I
AE
S
I
n
t
e
r
n
at
ion
al
Jou
r
n
al
of
Ar
t
if
icial
I
n
t
e
ll
ig
e
n
c
e
(
I
J
-
AI
)
Vol.
14
,
No.
4
,
Augus
t
2025
,
pp.
2849
~
2863
I
S
S
N:
2252
-
8938
,
DO
I
:
10
.
11591/i
jai
.
v
14
.i
4
.
pp
28
49
-
2863
2849
Jou
r
n
al
h
omepage
:
ht
tp:
//
ij
ai
.
iaes
c
or
e
.
c
om
Pe
r
f
o
r
m
a
n
c
e
an
al
ysi
s an
d
c
o
m
p
ar
is
o
n
of
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
al
gor
ith
m
s f
o
r
p
r
e
d
ic
t
in
g h
e
a
r
t
d
is
e
as
e
Ne
h
a
B
h
ad
u
,
Jas
win
d
e
r
S
in
gh
D
e
pa
r
tm
e
nt
of
C
omput
e
r
S
c
ie
nc
e
a
nd E
ngi
ne
e
r
in
g, F
a
c
ul
ty
of
E
ngi
ne
e
r
in
g a
nd T
e
c
hnol
ogy, G
ur
u
J
a
mbhe
s
hw
a
r
U
ni
ve
r
s
it
y of
S
c
ie
nc
e
a
nd T
e
c
hnol
ogy, Hi
s
a
r
, I
ndi
a
Ar
t
icle
I
n
f
o
AB
S
T
RA
CT
A
r
ti
c
le
h
is
tor
y
:
R
e
c
e
ived
Apr
19,
2024
R
e
vis
e
d
M
a
r
21,
2025
Ac
c
e
pted
J
un
8,
2025
H
eart
d
i
s
eas
e
(
HD
)
i
s
a
s
eri
o
u
s
med
i
cal
co
n
d
i
t
i
o
n
t
h
at
h
as
an
en
o
rm
o
u
s
effect
o
n
p
eo
p
l
e'
s
q
u
al
i
t
y
o
f
l
i
fe.
E
ar
l
y
a
s
w
el
l
as
acc
u
rat
e
i
d
e
n
t
i
fi
cat
i
o
n
i
s
cru
ci
a
l
fo
r
p
rev
e
n
t
i
n
g
an
d
t
reat
i
n
g
H
D
.
T
ra
d
i
t
i
o
n
a
l
met
h
o
d
s
o
f
d
i
a
g
n
o
s
i
s
may
n
o
t
al
w
ay
s
b
e
re
l
i
a
b
l
e.
N
o
n
-
i
n
t
r
u
s
i
v
e
me
t
h
o
d
s
l
i
k
e
mach
i
n
e
l
earn
i
n
g
(ML
)
are
p
ro
fi
c
i
en
t
i
n
d
i
s
t
i
n
g
u
i
s
h
i
n
g
b
e
t
w
ee
n
p
a
t
i
en
t
s
w
i
t
h
H
D
a
n
d
t
h
o
s
e
i
n
g
o
o
d
h
ea
l
t
h
.
T
h
e
p
ri
me
o
b
j
ect
i
v
e
o
f
t
h
i
s
s
t
u
d
y
i
s
t
o
fi
n
d
a
ro
b
u
s
t
ML
t
ech
n
i
q
u
e
t
h
a
t
can
accu
ra
t
el
y
d
e
t
ect
t
h
e
p
re
s
en
ce
o
f
H
D
.
Fo
r
t
h
i
s
p
u
r
p
o
s
e,
s
ev
era
l
ML
al
g
o
ri
t
h
m
s
w
ere
ch
o
s
e
n
b
a
s
ed
o
n
t
h
e
rel
e
v
a
n
t
l
i
t
erat
u
re
s
t
u
d
i
e
d
.
Fo
r
t
h
i
s
i
n
v
e
s
t
i
g
a
t
i
o
n
,
t
w
o
d
i
fferen
t
h
ear
t
d
at
a
s
et
s
t
h
e
C
l
ev
e
l
an
d
an
d
St
a
t
l
o
g
d
at
a
s
et
s
w
ere
d
o
w
n
l
o
ad
e
d
fro
m
K
ag
g
l
e.
T
h
e
an
a
l
y
s
i
s
w
as
carr
i
ed
o
u
t
u
t
i
l
i
zi
n
g
t
h
e
W
ai
k
at
o
en
v
i
ro
n
me
n
t
fo
r
k
n
o
w
l
e
d
g
e
an
a
l
y
s
i
s
(
W
E
K
A
)
3
.
9
.
6
s
o
ft
w
are.
T
o
a
s
s
e
s
s
h
o
w
w
el
l
v
ari
o
u
s
al
g
o
r
i
t
h
ms
p
re
d
i
ct
e
d
H
D
,
t
h
e
s
t
u
d
y
emp
l
o
y
e
d
a
v
ari
et
y
o
f
p
erf
o
rman
ce
ev
a
l
u
a
t
i
o
n
met
r
i
cs
an
d
erro
r
rat
e
s
.
T
h
e
fi
n
d
i
n
g
s
s
h
o
w
e
d
t
h
at
f
o
r
b
o
t
h
t
h
e
d
at
a
s
et
s
r
an
d
o
m
f
o
res
t
(RF)
i
s
a
b
et
t
er
o
p
t
i
o
n
fo
r
p
red
i
ct
i
n
g
H
D
w
i
t
h
an
accu
racy
an
d
recei
v
er
o
p
erat
i
n
g
ch
aract
er
i
s
t
i
c
(
RO
C
)
v
al
u
es
o
f
9
4
%
a
n
d
0
.
9
8
4
fo
r
t
h
e
Cl
ev
e
l
an
d
d
at
a
s
et
a
n
d
9
0
%
an
d
0
.
9
7
5
f
o
r
t
h
e
St
at
l
o
g
d
a
t
as
e
t
.
T
h
i
s
w
o
r
k
ma
y
ai
d
res
earc
h
ers
i
n
creat
i
n
g
earl
y
H
D
d
et
e
ct
i
o
n
mo
d
el
s
an
d
as
s
i
s
t
med
i
cal
p
rac
t
i
t
i
o
n
er
s
i
n
i
d
e
n
t
i
fy
i
n
g
H
D
.
K
e
y
w
o
r
d
s
:
De
c
is
ion
tr
e
e
He
a
r
t
dis
e
a
s
e
M
a
c
hine
lea
r
ning
P
e
r
f
or
manc
e
metr
ics
R
a
ndom
f
or
e
s
t
W
E
KA
Th
i
s
i
s
a
n
o
p
en
a
c
ces
s
a
r
t
i
c
l
e
u
n
d
e
r
t
h
e
CC
B
Y
-
SA
l
i
ce
n
s
e.
C
or
r
e
s
pon
din
g
A
u
th
or
:
Ne
ha
B
ha
du
De
pa
r
tm
e
nt
of
C
omput
e
r
S
c
ienc
e
a
nd
E
nginee
r
ing
,
F
a
c
ult
y
of
E
nginee
r
ing
a
nd
T
e
c
hnology
Gur
u
J
a
mbhes
hwa
r
Unive
r
s
it
y
of
S
c
ienc
e
a
nd
T
e
c
hnology
His
a
r
-
125001,
Ha
r
ya
na
,
I
ndia
E
mail:
ne
ha
bha
du100@gmail.
c
om
1.
I
NT
RODU
C
T
I
ON
P
e
ople
nowa
da
ys
a
r
e
f
a
c
ing
major
he
a
lt
h
c
ha
ll
e
nge
s
.
Util
iza
ti
on
of
tobac
c
o,
unhe
a
lt
hy
dieta
r
y
pa
tt
e
r
ns
,
a
nd
ins
uf
f
icie
nt
phys
ica
l
a
c
ti
vit
y
a
r
e
lea
ding
to
numer
ous
c
hr
onic
il
lnes
s
e
s
.
C
hr
onic
il
lnes
s
e
s
a
r
e
the
main
r
e
a
s
ons
f
or
de
a
th
a
nd
dis
a
bil
it
y
wor
ldwide
.
As
pe
r
the
US
Na
ti
ona
l
C
e
ntr
e
f
o
r
He
a
lt
h
S
tatis
ti
c
s
,
c
hr
onic
dis
e
a
s
e
s
pe
r
s
is
t
f
or
a
n
e
xtende
d
dur
a
ti
on
,
typ
ic
a
ll
y
e
xc
e
e
ding
thr
e
e
mont
hs
.
T
he
s
e
dis
e
a
s
e
s
a
r
e
ne
it
he
r
c
ur
a
ble
thr
ough
medic
a
ti
on
no
r
pr
e
ve
ntable
th
r
ou
gh
va
c
c
ination.
He
a
lt
h
c
ondit
ions
li
ke
he
a
r
t
dis
e
a
s
e
(
HD
)
,
c
a
nc
e
r
,
a
r
thr
i
ti
s
,
d
iabe
tes
,
obe
s
it
y,
de
pr
e
s
s
ion,
a
nd
other
s
f
a
ll
unde
r
thi
s
c
a
tegor
y
of
di
s
e
a
s
e
s
[
1]
.
On
e
of
the
de
a
dli
e
s
t
c
hr
onic
il
lnes
s
e
s
,
HD
,
will
be
the
s
ubjec
t
of
thi
s
inves
ti
ga
ti
on.
T
he
human
he
a
r
t
is
in
c
ha
r
ge
of
pumpi
ng
blood,
s
upplyi
ng
a
ll
body
o
r
ga
ns
with
n
utr
it
ion
a
nd
oxyge
n,
a
nd
r
e
movi
ng
ha
r
mf
ul
e
lem
e
nts
li
ke
c
a
r
bon
dioxi
de
.
S
e
ve
r
a
l
c
ondit
ions
that
a
f
f
e
c
t
the
s
tr
uc
tur
e
a
nd
f
unc
ti
on
of
the
he
a
r
t
a
r
e
c
oll
e
c
ti
ve
ly
r
e
f
e
r
r
e
d
to
a
s
HD
.
HD
is
c
las
s
if
ied
a
s
c
a
r
dio
va
s
c
ular
dis
e
a
s
e
(
C
VD
)
.
C
VD
e
nc
ompas
s
e
s
a
r
a
nge
of
he
a
r
t
a
nd
blood
ve
s
s
e
l
c
ondit
ions
,
s
uc
h
a
s
pe
r
ipher
a
l
a
r
ter
ial
dis
e
a
s
e
,
he
a
r
t
a
tt
a
c
ks
,
s
tr
oke
s
,
a
nd
c
or
ona
r
y
HD
.
I
t
is
e
s
s
e
nti
a
l
to
unde
r
s
tand
that
while
a
ll
HD
s
a
r
e
C
VD
s
,
not
a
ll
C
VD
s
a
r
e
c
la
s
s
if
ied
a
s
H
D
s
[
2]
.
