la
ti
ve
l
y
a
c
c
ur
a
te
.
A
c
c
or
di
ng
to
S
H
A
P
a
na
ly
s
i
s
,
th
e
va
r
ia
bl
e
s
of
a
ge
,
B
M
I
,
a
nd
a
v
e
r
a
ge
gl
uc
o
s
e
le
ve
l
a
r
e
th
os
e
th
a
t
m
os
t
in
f
lu
e
nc
e
th
e
m
od
e
l'
s
pr
e
di
c
ti
ons
.
T
he
s
e
f
in
di
ngs
a
r
e
c
on
s
is
te
nt
w
it
h
pr
e
vi
ous
s
tu
di
e
s
.
H
ow
e
v
e
r
,
D
C
A
s
how
s
th
a
t
th
e
m
ode
l
doe
s
not
pr
ov
id
e
a
gr
e
a
te
r
ne
t
be
ne
f
it
c
om
pa
r
e
d
to
th
e
“
s
e
r
ve
a
ll
”
a
nd
“
s
e
r
ve
none
”
s
tr
a
te
gi
e
s
.
A
s
a
r
e
s
ul
t,
th
e
m
ode
l
c
a
nnot
be
u
s
e
d
f
or
th
r
e
s
hol
d
-
ba
s
e
d
de
c
is
io
n
m
a
ki
ng.
T
he
f
ol
lo
w
in
g
a
r
e
th
e
r
e
s
ul
ts
of
c
la
s
s
if
ic
a
ti
on
u
s
in
g
th
e
X
G
B
oos
t
m
e
th
od
in
th
e
s
e
c
ond
da
ta
s
a
m
pl
in
g
c
ondi
ti
on
w
it
h
th
e
R
O
S
te
c
hni
que
, a
s
s
how
n i
n F
ig
ur
e
9.
Evaluation Warning : The document was created with Spire.PDF for Python.