S
e
ve
r
a
l
f
a
c
tor
s
,
i
nc
lu
ding
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14
,
No.
4
,
Augus
t
2025
:
284
9
-
2863
2850
a
ge
,
s
e
x,
tobac
c
o
us
e
,
ha
ving
a
f
a
mi
ly
his
tor
y
of
HD
,
high
blood
pr
e
s
s
ur
e
,
h
igh
c
holes
ter
ol,
e
a
ti
ng
a
n
unhe
a
lt
hy
diet,
hype
r
tens
ion,
be
ing
ove
r
we
ight
,
i
na
c
ti
vit
y,
a
nd
a
lcohol
c
ons
umpt
ion,
c
a
n
r
a
is
e
one
's
c
ha
nc
e
of
de
ve
lopi
ng
HD
[
3]
.
T
he
r
e
e
xis
t
va
r
ious
f
or
ms
of
HD
,
s
uc
h
a
s
“
c
or
ona
r
y
HD
,
a
ngina
pe
c
tor
is
,
c
o
nge
s
ti
ve
he
a
r
t
f
a
il
u
r
e
,
c
a
r
diom
yopa
thy,
c
onge
nit
a
l
HD
,
a
r
r
hythm
ias
,
a
nd
myoca
r
dit
is
”
[
4]
.
T
he
mos
t
p
r
e
va
lent
c
ondit
ion
a
mong
thes
e
is
c
or
ona
r
y
HD
.
As
a
r
e
s
ult
of
thi
s
c
ondit
ion
,
the
c
or
ona
r
y
a
r
ter
ies
,
whic
h
f
e
e
d
the
he
a
r
t
with
blood
r
ich
in
oxyge
n
,
s
hr
ink
o
r
b
lock.
C
omm
on
s
igns
of
HD
include
c
he
s
t
dis
c
omf
or
t
,
d
if
f
iculty
br
e
a
thi
ng,
li
g
ht
-
he
a
de
dne
s
s
,
na
us
e
a
,
puf
f
y
f
e
e
t,
e
xt
r
e
me
s
we
a
ti
ng,
a
nd
ge
ne
r
a
l
f
a
ti
gue
.
T
im
e
ly
identif
ica
ti
on
of
HD
c
a
n
he
lp
r
e
duc
e
the
mor
talit
y
r
a
te
a
nd
mi
ni
mi
z
e
ove
r
a
ll
c
ons
e
que
nc
e
s
.
T
r
a
dit
ionally,
HD
is
d
iagnos
e
d
by
a
na
lyzing
th
e
pa
ti
e
nt's
medic
a
l
ba
c
kgr
ound,
c
a
r
r
yi
ng
out
a
thor
ough
phys
ica
l
e
xa
mi
na
ti
on,
a
nd
a
s
s
e
s
s
ing
the
r
e
leva
nt
s
igns
by
the
phys
icia
n.
T
his
tr
a
dit
ional
d
iagnos
is
,
howe
ve
r
,
c
a
n
be
inac
c
ur
a
te
a
nd
is
c
os
tl
y
a
nd
ti
me
-
c
ons
umi
ng.
T
he
us
e
of
a
r
ti
f
icia
l
int
e
l
li
ge
nc
e
(
AI
)
methods
,
pa
r
ti
c
ular
ly
mac
hine
lea
r
ni
ng
(
M
L
)
a
lgor
it
h
ms
,
is
one
pos
s
ibl
e
a
ppr
oa
c
h
to
ove
r
c
omi
ng
thes
e
obs
tac
les
.
M
L
,
a
br
a
nc
h
of
AI
,
a
ppli
e
s
a
lgor
it
hms
to
da
ta
a
na
lys
is
s
o
that
c
omput
e
r
s
c
a
n
r
e
c
ognize
,
lea
r
n,
s
pot
pa
tt
e
r
ns
,
a
nd
make
inf
or
med
judgm
e
nts
.
M
L
a
lgor
it
hms
ope
r
a
t
e
on
a
mathe
matica
l
model
that
r
e
li
e
s
on
a
tr
a
ini
n
g
da
tas
e
t
to
pr
e
dict
outcome
s
or
make
de
c
is
ions
without
e
xpli
c
it
pr
og
r
a
mm
ing
[
5]
.
B
y
a
na
lyzing
medic
a
l
r
e
c
or
ds
,
thes
e
a
lgor
it
hms
c
a
n
r
e
c
ognize
pe
r
s
ons
who
mi
g
ht
de
ve
lop
HD
,
lea
ding
to
e
a
r
li
e
r
diagnos
is
a
nd
t
r
e
a
tm
e
nt
a
nd
e
ve
ntually
lowe
r
ing
mo
r
talit
y
r
a
tes
.
E
ve
r
y
ye
a
r
,
a
ppr
oxim
a
tely
17.
9
m
il
li
on
li
ve
s
a
r
e
c
l
a
im
e
d
by
C
VD
s
,
making
them
the
ma
jor
c
a
us
e
of
f
a
talit
ies
globally,
a
c
c
or
ding
to
da
ta
f
r
om
the
W
or
ld
He
a
lt
h
Or
ga
niza
ti
on
(
W
HO
)
[
6]
.
Ac
c
or
di
ng
to
the
W
or
ld
He
a
lt
h
F
e
de
r
a
ti
on's
(
W
HF)
W
or
ld
He
a
lt
h
R
e
por
t
2023,
C
VD
s
c
laimed
the
li
ve
s
of
20.
5
mi
ll
io
n
pe
ople
in
2021,
a
c
c
ounti
ng
f
or
r
oughly
one
-
thi
r
d
o
f
glob
a
l
mor
talit
y
.
I
n
1990,
the
r
e
we
r
e
12.
1
mi
ll
ion
de
a
ths
f
r
om
C
VD
.
How
e
ve
r
,
thi
s
number
ha
s
s
igni
f
ica
ntl
y
incr
e
a
s
e
d.
I
f
nothi
ng
is
done
to
pr
e
ve
nt
it
,
by
2030
,
th
e
global
de
a
th
tol
l
is
e
xpe
c
ted
to
r
e
a
c
h
a
r
ound
22
mi
ll
ion
[
7]
.
Ac
c
or
ding
to
the
da
ta,
HD
is
a
s
e
r
ious
univer
s
a
l
he
a
lt
h
c
onc
e
r
n,
highl
ight
ing
the
ne
e
d
f
or
mor
e
s
tudy
in
t
h
is
a
r
e
a
.
R
e
c
e
nt
de
ve
lopm
e
nts
in
M
L
ha
ve
gr
e
a
tl
y
e
nha
n
c
e
d
HD
pr
e
diction
thr
ough
the
us
e
of
e
ns
e
mbl
e
tec
hniques
s
uc
h
a
s
r
a
ndom
f
or
e
s
t
(
R
F
)
a
nd
e
xt
r
e
me
gr
a
dient
boos
ti
ng
(
XG
B
)
,
f
e
a
tur
e
s
e
lec
ti
on
methods
,
int
e
gr
a
ti
on
of
f
e
a
tu
r
e
s
e
lec
ti
on
methods
with
m
e
tahe
ur
is
ti
c
opti
mi
z
a
ti
on
tec
hniques
,
a
nd
the
c
r
e
a
ti
on
of
hybr
id
models
that
c
ombi
ne
tr
a
dit
ional
M
L
lea
r
ni
ng
with
de
e
p
lea
r
ning
.
T
he
s
e
models
pe
r
f
or
m
be
t
ter
than
c
onve
nti
ona
l
M
L
tec
hniques
by
identif
ying
in
tr
i
c
a
te
da
ta
pa
tt
e
r
ns
.
B
ut
e
ve
n
with
thes
e
a
dva
nc
e
ments
,
a
thor
ough
e
va
luation
of
va
r
ious
M
L
a
lgor
it
hms
is
s
ti
ll
r
e
quir
e
d
to
a
s
c
e
r
tain
how
we
ll
they
pe
r
f
or
m
in
diver
s
e
s
c
e
na
r
ios
.
M
a
ny
c
ur
r
e
ntl
y
a
va
il
a
ble
r
e
s
e
a
r
c
h
c
onc
e
ntr
a
tes
on
s
pe
c
if
ic
models
without
a
s
s
e
s
s
ing
their
r
e
lative
a
dva
ntage
s
,
dis
a
dva
ntage
s
,
a
nd
e
f
f
e
c
ti
ve
ne
s
s
.
T
h
e
c
hief
pur
pos
e
of
thi
s
a
na
lyt
ica
l
s
tudy
is
to
e
va
l
ua
te
a
nd
c
ontr
a
s
t
s
e
ve
r
a
l
M
L
models
,
of
f
e
r
ing
a
s
ys
tema
ti
c
pe
r
f
or
manc
e
a
na
lys
is
to
c
hoos
e
the
mos
t
pr
e
c
is
e
,
e
f
f
e
c
ti
ve
,
a
nd
r
e
li
a
ble
a
lgor
it
hm
by
e
xa
mi
ning
r
e
s
e
a
r
c
h
que
s
ti
ons
(
R
Qs
)
that
will
he
lp
he
a
lt
hc
a
r
e
ins
ti
tut
ions
a
s
we
ll
a
s
hos
pit
a
ls
in
a
dva
nc
ing
the
knowle
dge
a
nd
dir
e
c
t
ing
the
de
ve
lopm
e
nt
of
ne
w
he
a
lt
hc
a
r
e
a
ppli
c
a
ti
ons
.
T
he
R
Qs
include
:
i)
w
hich
M
L
a
lgor
it
hms
a
r
e
f
r
e
q
ue
ntl
y
us
e
d
f
or
p
r
e
dicting
HD
?
a
nd
ii
)
w
hich
of
thes
e
a
lgor
it
hms
de
mons
tr
a
te
s
upe
r
ior
pe
r
f
or
manc
e
i
n
HD
pr
e
diction?
T
o
a
ns
we
r
thes
e
R
Qs
,
a
thor
ough
e
xa
mi
na
ti
on
of
r
e
leva
nt
li
ter
a
tur
e
is
r
e
qui
r
e
d,
a
s
e
l
a
bo
r
a
ted
in
the
f
oll
owing
s
e
gment.
T
his
wor
k
is
or
ga
nize
d
int
o
dif
f
e
r
e
nt
s
e
c
ti
ons
.
An
ove
r
view
of
HD
,
including
it
s
types
,
s
ym
ptom
s
,
pr
im
a
r
y
r
is
k
f
a
c
tor
s
,
s
tatis
ti
c
s
,
c
ur
r
e
nt
s
tate
of
t
he
a
r
t,
a
nd
objec
ti
ve
o
f
the
s
tudy
is
given
in
s
e
c
ti
on
1.
T
he
wor
k
of
mul
ti
ple
r
e
s
e
a
r
c
he
r
s
on
the
e
a
r
ly
de
tec
ti
on
of
HD
us
ing
va
r
ious
c
onve
nti
ona
l
a
nd
hy
br
id
M
L
models
is
c
ompi
led
in
s
e
c
ti
on
2
.
T
h
e
tec
hnique
s
e
mpl
oye
d
in
thi
s
inves
ti
ga
ti
on
f
or
identi
f
ying
HD
a
r
e
de
s
c
r
ibed
in
s
e
c
ti
on
3.
T
he
f
indi
ngs
f
r
om
the
e
xpe
r
im
e
nt
a
nd
a
c
ompr
e
he
ns
ive
a
na
lys
is
a
r
e
pr
ovided
in
s
e
c
ti
on
4.
F
inally
,
the
las
t
s
e
c
ti
on
s
ums
up
the
f
in
dings
a
nd
make
s
r
e
c
omm
e
nda
ti
ons
f
or
a
ddit
ional
r
e
s
e
a
r
c
h
a
nd
s
tudy
im
pli
c
a
ti
ons
.
2.
RE
L
AT
E
D
WORK
R
e
s
e
a
r
c
he
r
s
pr
e
dicte
d
HD
us
ing
a
r
a
nge
o
f
M
L
a
ppr
oa
c
he
s
.
E
xtens
ive
r
e
s
e
a
r
c
h
ha
s
a
l
r
e
a
dy
be
e
n
done
a
nd
is
c
onti
nuing
f
o
r
f
ur
ther
e
nha
nc
e
ments
in
pr
e
diction
.
Nume
r
ous
publi
c
a
ti
ons
c
ove
r
ing
t
he
ye
a
r
s
2018
to
2024
ha
ve
be
e
n
c
omp
il
e
d
f
r
om
r
e
s
our
c
e
s
li
ke
I
E
E
E
Xplo
r
e
,
Google
S
c
holar
,
R
e
s
e
a
r
c
hGa
te,
a
nd
S
c
ienc
e
Dir
e
c
t
to
a
ddr
e
s
s
R
Q1.
T
his
s
e
c
ti
on
pr
ovides
ins
ight
int
o
d
if
f
e
r
e
nt
M
L
p
r
e
diction
mo
de
ls
f
or
pr
e
dicting
HD
.
Ha
q
e
t
al
.
[
8
]
p
r
opos
e
d
a
hybr
id
s
mar
t
M
L
pr
e
dictive
a
ppr
oa
c
h
f
o
r
identi
f
ying
HD
.
S
e
ve
n
we
ll
-
known
c
las
s
if
ier
s
logi
s
ti
c
r
e
gr
e
s
s
ion
(
L
R
)
,
a
r
ti
f
icia
l
ne
ur
a
l
ne
twor
k
(
AN
N)
,
K
-
ne
a
r
e
s
t
ne
ighbor
(
KN
N)
,
na
ïve
B
a
ye
s
(
NB
)
,
s
uppor
t
ve
c
tor
mac
hine
(
S
V
M
)
,
R
F
,
a
nd
de
c
is
ion
tr
e
e
(
D
T
)
we
r
e
us
e
d
to
a
c
hieve
thi
s
.
T
hr
e
e
a
lgor
it
hms
we
r
e
us
e
d
to
f
ind
out
the
mos
t
s
igni
f
ica
nt
f
e
a
tur
e
s
:
r
e
li
e
f
,
lea
s
t
a
bs
olut
e
s
hr
ink
a
ge
a
nd
s
e
lec
ti
on
ope
r
a
tor
(
L
ASS
O)
,
a
nd
mi
nim
um
r
e
dund
a
nc
y
maximum
r
e
leva
nc
e
(
m
R
M
R
)
.
T
he
C
leve
lan
d
da
tas
e
t
wa
s
uti
li
z
e
d
f
or
model
a
s
s
e
s
s
ment,
a
nd
the
outcome
s
we
r
e
va
li
da
ted
us
ing
K
-
f
old
c
r
os
s
-
va
li
da
ti
on.
T
he
r
e
li
e
f
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
P
e
r
for
manc
e
analys
is
and
c
ompar
i
s
on
of
mac
hine
lear
ning
algor
it
hms
for
pr
e
dicting
he
ar
t
(
N
e
ha
B
h
adu)
2851
a
lgor
it
hm
he
lped
a
c
hieve
a
n
a
c
c
ur
a
c
y
of
89%
with
L
R
us
ing
10
-
f
old
c
r
os
s
-
va
li
da
ti
on.
M
oha
n
e
t
al
.
[
9]
mer
ge
d
the
be
ne
f
it
s
o
f
the
li
ne
a
r
method
(
L
M
)
a
lo
ng
with
R
F
to
c
r
e
a
te
the
hy
br
id
r
a
ndom
f
or
e
s
t
li
ne
a
r
model
(
HR
F
L
M
)
hybr
id
methodo
logy.
T
he
model's
a
c
c
u
r
a
c
y
s
c
or
e
on
the
C
leve
land
da
tas
e
t
wa
s
88
.
7%
,
i
ndica
ti
ng
im
pr
ove
d
pe
r
f
or
manc
e
with
the
us
e
of
a
n
R
s
tudi
o
r
a
tt
le.
B
a
s
hir
e
t
al
.
[
10]
int
e
nde
d
to
incr
e
a
s
e
the
leve
l
of
a
c
c
ur
a
c
y
of
HD
identif
ica
ti
on
by
uti
li
z
ing
f
e
a
tur
e
s
e
lec
ti
on
methods
.
T
he
y
c
onduc
ted
e
xpe
r
im
e
n
ts
us
ing
va
r
ious
M
L
c
la
s
s
if
ier
s
na
mely
S
V
M
,
L
R
,
NB
,
DT
,
a
nd
R
F
on
a
n
HD
da
tas
e
t
obtaine
d
f
r
om
Unive
r
s
it
y
of
C
a
li
f
or
nia,
I
r
v
i
ne
(
UC
I
)
us
ing
the
r
a
pid
mi
ne
r
tool
.
T
he
f
ind
ings
indi
c
a
ted
that
LR
a
nd
N
B
,
e
xhibi
ted
im
pr
ove
d
a
c
c
ur
a
c
y.
R
e
pa
ka
e
t
al
.
[
11]
,
pr
opos
e
d
a
s
mar
t
he
a
r
t
d
is
e
a
s
e
pr
e
diction
s
ys
tem
(
S
HD
P
)
,
by
incor
por
a
ti
ng
the
NB
c
las
s
if
ier
a
long
with
a
n
a
dv
a
nc
e
d
e
nc
r
ypti
on
s
tanda
r
d
(
AE
S
)
f
o
r
p
r
e
dicting
HD
.
T
he
r
e
s
ult
s
indi
c
a
ted
that
thi
s
a
ppr
oa
c
h
outper
f
or
med
NB
,
a
c
hieving
a
n
a
c
c
ur
a
c
y
r
a
te
of
89.
77%
.
F
ur
t
he
r
mor
e
,
AE
S
de
mons
t
r
a
ted
s
upe
r
ior
s
e
c
ur
it
y
pe
r
f
or
man
c
e
whe
n
c
ompar
e
d
to
pa
r
a
ll
e
l
homom
or
phic
e
nc
r
ypti
on
a
lgor
it
hm
(
P
H
E
A
)
.
F
it
r
iyani
e
t
al
.
[
12]
pr
opos
e
d
HD
P
M
to
pr
e
dict
HD
.
T
o
e
nha
nc
e
the
a
c
c
ur
a
c
y,
the
model
int
e
gr
a
ted
s
ynthetic
mi
nor
it
y
ove
r
s
a
mpl
ing
tec
hni
que
-
e
dit
e
d
ne
a
r
e
s
t
ne
ighbor
s
(
S
M
OT
E
-
E
NN
)
,
a
nd
de
ns
it
y
-
ba
s
e
d
s
pa
ti
a
l
c
lus
ter
ing
of
a
ppli
c
a
ti
ons
with
noi
s
e
(
DB
S
C
AN
)
a
long
with
XG
B
oos
t
M
L
c
las
s
if
ier
.
T
he
tr
a
ini
ng
da
tas
e
t
wa
s
ba
lanc
e
d
us
ing
S
M
OT
E
-
E
NN
.
DB
S
C
AN
wa
s
us
e
d
f
or
de
tec
ti
ng
a
nd
r
e
movi
ng
outl
ier
da
ta,
a
nd
XG
B
oos
t
wa
s
us
e
d
f
or
ge
ne
r
a
ti
ng
the
pr
e
dictive
model
.
T
he
model
wa
s
c
ons
tr
uc
ted
us
ing
the
C
leve
land
a
nd
the
S
tatlog
da
tas
e
ts
.
I
n
the
e
va
luation
s
tage
,
he
a
r
t
dis
e
a
s
e
pr
e
diction
model
(
HD
P
M
)
outper
f
or
med
s
ix
other
M
L
a
lgor
it
hms
,
e
xhibi
ti
ng
a
s
upe
r
ior
a
c
c
ur
a
c
y
s
c
or
e
of
98.
40%
on
the
C
leve
land
a
nd
95.
90%
on
the
s
tatlog
da
tas
e
t.
Ka
tar
ya
a
nd
M
e
e
na
[
13]
us
e
d
the
UC
I
da
tas
e
t
to
e
xa
mi
ne
the
e
f
f
e
c
ti
v
e
ne
s
s
of
many
M
L
methods
,
c
ompr
is
ing
KN
N,
L
R
,
NB
,
S
VM
,
DT
,
R
F
,
M
L
P
,
AN
N,
a
nd
DN
N,
in
p
r
e
dic
ti
ng
HD
.
R
F
wa
s
identif
ied
a
s
the
mos
t
a
c
c
ur
a
te
a
lgor
it
hm
of
a
ll
.
L
i
e
t
al
.
[
14
]
,
de
ve
loped
a
n
HD
pr
e
dictio
n
model
us
ing
KN
N,
S
VM
,
L
R
,
NB
,
A
NN
,
a
nd
DT
c
las
s
if
ier
s
of
M
L
.
Dif
f
e
r
e
nt
methods
s
uc
h
a
s
m
R
M
R
,
r
e
li
e
f
,
loca
l
lea
r
ning,
a
nd
L
ASS
O
we
r
e
us
e
d
to
e
li
mi
na
te
i
r
r
e
leva
nt
a
nd
r
e
dunda
nt
a
tt
r
ibu
tes
.
T
he
c
r
os
s
-
v
a
li
da
ti
on
tec
hnique
uti
li
z
e
d
wa
s
“
lea
ve
-
one
-
s
ubjec
t
-
out”.
Ac
c
or
ding
to
the
s
tudy,
the
s
ug
ge
s
ted
f
e
a
tur
e
s
e
lec
ti
on
method
(
F
C
M
I
M
)
wor
ks
we
ll
whe
n
pa
ir
e
d
with
S
VM
to
c
r
e
a
te
a
n
a
dva
nc
e
d
int
e
ll
igent
s
ys
tem
f
or
HD
identif
ica
ti
on.
T
ha
kka
r
e
t
al
.
[
15
]
de
ve
loped
a
f
r
a
mew
or
k
to
c
onduc
t
a
c
ompr
e
he
ns
ive
pe
r
f
or
manc
e
a
na
lys
is
of
f
ive
M
L
methods
s
pe
c
if
ica
ll
y
KN
N,
L
R
,
S
VM
,
NB
,
a
nd
R
F
.
T
he
tes
ti
ng
wa
s
done
us
ing
the
C
leve
land
HD
da
tas
e
t.
T
he
major
it
y
of
pe
r
f
o
r
manc
e
metr
ics
indi
c
a
ted
that
L
R
outper
f
o
r
med
the
other
c
las
s
if
ier
s
c
ons
is
tently.
S
ha
h
e
t
al
.
[
16
]
a
ppli
e
d
the
C
leve
land
HD
da
tas
e
t
to
f
our
M
L
c
las
s
if
ica
ti
on
tec
hniques
:
DT
,
R
F
,
KN
N,
a
nd
NB
.
W
a
ikato
e
nvir
on
ment
f
o
r
know
ledge
a
na
lys
is
(
W
E
KA
)
wa
s
us
e
d
f
or
c
a
r
r
ying
out
the
a
na
lys
is
.
T
he
f
indi
ngs
r
e
ve
a
led
that
KN
N
y
ielde
d
the
highes
t
a
c
c
ur
a
c
y
s
c
or
e
.
S
ha
r
ma
e
t
al
.
[
17]
c
r
e
a
ted
a
n
M
L
model
us
ing
f
ou
r
d
if
f
e
r
e
nt
c
las
s
if
ier
s
:
R
F
,
S
V
M
,
NB
,
a
nd
D
T
.
T
he
e
xpe
r
im
e
nt
us
e
d
a
n
HD
da
ta
s
e
t
f
r
om
UC
I
.
T
he
r
e
s
ult
s
s
howe
d
that
R
F
a
tt
a
ined
a
99
%
a
c
c
ur
a
c
y
r
a
te
in
a
mor
e
e
f
f
icie
nt
pr
e
diction
ti
mef
r
a
me.
Hos
s
e
n
e
t
al
.
[
18]
uti
li
z
e
d
thr
e
e
M
L
c
las
s
if
ier
s
na
mely
R
F
,
DT
,
a
nd
L
R
f
or
p
r
e
dicting
HD
,
a
nd
their
c
ompar
a
ti
ve
a
s
s
e
s
s
ment
wa
s
done
.
T
he
e
xpe
r
im
e
ntation
wa
s
c
a
r
r
ied
out
us
ing
the
UC
I
C
leve
land
da
taba
s
e
.
L
R
ha
d
the
highes
t
a
c
c
ur
a
c
y
s
c
or
e
of
92
.
10%
,
making
it
the
be
s
t
pe
r
f
or
mer
ove
r
a
ll
.
B
a
s
hir
e
t
al
.
[
19]
pr
opos
e
d
a
vo
ti
ng
s
ys
tem
us
ing
a
n
e
ns
e
mbl
e
a
ppr
oa
c
h
to
a
c
c
ur
a
tely
pr
e
dict
HD
.
F
o
r
tes
ti
ng
pur
pos
e
s
,
f
our
HD
da
tas
e
ts
s
our
c
e
d
f
r
om
the
UC
I
r
e
pos
it
or
y
we
r
e
uti
li
z
e
d.
Outc
omes
s
howe
d
that
the
e
ns
e
mbl
e
s
c
he
me
a
c
hieve
d
a
n
a
c
c
ur
a
c
y
of
83%
,
outper
f
or
m
ing
other
e
ns
e
mbl
e
s
c
he
mes
a
nd
indi
vidual
c
l
a
s
s
if
ier
s
.
R
a
ni
e
t
al
.
[
20
]
c
r
e
a
ted
a
hybr
id
a
ppr
oa
c
h
-
ba
s
e
d
de
c
is
ion
s
uppor
t
s
ys
tem
f
o
r
HD
pr
e
diction
.
F
or
s
e
lec
ti
ng
the
mos
t
r
e
leva
nt
f
e
a
tur
e
s
,
a
hybr
id
a
lgo
r
it
hm
that
in
tegr
a
ted
r
e
c
ur
s
ive
f
e
a
tur
e
e
li
mi
na
ti
on
(
R
F
E
)
a
lon
g
with
a
ge
ne
ti
c
a
lgor
it
hm
(
GA
)
wa
s
uti
li
z
e
d.
T
he
C
leve
land
HD
da
tas
e
t
wa
s
us
e
d
f
o
r
model
tes
ti
ng.
P
r
e
-
pr
oc
e
s
s
ing
of
the
da
ta
wa
s
done
us
ing
s
tanda
r
d
s
c
a
lar
tec
hniq
ue
s
a
nd
S
M
OT
E
.
M
is
s
ing
va
lues
we
r
e
ha
ndled
by
a
pplyi
ng
the
mul
ti
va
r
iate
im
putation
by
c
ha
ined
e
qua
ti
ons
t
e
c
hnique.
F
inally
,
f
ive
M
L
tec
hniques
:
L
R
,
S
VM
,
NB
,
R
F
,
a
nd
a
da
pti
ve
boos
ti
ng
(
Ada
B
oos
t
)
we
r
e
us
e
d.
T
he
hybr
id
s
ys
tem
pe
r
f
o
r
med
e
xc
e
pti
ona
ll
y
we
ll
with
a
n
a
c
c
ur
a
c
y
of
86.
6
%
.
Ghos
h
e
t
al
.
[
21]
de
ve
lope
d
a
hybr
id
model
by
c
ombi
ning
ba
gging
a
nd
boos
ti
ng
tec
hniques
with
f
ive
c
onve
nti
ona
l
M
L
c
las
s
if
ier
s
.
B
a
gging
wa
s
a
ppli
e
d
to
KN
N,
DT
,
a
nd
R
F
r
e
s
ult
ing
in
K
-
ne
a
r
e
s
t
ne
ighbor
s
ba
gging
method
(
KN
NB
M
)
,
de
c
is
ion
tr
e
e
ba
gging
method
(
D
T
B
M
)
,
a
nd
r
a
ndo
m
f
o
r
e
s
t
ba
gging
method
(
R
F
B
M
)
hybr
id
methods
.
B
oos
ti
n
g
wa
s
a
ppli
e
d
to
Ada
B
oos
t
a
nd
gr
a
dient
boos
ti
ng
r
e
s
ult
ing
in
Ada
B
oos
t
boos
ti
ng
method
(
AB
B
M
)
a
nd
g
r
a
dient
boos
ti
ng
boos
ti
ng
method
(
GB
B
M
)
hybr
id
methods
.
F
or
s
e
lec
ti
ng
r
e
leva
nt
f
e
a
tur
e
s
L
ASS
O
a
nd
r
e
li
e
f
tec
hniques
we
r
e
e
mpl
oye
d.
A
c
ompr
e
he
ns
ive
da
tas
e
t
c
ompr
is
ing
f
ive
be
nc
hmar
k
da
tas
e
ts
,
C
leve
land,
S
tatlog,
Hung
a
r
ian
,
S
witze
r
land
,
a
nd
L
ong
B
e
a
c
h
VA
f
or
HD
,
wa
s
us
e
d
to
c
onduc
t
the
s
tudi
e
s
.
T
he
f
indi
n
gs
r
e
ve
a
led
that
R
F
B
M
a
long
with
r
e
li
e
f
f
e
a
tur
e
s
e
lec
ti
on
outper
f
or
med
other
s
with
a
n
a
c
c
ur
a
c
y
of
99.
05
%
.
As
hr
i
e
t
al
.
[
22]
p
r
opos
e
d
a
n
innovative
hybr
id
in
telli
ge
nt
f
r
a
mew
or
k,
int
e
gr
a
ti
ng
f
ive
M
L
methodologi
e
s
i
nc
ludi
ng
KN
N,
S
VM
,
L
R
,
DT
,
a
nd
R
F
with
a
major
it
y
voti
ng
tec
hnique.
Additi
ona
ll
y
,
a
s
im
ple
ge
ne
ti
c
a
lgor
it
hm
(
S
GA
)
wa
s
e
mpl
oye
d
f
o
r
f
e
a
tur
e
s
e
lec
ti
on,
im
pr
oving
pr
e
di
c
ti
on
pe
r
f
or
manc
e
a
nd
r
e
duc
ing
ove
r
a
ll
ti
me
c
ons
umpt
ion.
Ove
r
f
it
ti
ng
wa
s
a
ddr
e
s
s
e
d
by
us
ing
10
-
f
old
c
r
os
s
-
va
li
da
ti
on.
T
he
UC
I
HD
da
tas
e
t
wa
s
uti
li
z
e
d
f
or
the
e
xpe
r
im
e
nts
.
T
he
outcome
s
s
howe
d
that
the
e
ns
e
mbl
e
tec
hnique
a
c
c
ompl
is
he
d
a
r
e
m
a
r
ka
ble
a
c
c
ur
a
c
y
of
98
.
18%
.
Ali
e
t
al
.
[
23
]
c
a
r
r
i
e
d
out
a
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14
,
No.
4
,
Augus
t
2025
:
284
9
-
2863
2852
c
ompar
a
ti
ve
e
va
luation
of
va
r
ious
M
L
c
las
s
if
ier
s
.
A
f
e
a
tur
e
im
por
tanc
e
s
c
or
e
wa
s
c
omput
e
d
a
c
r
os
s
a
ll
c
las
s
if
ier
s
e
xc
e
pt
f
or
KN
N
a
nd
M
L
P
.
T
h
is
s
c
or
e
wa
s
us
e
d
to
r
a
te
e
a
c
h
f
e
a
tur
e
.
T
he
HD
da
tas
e
t
wa
s
obtaine
d
Ka
ggle
M
L
r
e
pos
it
or
y.
T
he
f
indi
ngs
r
e
ve
a
led
th
a
t
thr
e
e
c
las
s
if
ier
s
na
mely
DT
,
R
F
,
a
nd
KN
N
a
c
hieve
d
e
qua
ll
y
outs
tanding
pe
r
f
or
manc
e
with
100%
a
c
c
ur
a
c
y,
s
e
ns
it
ivi
ty,
a
nd
s
pe
c
if
icity.
I
s
ha
q
e
t
al
.
[
24
]
e
mpl
oye
d
nine
M
L
c
la
s
s
if
ier
s
s
uc
h
a
s
L
R
,
S
VM
,
DT
,
R
F
,
s
tocha
s
ti
c
gr
a
dient
c
las
s
if
ier
(
S
G
C
)
,
Ada
B
oos
t,
gr
a
dient
boos
ti
ng
c
las
s
if
ier
(
GB
M
)
,
g
a
us
s
ian
na
ive
B
a
ye
s
(
GN
B
)
,
a
nd
e
xtr
a
tr
e
e
c
las
s
if
ier
(
E
T
C
)
in
thi
s
s
tudy.
T
he
c
las
s
im
ba
lanc
e
is
s
ue
wa
s
a
ddr
e
s
s
e
d
with
S
M
OT
E
.
Additi
ona
ll
y,
the
models
we
r
e
tr
a
ined
on
top
f
e
a
tur
e
s
c
hos
e
n
by
R
F
.
T
he
r
e
s
ult
s
s
howe
d
that
E
T
C
wi
th
S
M
OT
E
pe
r
f
or
med
the
be
s
t,
r
e
a
c
hing
a
n
a
c
c
ur
a
c
y
of
92.
62%
.
C
ha
ng
e
t
al
.
[
25]
c
r
e
a
ted
a
P
ython
-
ba
s
e
d
a
ppli
c
a
ti
on
to
de
tec
t
HD
with
im
pr
ove
d
pr
e
c
is
ion.
T
he
model
wa
s
c
ons
tr
uc
ted
us
ing
a
n
R
F
c
las
s
if
ier
.
T
he
a
ppli
c
a
ti
on
a
tt
a
ined
a
r
e
mar
ka
ble
a
c
c
ur
a
c
y
r
a
te
of
83%
.
A
b
de
ll
a
t
if
e
t
a
l
.
[
26]
s
ug
ge
s
te
d
a
n
e
f
f
ic
ie
nt
a
p
pr
oa
c
h
to
c
on
s
tr
u
c
t
th
e
m
od
e
l
b
y
c
om
bi
nin
g
S
M
OT
E
,
e
x
tr
a
tr
e
e
s
(
E
T
)
,
a
nd
h
yp
e
r
b
a
nd
(
H
B
)
te
c
hn
iq
ue
s
.
S
M
OT
E
w
a
s
u
s
e
d
to
r
e
s
ol
ve
c
la
s
s
ine
q
ua
li
ty,
E
T
w
a
s
u
s
e
d
f
or
c
la
s
s
if
i
c
a
ti
on
a
n
d
H
B
w
a
s
u
s
e
d
f
or
op
ti
mi
z
a
ti
on
of
hy
pe
r
-
p
a
r
a
m
e
t
e
r
s
.
F
o
r
pr
e
d
ic
ti
ng
t
h
e
s
e
v
e
r
i
ty
l
e
v
e
l
of
H
D,
s
i
x
d
i
s
t
in
c
t
M
L
c
l
a
s
s
i
f
i
e
r
s
,
na
me
ly
L
R
,
S
V
M
,
K
N
N,
E
T
,
s
to
c
h
a
s
t
ic
g
r
a
di
e
n
t
d
e
s
c
e
nt
(
S
G
D)
,
a
nd
X
G
B
o
o
s
t
w
e
r
e
e
mp
lo
ye
d
.
T
he
e
x
pe
r
im
e
n
t
a
ti
on
wa
s
c
o
nd
uc
te
d
ut
il
i
z
i
ng
t
h
e
C
l
e
v
e
la
nd
a
nd
S
ta
tl
og
d
a
t
a
s
e
t
s
.
T
h
e
ou
tc
om
e
s
r
e
v
e
a
l
e
d
t
ha
t
th
e
hi
gh
e
s
t
a
c
c
ur
a
c
y
of
99.
2%
a
n
d
9
8.
5
2%
w
a
s
a
c
h
ie
ve
d
b
y
S
M
OT
E
a
n
d
E
T
o
pt
im
i
z
e
d
b
y
HB
,
r
e
s
pe
c
ti
v
e
ly.
Ah
ma
d
e
t
a
l.
[
2
7]
c
o
ndu
c
te
d
a
p
e
r
f
or
ma
nc
e
i
nv
e
s
t
ig
a
t
io
n
of
va
r
io
u
s
M
L
c
l
a
s
s
if
ier
s
in
c
l
ud
in
g
S
V
M
,
KN
N,
DT
,
R
F
,
G
B
C
,
a
nd
li
ne
a
r
d
i
s
c
r
im
in
a
nt
s
a
n
a
l
y
s
i
s
(
L
DA
)
.
T
o
s
e
le
c
t
th
e
m
o
s
t
s
i
gn
if
i
c
a
n
t
f
e
a
t
ur
e
s
,
a
s
e
q
u
e
nt
i
a
l
f
e
a
tur
e
s
e
l
e
c
ti
on
t
e
c
h
ni
qu
e
w
a
s
u
s
e
d.
E
m
plo
yi
ng
t
h
e
K
-
f
old
c
r
o
s
s
-
v
a
li
da
ti
on
t
e
c
hn
iq
u
e
,
ve
r
if
i
c
a
t
io
n
w
a
s
c
o
mp
le
te
d.
T
h
e
c
o
mbi
n
e
d
(
S
t
a
t
lo
g+
C
l
e
ve
la
nd
+
Hu
ng
a
r
y)
d
a
t
a
s
e
t,
t
og
e
th
e
r
w
i
t
h
t
he
in
di
vi
du
a
l
da
ta
s
e
t
s
f
r
o
m
C
le
v
e
l
a
n
d,
H
u
ng
a
r
y,
S
wi
tz
e
r
la
nd,
a
n
d
L
on
g
B
e
a
c
h
V,
w
e
r
e
u
s
e
d
t
o
e
v
a
l
u
a
t
e
ho
w
w
e
l
l
th
e
mo
de
l
pe
r
f
or
med
.
W
i
th
n
e
a
r
l
y
s
im
il
a
r
f
i
nd
in
g
s
of
1
00
a
n
d
99.
40%
f
or
t
he
f
ir
s
t
d
a
t
a
s
e
t
a
nd
10
0
a
n
d
9
9.
7
6%
f
or
t
h
e
s
e
c
on
d,
r
e
s
p
e
c
ti
v
e
l
y,
t
h
e
R
F
s
e
qu
e
n
ti
a
l
f
e
a
tur
e
s
e
l
e
c
t
io
n
(
S
F
S
)
a
nd
DT
S
F
S
s
ho
we
d
th
e
gr
e
a
te
s
t
a
c
c
ur
a
c
y
v
a
lu
e
s
f
or
bo
th
d
a
t
a
s
e
t
s
.
A
hm
a
d
e
t
al
.
[
28]
u
ti
li
z
e
d
Gr
idS
e
a
r
c
h
C
V
i
n
c
on
ju
nc
ti
on
w
it
h
mul
ti
pl
e
M
L
m
e
t
ho
d
s
s
u
c
h
a
s
S
V
M
,
L
R
,
K
NN
,
a
n
d
X
G
B
oo
s
t
f
o
r
i
d
e
nt
if
yi
ng
H
D.
F
ur
th
e
r
,
a
c
om
pa
r
a
ti
ve
s
tu
dy
w
a
s
c
on
d
uc
te
d.
F
iv
e
f
old
c
r
o
s
s
-
v
a
li
d
a
ti
o
n
wa
s
u
s
e
d
a
s
a
v
e
r
i
f
i
c
a
ti
o
n
a
p
pr
o
a
c
h.
T
h
e
d
a
t
a
s
e
t
s
f
r
om
U
C
I
K
a
g
gl
e
,
L
o
ng
B
e
a
c
h
V
,
H
un
ga
r
y,
S
wit
z
e
r
l
a
nd,
a
n
d
C
le
v
e
l
a
n
d
w
e
r
e
ut
il
iz
e
d
to
a
s
s
e
s
s
t
h
e
s
y
s
t
e
m.
T
he
ou
tc
om
e
s
d
e
mo
n
s
t
r
a
te
d
t
h
a
t,
wh
e
n
c
o
mb
in
e
d,
X
GB
oo
s
t
a
nd
Gr
i
dS
e
a
r
c
hC
V
g
e
n
e
r
a
t
e
d
th
e
u
tm
os
t
a
n
d
a
ppr
ox
im
a
t
e
l
y
e
qu
iv
a
l
e
nt
t
e
s
t
in
g
a
s
w
e
l
l
a
s
tr
a
i
ni
ng
a
c
c
ur
a
t
e
ne
s
s
l
e
v
e
ls
of
1
00
a
nd
9
9.
0
3%
o
n
bo
th
d
a
t
a
s
e
t
s
.
A
bd
e
l
la
ti
f
e
t
al
.
[
29]
o
f
f
e
r
e
d
a
n
ov
e
l
s
t
r
a
te
gy
t
h
a
t
u
s
e
d
i
mpr
ov
e
d
we
igh
t
e
d
r
a
n
dom
f
or
e
s
t
(
I
W
R
F
)
f
or
i
d
e
nt
if
y
in
g
H
D,
B
a
y
e
s
i
a
n
o
pt
im
iz
a
t
io
n
f
or
op
ti
m
i
z
in
g
I
W
R
F
'
s
hy
pe
r
-
p
a
r
a
me
t
e
r
s
a
n
d
s
up
e
r
vi
s
e
d
“
inf
in
it
e
f
e
a
t
ur
e
s
e
le
c
ti
o
n
(
I
n
f
-
F
S
s
)
”
t
o
de
te
r
mi
n
e
i
mp
or
t
a
n
t
f
e
a
tu
r
e
s
.
T
h
e
HD
c
li
ni
c
a
l
r
e
c
or
d
s
a
n
d
t
he
S
t
a
tl
o
g
da
ta
s
e
ts
w
e
r
e
u
s
e
d
i
n
th
e
m
od
e
l
'
s
d
e
ve
lo
pm
e
n
t
a
nd
t
e
s
t
in
g.
T
h
e
r
e
s
ult
s
d
e
mo
ns
tr
a
t
e
d
t
h
a
t,
c
on
c
e
r
n
in
g
a
c
c
ur
a
c
y
a
nd
F
-
m
e
a
s
ur
e
,
I
nf
-
FSs
-
I
W
R
F
out
p
e
r
f
o
r
m
e
d
ot
he
r
m
od
e
l
s
on
bo
th
da
t
a
s
e
t
s
.
C
e
n
it
t
a
e
t
al.
[
30
]
d
e
s
i
gn
e
d
a
no
ve
l
f
e
a
tu
r
e
s
e
l
e
c
ti
on
t
e
c
h
ni
qu
e
f
or
i
s
c
h
e
m
ic
H
D
na
m
e
ly
i
s
c
h
e
m
i
c
h
e
a
r
t
d
is
e
a
s
e
s
q
uir
r
e
l
s
e
a
r
c
h
op
ti
m
iz
a
ti
o
n
(
I
H
DSS
O
)
.
T
he
m
od
e
l
'
s
e
f
f
e
c
t
iv
e
ne
s
s
wa
s
c
o
nf
ir
me
d
b
y
ut
il
iz
in
g
t
h
e
U
C
I
H
D
d
a
t
a
s
e
t.
T
h
e
o
ut
c
o
me
s
de
mo
n
s
tr
a
t
e
d
t
ha
t
th
e
I
HD
S
S
O
mo
d
e
l
c
o
uld
id
e
n
ti
f
y
t
he
mo
s
t
s
ig
nif
ica
n
t
a
tt
r
i
bu
t
e
s
wi
th
a
n
a
c
c
ur
a
c
y
r
a
t
e
of
m
or
e
t
ha
n
9
8.
3
8%
b
y
u
s
i
ng
th
e
R
F
c
la
s
s
if
i
e
r
.
Kh
a
n
e
t
a
l
.
[
31]
e
v
a
lu
a
te
d
t
h
e
e
f
f
e
c
ti
v
e
ne
s
s
of
f
iv
e
pr
e
d
ic
ti
ve
M
L
c
la
s
s
if
ier
s
i
n
c
lu
di
ng
L
R
,
S
V
M
,
NB
,
DT
,
a
nd
R
F
,
f
or
p
a
t
ie
n
t
s
wi
th
C
V
D.
T
h
e
da
t
a
wa
s
pr
o
vi
de
d
by
t
h
e
K
hy
be
r
T
e
a
c
hi
ng
H
o
s
pi
t
a
l
a
s
w
e
l
l
a
s
th
e
L
a
dy
R
e
a
ding
Ho
s
pit
a
l,
l
o
c
a
te
d
in
K
hy
b
e
r
P
r
o
vi
nc
e
,
P
a
ki
s
ta
n.
U
po
n
c
o
nd
uc
ti
ng
e
x
pl
or
a
to
r
y
a
n
a
l
y
s
i
s
,
i
t
w
a
s
r
e
v
e
a
l
e
d
t
h
a
t
R
F
h
a
d
a
tt
a
i
ne
d
t
he
gr
e
a
te
s
t
p
e
r
c
e
nt
a
ge
s
of
85.
01,
9
2.
11,
a
nd
87.
73
%
f
or
a
c
c
ur
a
c
y,
s
e
n
s
it
i
vi
ty,
a
nd
r
e
c
e
i
ve
r
o
pe
r
a
ti
ng
c
ha
r
a
c
t
e
r
is
ti
c
(
R
O
C
)
c
ur
v
e
,
r
e
s
p
e
c
ti
v
e
l
y.
U
ll
a
h
et
al
.
[
32]
i
ntr
od
uc
e
d
a
s
c
a
l
a
bl
e
M
L
-
ba
s
e
d
f
r
a
m
e
w
or
k
b
y
i
nt
e
g
r
a
ti
ng
s
op
hi
s
ti
c
a
t
e
d
f
e
a
tu
r
e
s
e
le
c
t
io
n
t
e
c
h
ni
qu
e
s
i
n
c
l
udi
ng
f
a
s
t
c
or
r
e
l
a
t
io
n
-
b
a
s
e
d
f
il
ter
(
F
C
B
F
)
,
m
R
M
R
,
r
e
li
e
f
,
a
nd
pa
r
ti
c
l
e
s
w
a
r
m
o
pt
im
i
z
a
t
io
n
(
P
S
O)
.
T
h
e
s
e
me
th
od
s
w
e
r
e
a
p
pli
e
d
t
o
e
xtr
a
c
t
a
nd
i
d
e
n
ti
f
y
t
he
mo
s
t
s
ig
ni
f
i
c
a
nt
f
e
a
tu
r
e
s
f
r
om
E
C
G
s
i
gn
a
l
s
.
T
h
e
r
e
f
i
ne
d
f
e
a
tur
e
s
e
t
wa
s
th
e
n
u
s
e
d
to
t
r
a
in
ML
c
l
a
s
s
if
i
e
r
s
s
u
c
h
a
s
ET
a
n
d
RF
,
w
hi
c
h
a
c
hi
e
v
e
d
o
ut
s
ta
nd
in
g
a
c
c
ur
a
c
y
r
a
t
e
s
o
f
10
0%
on
b
ot
h
s
m
a
l
l
a
n
d
l
a
r
g
e
d
a
t
a
s
e
t
s
.
B
i
s
wa
s
e
t
a
l
.
[
33]
u
s
e
d
thr
e
e
di
s
t
in
c
t
t
e
c
h
niq
u
e
s
to
c
ho
o
s
e
i
mp
or
ta
nt
f
e
a
tur
e
s
n
a
m
e
l
y
an
a
l
y
s
i
s
of
va
r
ia
n
c
e
(
AN
O
VA
)
,
c
h
i
-
s
q
ua
r
e
,
a
nd
mut
u
a
l
i
nf
or
m
a
t
io
n.
F
ur
t
he
r
mor
e
,
s
i
x
d
i
s
t
in
c
t
M
L
me
th
od
s
w
e
r
e
u
ti
l
iz
e
d,
c
om
pr
i
s
in
g
S
V
M
,
L
R
,
K
NN
,
N
B
,
DT
,
a
nd
R
F
.
T
h
e
s
e
mo
de
l
s
w
e
r
e
u
s
e
d
to
d
e
t
e
r
m
in
e
th
e
m
o
s
t
e
f
f
e
c
t
iv
e
mo
d
e
l
a
nd
f
e
a
tu
r
e
s
u
b
s
e
t.
F
in
a
l
ly,
i
t
wa
s
f
o
un
d
th
a
t
w
h
e
n
m
ut
ua
l
inf
or
m
a
t
io
n
f
e
a
t
ur
e
s
ub
s
e
t
s
w
e
r
e
u
s
e
d,
R
F
h
a
d
t
he
h
ig
h
e
s
t
a
c
c
u
r
a
c
y
r
a
t
e
,
a
t
9
4.
5
1%
.
R
e
s
h
a
n
et
al
.
[
34
]
d
e
v
e
l
op
e
d
a
n
e
w
hy
br
id
d
e
e
p
ne
ur
a
l
n
e
t
w
or
k
(
H
D
NN
)
m
od
e
l.
T
h
e
mo
de
l
u
s
e
d
c
o
nv
ol
uti
on
a
l
n
e
ur
a
l
ne
t
w
or
k
s
(
C
NN
)
,
AN
N,
l
on
g
s
ho
r
t
-
ter
m
m
e
m
or
y
(
L
S
T
M
)
,
a
n
d
a
n
i
nt
e
g
r
a
ti
on
of
L
S
T
M
w
it
h
C
N
N
o
ve
r
ma
n
y
la
ye
r
s
.
F
ur
t
h
e
r
to
e
nh
a
nc
e
th
e
q
ua
li
t
y
of
da
t
a
,
da
t
a
i
mp
ut
a
t
io
n
t
e
c
hn
iq
ue
s
w
e
r
e
ut
il
iz
e
d.
T
h
e
mo
d
e
l
wa
s
tr
a
in
e
d
u
s
i
ng
tw
o
d
a
t
a
s
e
t
s
,
t
h
e
C
le
v
e
l
a
n
d
a
n
d
t
he
c
o
mbi
n
e
d
HD
d
a
t
a
s
e
t,
w
hi
c
h
in
c
l
ud
e
s
da
ta
f
r
o
m
f
iv
e
b
e
n
c
hm
a
r
k
d
a
t
a
s
e
t
s
.
A
r
e
mar
k
a
bl
e
a
c
c
ur
a
c
y
r
a
te
of
9
8.
86%
wa
s
s
ho
wn
by
th
e
s
ug
ge
s
te
d
t
e
c
h
niq
u
e
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
P
e
r
for
manc
e
analys
is
and
c
ompar
i
s
on
of
mac
hine
lear
ning
algor
it
hms
for
pr
e
dicting
he
ar
t
(
N
e
ha
B
h
adu)
2853
Qa
dr
i
et
al
.
[
35]
s
ugge
s
ted
a
ne
w
method
f
or
f
e
a
tur
e
e
nginee
r
ing
in
pr
incipa
l
c
omponent
he
a
r
t
f
a
il
ur
e
(
P
C
HF
)
,
f
oc
us
ing
o
n
s
e
lec
ti
ng
the
top
e
igh
t
f
e
a
tur
e
s
to
im
p
r
ove
pe
r
f
o
r
manc
e
.
B
y
int
r
oduc
ing
a
nove
l
f
e
a
tur
e
s
e
t,
P
C
HF
wa
s
f
ine
-
tuned
to
a
c
hieve
opti
mal
a
c
c
ur
a
c
y
s
c
or
e
s
.
T
he
s
tudy
uti
li
z
e
d
nine
M
L
c
las
s
if
ier
s
to
c
onduc
t
thor
ough
a
na
lys
is
a
nd
e
va
luations
.
T
he
f
indi
ngs
indi
c
a
t
e
d
that
the
DT
method
s
ur
pa
s
s
e
d
other
M
L
models
,
a
c
hieving
a
r
e
mar
ka
ble
a
c
c
ur
a
c
y
s
c
or
e
of
100%
.
P
a
tr
a
et
al
.
[
36
]
de
ve
loped
a
highl
y
e
f
f
e
c
ti
v
e
hybr
id
voti
ng
e
ns
e
mbl
e
a
ppr
oa
c
h
to
a
c
c
ur
a
tely
identif
y
th
e
r
is
k
of
HD
.
T
he
F
r
a
mi
ngha
m
HD
da
tas
e
t's
c
ha
r
a
c
ter
is
ti
c
s
we
r
e
opti
mi
z
e
d
f
or
the
model,
a
nd
their
r
e
leva
nc
e
to
the
r
e
s
ult
wa
s
e
va
luate
d.
T
he
f
or
wa
r
d
f
e
a
tur
e
s
e
lec
ti
on
a
ppr
oa
c
h
wa
s
then
us
e
d
to
int
e
gr
a
te
thes
e
r
a
nking
f
e
a
tur
e
s
us
i
ng
t
r
a
dit
ional
c
las
s
if
ier
s
to
pr
oduc
e
meta
-
models
with
f
e
a
tur
e
we
ight
s
.
T
he
s
ugge
s
ted
hybr
id
model
wa
s
ult
im
a
tely
f
o
r
med
by
s
e
lec
ti
ng
t
he
top
5
pe
r
f
or
mi
ng
c
las
s
if
ier
s
.
T
he
r
e
s
ult
s
s
howe
d
a
r
e
mar
ka
ble
a
c
c
ur
a
c
y
r
a
te
o
f
95
.
87%
.
Ahma
d
a
nd
P
o
lat
[
37]
s
ugge
s
ted
a
n
M
L
-
ba
s
e
d
int
e
ll
igent
HD
diagnos
ti
c
model.
A
s
wa
r
m
-
ba
s
e
d
meta
he
ur
is
ti
c
tec
hnique
c
a
ll
e
d
jellyf
is
h
opti
mi
z
a
ti
o
n
wa
s
us
e
d
to
c
hoos
e
the
opti
mal
f
e
a
tur
e
s
to
ove
r
c
ome
the
ove
r
f
it
ti
ng
pr
oblem
br
ought
on
by
the
a
bunda
nc
e
of
c
ha
r
a
c
ter
is
ti
c
s
in
the
C
leve
land
da
tas
e
t.
T
he
be
s
t
c
ha
r
a
c
ter
is
ti
c
s
f
r
om
the
da
tas
e
t
we
r
e
then
c
hos
e
n,
a
nd
f
our
dis
ti
nc
t
M
L
a
lgor
it
hms
na
mely
S
VM
,
A
NN
,
D
T
,
a
nd
Ada
B
oos
t
we
r
e
e
mpl
oye
d
f
or
s
im
ulation.
All
M
L
methods
de
mons
tr
a
ted
higher
a
c
c
ur
a
c
y
r
a
tes
whe
n
us
ing
the
jellyf
is
h
tec
hnique.
T
he
S
VM
model
in
pa
r
ti
c
ular
ha
d
the
be
s
t
a
c
c
ur
a
c
y
of
98.
47
%
.
Noo
r
e
t
a
l
.
[
38]
pr
e
s
e
nted
P
a
R
S
E
L
,
a
nove
l
s
tac
king
model.
T
he
ba
s
e
laye
r
is
c
omp
r
is
e
d
of
t
he
r
idge
c
las
s
if
ier
(
R
C
)
,
the
pa
s
s
ive
-
a
ggr
e
s
s
ive
c
las
s
if
ier
(
P
AC
)
,
XG
B
oos
t,
a
n
d
the
s
tocha
s
ti
c
gr
a
dient
de
s
c
e
nt
c
las
s
if
ier
(
S
GD
C
)
.
On
the
meta
laye
r
,
L
ogit
B
oos
t
wa
s
e
mpl
oye
d.
R
F
E
,
li
ne
a
r
dis
c
r
im
inant
a
na
lys
is
(
L
DA
)
,
a
nd
f
a
c
tor
a
na
lys
is
(
F
A)
we
r
e
the
thr
e
e
methods
e
mpl
oye
d
to
r
e
duc
e
dim
e
ns
ionalit
y.
T
o
a
ddr
e
s
s
the
im
ba
lanc
e
d
na
tur
e
of
the
da
tas
e
t,
e
ight
b
a
lanc
ing
pr
oc
e
dur
e
s
we
r
e
a
ppli
e
d.
T
he
outcome
s
s
howe
d
that
P
a
R
S
E
L
outper
f
or
med
other
s
tand
-
a
lone
c
las
s
if
ier
s
,
with
a
n
a
c
c
ur
a
c
y
o
f
97
%
.
J
a
f
a
r
a
nd
L
e
e
[
39]
de
ve
loped
a
n
a
utom
a
ti
c
M
L
s
ys
tem
c
a
ll
e
d
HypG
B
.
I
t
us
e
d
the
GB
c
las
s
if
ier
f
or
c
las
s
if
ica
ti
on.
T
o
c
hoos
e
the
be
s
t
f
e
a
tur
e
s
ubs
e
t
a
nd
e
li
mi
na
te
dupli
c
a
te
a
nd
nois
y
a
t
tr
ibut
e
s
,
a
tr
a
dit
ional
L
ASS
O
tec
hnique
wa
s
e
mpl
oye
d
.
T
h
e
GB
model
wa
s
e
nha
nc
e
d
us
ing
the
mos
t
r
e
c
e
nt
v
e
r
s
ion
of
the
Hype
r
Opt
opti
mi
z
a
ti
on
f
r
a
mew
or
k
.
E
xpe
r
i
menta
l
r
e
s
ult
s
f
o
r
the
C
leve
land
HD
a
nd
Ka
ggle
he
a
r
t
f
a
il
ur
e
da
tas
e
ts
s
how
that
HypG
B
w
a
s
a
ble
to
s
uc
c
e
s
s
f
u
ll
y
identif
y
f
e
a
tur
e
s
a
nd
obtain
outs
tanding
c
las
s
if
ica
ti
on
a
c
c
ur
a
c
ies
of
97
.
32
a
nd
97
.
72%
.
C
ha
ndr
a
s
e
kha
r
a
nd
P
e
dda
kr
is
hna
[
40
]
tes
ted
s
ix
M
L
tec
hniques
c
o
mpr
is
ing
L
R
,
KN
N,
NB
,
R
F
,
GB
,
a
nd
Ada
B
oos
t,
us
ing
the
da
ta
f
r
om
C
leve
land
a
nd
I
E
E
E
Da
tapor
t.
T
o
incr
e
a
s
e
model
c
or
r
e
c
tnes
s
,
the
s
tudy
e
mpl
oye
d
Gr
ids
e
a
r
c
hC
V
a
long
with
f
ive
-
f
old
c
r
os
s
-
va
li
da
ti
on.
I
n
the
C
leve
land
da
tas
e
t,
L
R
pe
r
f
or
med
be
tt
e
r
than
the
other
a
lgor
i
t
hms
with
90.
16%
a
c
c
ur
a
c
y,
whe
r
e
a
s
Ada
B
oos
t
pe
r
f
or
med
be
tt
e
r
with
90%
a
c
c
ur
a
c
y
in
the
I
E
E
E
Da
tapor
t
da
tas
e
t.
T
he
a
c
c
ur
a
c
y
of
the
model
wa
s
f
ur
ther
r
a
is
e
d
to
93.
44%
a
nd
95
%
f
or
the
C
leve
land
a
nd
I
E
E
E
Da
tapor
t
da
tas
e
ts
,
c
or
r
e
s
pondingl
y,
by
int
e
gr
a
ti
n
g
a
ll
s
ix
a
ppr
oa
c
he
s
with
the
s
of
t
voti
ng
e
ns
e
mbl
e
c
las
s
if
ier
.
Hos
s
a
in
e
t
al
.
[
41]
e
mpl
oye
d
the
be
s
t
f
ir
s
t
s
e
a
r
c
h
a
long
with
a
f
e
a
tur
e
s
ubs
e
t
s
e
lec
ti
on
method
ba
s
e
d
on
c
or
r
e
lation
to
dis
c
ove
r
the
be
s
t
f
e
a
tur
e
s
in
the
da
ta.
T
wo
types
of
HD
da
tas
e
ts
one
with
a
ll
f
e
a
tur
e
s
a
nd
the
other
with
c
hos
e
n
f
e
a
tur
e
s
we
r
e
us
e
d
to
tes
t
nume
r
ous
M
L
a
ppr
oa
c
he
s
.
T
he
s
e
include
d
S
VM
,
L
R
,
KN
N,
NB
,
DT
,
R
F
,
a
nd
M
L
P
.
Among
thes
e
tec
hniques
,
R
F
us
ing
the
s
e
lec
ted
f
e
a
tur
e
s
de
mons
tr
a
ted
the
highes
t
a
c
c
ur
a
c
y
of
90%
.
J
a
wa
lkar
e
t
al
.
[
42]
pr
opos
e
d
a
n
M
L
-
ba
s
e
d
a
ppr
oa
c
h
f
or
identi
f
ying
HD
by
e
mpl
oying
a
l
os
s
-
opti
mi
z
e
d
de
c
is
ion
tr
e
e
-
ba
s
e
d
r
a
ndom
f
or
e
s
t
(
DT
R
F
)
c
las
s
if
ier
.
F
ur
ther
mor
e
,
the
DT
R
F
c
las
s
if
ier
wa
s
tr
a
ined
uti
li
z
ing
a
los
s
opti
mi
z
a
ti
on
te
c
hnique
c
a
ll
e
d
s
tocha
s
ti
c
gr
a
dient
boos
ti
ng
(
S
GB
)
.
Ac
c
or
ding
to
the
r
e
s
ult
s
,
the
s
ugge
s
ted
HD
P
-
DT
R
F
a
ppr
oa
c
h
o
btaine
d
a
96%
a
c
c
ur
a
c
y
r
a
te
on
publi
c
ly
a
va
il
a
ble
r
e
a
l
-
wor
ld
da
tas
e
ts
.
M
a
nikanda
n
e
t
al.
[
43
]
e
va
luate
d
a
nd
c
ontr
a
s
ted
the
r
e
s
ult
s
of
the
S
VM
,
L
R
,
a
nd
D
T
a
lgor
it
h
ms
both
in
c
onjunction
with
a
nd
without
us
ing
the
f
e
a
tur
e
s
e
lec
ti
on
a
ppr
oa
c
h
na
med
bor
uta
.
T
his
inves
ti
ga
ti
on
wa
s
c
onduc
ted
us
ing
the
C
leve
land
HD
da
tas
e
t.
I
t
wa
s
dis
c
ove
r
e
d
that
the
B
or
uta
a
lgor
it
hm
e
nha
nc
e
d
t
he
r
e
s
ult
s
of
the
a
lgo
r
it
hms
.
Among
a
ll
,
L
R
a
c
hieve
d
the
highes
t
a
c
c
ur
a
c
y
of
88.
52%
.
Als
hr
a
ideh
e
t
al
.
[
44
]
a
im
e
d
to
e
nha
nc
e
HD
pr
e
diction
us
ing
M
L
mod
e
ls
with
the
HD
da
tas
e
t
obtaine
d
f
r
om
the
J
or
da
n
U
niver
s
it
y
Hos
pit
a
l
(
J
UH
)
.
T
o
c
hoos
e
f
e
a
tur
e
s
,
s
e
ve
r
a
l
M
L
c
las
s
if
ier
s
,
c
ompr
is
ing
KN
N,
S
VM
,
NB
,
DT
,
a
nd
R
F
we
r
e
e
xa
mi
ne
d
us
ing
P
S
O.
T
he
f
indi
ngs
s
howe
d
that
S
VM
c
ombi
ne
d
with
P
S
O
s
howe
d
outs
tanding
pe
r
f
or
manc
e
,
indi
c
a
ti
ng
it
s
e
f
f
icie
nc
y
in
c
las
s
if
ying
pa
ti
e
nts
a
c
c
or
ding
to
their
HD
r
is
k,
r
e
a
c
hing
a
n
a
c
c
ur
a
c
y
o
f
94.
3
%
.
B
y
r
e
view
ing
the
r
e
leva
nt
li
ter
a
tur
e
,
it
is
c
lea
r
th
a
t
M
L
methods
a
id
in
the
e
a
r
ly
identif
ica
ti
on
of
HD
.
How
e
ve
r
,
thes
e
methods
a
ls
o
ha
ve
c
e
r
tain
dr
a
wba
c
ks
a
nd
pr
oblems
.
T
he
f
oll
owing
r
e
s
e
a
r
c
h
g
a
ps
we
r
e
identif
ied:
‒
S
ome
models
a
r
e
va
li
da
ted
with
jus
t
one
da
tas
e
t
.
‒
I
n
c
e
r
tain
c
a
s
e
s
,
the
s
a
mpl
e
s
ize
is
ve
r
y
s
mall
.
‒
S
ome
s
tudi
e
s
us
e
d
a
f
e
w
pe
r
f
o
r
manc
e
e
va
luation
metr
ics
to
a
s
s
e
s
s
their
models
.
‒
S
ome
s
tudi
e
s
ha
ve
not
c
omput
e
d
the
e
r
r
o
r
r
a
tes
in
pr
e
dict
ion
.
‒
S
ome
models
a
r
e
not
va
li
da
ted
us
ing
R
OC
c
ur
ve
.
‒
T
im
e
c
ompl
e
xit
y
is
s
ometim
e
s
ove
r
looked
by
r
e
s
e
a
r
c
he
r
s
.
‒
O
ve
r
f
it
ti
ng
ha
s
be
e
n
identi
f
ied
in
s
ome
s
tudi
e
s
.
‒
C
e
r
tain
a
r
ti
c
les
only
c
ompar
e
d
the
pe
r
f
o
r
manc
e
of
2
M
L
c
las
s
if
ier
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14
,
No.
4
,
Augus
t
2025
:
284
9
-
2863
2854
T
a
ble
1
s
howc
a
s
e
s
a
va
r
iety
of
M
L
a
lgor
it
h
ms
uti
li
z
e
d
by
r
e
s
e
a
r
c
he
r
s
in
de
tec
ti
ng
HD
.
T
a
ble
1.
M
L
a
lgor
it
hms
f
o
r
HD
p
r
e
diction
a
long
with
their
r
e
f
e
r
e
nc
e
c
ount
M
L
a
lg
or
it
hm
R
e
f
e
r
e
nc
e
s
R
e
f
. c
ount
LR
[
8]
, [
10]
,
[
13]
–
[
15
]
,
[
18]
,
[
20]
,
[
22]
–
[
24]
,
[
26
]
,
[
28]
,
[
31]
–
[
33
]
,
[
35]
, [
40]
,
[
41]
, [
43
]
19
KNN
[
8]
,
[
13]
–
[
16]
,
[
21
]
–
[
23]
,
[
26]
–
[
28]
,
[
32]
, [
33
]
, [
35]
,
[
36]
, [
44
]
17
ANN
[
8]
, [
13]
,
[
14]
, [
34
]
, [
37]
5
S
V
M
[
8]
,
[
10]
,
[
13]
–
[
15
]
,
[
17]
,
[
19]
, [
20]
,
[
22]
, [
24
]
,
[
26]
–
[
28]
,
[
31
]
,
[
33]
,
[
35]
,
[
37]
, [
41
]
, [
43]
,
[
44]
22
NB
[
8]
,
[
10]
,
[
11]
,
[
13
]
–
[
17]
,
[
19]
, [
20]
,
[
31]
, [
33
]
,
[
35]
,
[
40]
,
[
41
]
,
[
44]
17
DT
[
8]
,
[
10]
,
[
13]
,
[
14
]
,
[
16]
–
[
19]
,
[
21]
–
[
24]
,
[
27
]
,
[
31]
,
[
33]
,
[
35
]
–
[
37]
,
[
41]
–
[
44]
22
RF
[
8]
–
[
10]
,
[
13]
,
[
15
]
–
[
18]
,
[
20]
–
[
25]
,
[
27]
,
[
29
]
–
[
31]
,
[
33]
,
[
35
]
,
[
36]
,
[
40]
–
[
42]
,
[
44
]
25
GB
[
21]
,
[
24]
,
[
27]
,
[
35
]
,
[
39]
,
[
40]
6
X
G
B
oos
t
[
12]
,
[
26]
,
[
28]
,
[
35
]
, [
36]
,
[
38]
6
M
L
P
[
13]
,
[
19]
,
[
23]
,
[
35
]
,
[
41]
5
A
da
B
oos
t
[
20]
,
[
21]
,
[2
3]
,
[
24
]
,
[
36]
,
[
37]
,
[
40]
7
C
N
N
[
34]
1
ET
[
36]
1
S
G
B
[
42]
1
3.
M
AT
E
R
I
AL
S
AN
D
M
E
T
HO
D
T
he
r
e
s
e
a
r
c
h
methodology
e
mpl
oye
d
f
or
c
onduc
ti
ng
the
r
e
s
e
a
r
c
h
is
outl
ined
in
thi
s
s
e
c
ti
on.
F
igu
r
e
1
de
picts
the
s
e
ve
r
a
l
pr
oc
e
s
s
e
s
a
s
s
oc
iat
e
d
with
pr
e
dicting
HD
,
including
:
i)
s
e
lec
ti
ng
the
da
tas
e
t
to
be
us
e
d,
ii
)
pr
oc
e
s
s
ing
da
ta,
ii
i
)
the
c
r
os
s
-
va
li
da
ti
on
,
iv)
c
hoos
ing
M
L
methods
,
v
)
pe
r
f
o
r
mi
ng
pr
e
dicti
ons
,
a
nd
v)
e
va
luating
pe
r
f
or
manc
e
.
T
he
ne
xt
s
ub
-
s
e
c
ti
on
g
oe
s
int
o
f
ur
the
r
de
pth
a
bout
thes
e
s
tage
s
.
F
igur
e
1.
F
low
of
s
teps
invol
ve
d
in
HD
p
r
e
diction
3.
1.
Dat
as
e
t
Da
ta
is
of
the
utm
os
t
im
po
r
tanc
e
f
or
M
L
to
pr
odu
c
e
a
c
c
ur
a
te
a
nd
r
e
li
a
ble
r
e
s
ult
s
.
T
his
a
na
lys
is
us
e
d
two
ope
nly
a
c
c
e
s
s
ibl
e
HD
da
ta
s
e
ts
f
r
om
Ka
ggle:
the
C
leve
land
a
nd
S
tatlog
(
He
a
r
t)
[
45]
,
[
46]
.
T
he
s
e
da
tas
e
ts
we
r
e
s
e
lec
ted
be
c
a
us
e
r
e
s
e
a
r
c
he
r
s
f
r
e
que
ntl
y
u
s
e
them
to
a
s
s
e
s
s
the
pe
r
f
or
manc
e
o
f
thei
r
HD
p
r
e
diction
methods
.
T
he
C
leve
land
da
ta
ha
s
303
c
a
s
e
s
,
whe
r
e
a
s
the
S
tatlog
da
tas
e
t
include
s
270
oc
c
u
r
r
e
nc
e
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
P
e
r
for
manc
e
analys
is
and
c
ompar
i
s
on
of
mac
hine
lear
ning
algor
it
hms
for
pr
e
dicting
he
ar
t
(
N
e
ha
B
h
adu)
2855
E
a
c
h
da
tas
e
t
ha
s
14
c
ha
r
a
c
ter
is
ti
c
s
,
with
the
ini
ti
a
l
13
in
a
f
e
a
tur
e
type
a
nd
the
las
t
in
the
tar
ge
t
type.
T
a
ble
2
de
s
c
r
ibes
the
pr
ope
r
ti
e
s
of
both
da
tas
e
ts
,
whic
h
inc
lude
the
s
a
me
kind
a
nd
a
mount
of
f
e
a
tur
e
s
.
T
a
ble
2.
F
e
a
tur
e
s
inf
o
r
mation
o
f
the
C
leve
land
a
nd
the
S
tatlog
HD
da
tas
e
t
S
.N
o.
F
e
a
tu
r
e
na
me
T
ype
of
da
ta
E
xpl
a
na
ti
on
D
oma
in
of
t
a
r
ge
t
a
tt
r
ib
ut
e
1.
A
ge
N
ume
r
ic
A
ge
(
ye
a
r
s
)
29
-
77
2.
S
e
x
C
a
te
gor
ic
a
l
G
e
nde
r
0:
F
e
ma
le
1:
M
a
le
3.
Cp
C
a
te
gor
ic
a
l
N
a
tu
r
e
of
P
a
in
i
n t
he
C
he
s
t
1:
T
ypi
c
a
l
a
ngi
na
2:
A
ty
pi
c
a
l
a
ngi
na
3:
N
on
-
a
ngi
na
l
pa
in
4:
A
s
ympt
oma
ti
c
4.
T
r
e
s
tb
ps
N
ume
r
ic
R
e
s
ti
ng blood pr
e
s
s
ur
e
(
mm
hg)
94
-
200
5.
C
hol
N
ume
r
ic
S
e
r
um c
hol
e
s
te
r
ol
(
mg/
dL
)
126
-
564
6.
F
bs
C
a
te
gor
ic
a
l
F
a
s
ti
ng blood s
uga
r
>
120 mg/
dL
0:
F
a
ls
e
1:
T
r
ue
7.
R
e
s
te
c
g
C
a
te
gor
ic
a
l
R
e
s
ti
ng e
le
c
tr
oc
a
r
di
ogr
a
m f
in
di
ngs
0:
N
or
ma
l
1:
S
T
-
T
w
a
ve
a
bnor
ma
li
ty
2:
P
r
oba
bl
e
8.
T
ha
la
c
h
N
ume
r
ic
M
a
xi
ma
l
he
a
r
t
r
a
te
71
-
202
9.
E
xa
ng
C
a
te
gor
ic
a
l
E
xe
r
c
is
e
-
r
e
la
te
d a
ngi
na
0:
N
o
1:
Y
e
s
10.
O
ld
pe
a
k
N
ume
r
ic
E
xe
r
c
is
e
-
in
duc
e
d S
T
de
pr
e
s
s
io
n i
n c
ompa
r
is
on t
o r
e
s
t
0
-
6.2
11.
S
lo
pe
C
a
te
gor
ic
a
l
S
lo
pe
of
pe
a
k e
xe
r
c
is
e
S
T
s
e
gme
nt
1:
U
ps
lo
pi
ng
2:
F
la
t
3:
D
ow
ns
lo
pi
ng
12.
Ca
C
a
te
gor
ic
a
l
C
ount
of
ma
jo
r
ve
s
s
e
ls
1
-
4
13.
T
ha
l
C
a
te
gor
ic
a
l
T
he
T
h
a
ll
iu
m i
ma
gi
ng
3:
N
or
ma
l
6:
F
ix
e
d
7:
R
e
ve
r
s
ib
le
de
f
e
c
t
14.
T
a
r
ge
t
C
a
te
gor
ic
a
l
O
ut
put
va
r
ia
bl
e
0:
H
D
i
s
a
bs
e
nt
1:
H
D
i
s
pr
e
s
e
nt
3.
2.
Dat
a
p
re
-
p
r
oc
e
s
s
in
g
T
he
unpr
oc
e
s
s
e
d
da
ta
mus
t
f
i
r
s
t
be
pr
e
-
pr
oc
e
s
s
e
d
be
f
or
e
be
ing
us
e
d
with
the
M
L
a
lgor
it
hm.
P
r
e
-
pr
oc
e
s
s
ing
tr
a
ns
f
or
ms
les
s
s
igni
f
ica
nt
inf
or
mation
int
o
mo
r
e
r
e
leva
nt
da
ta.
T
he
r
e
a
r
e
s
e
ve
r
a
l
s
teps
invol
ve
d
in
thi
s
p
r
oc
e
s
s
,
s
uc
h
a
s
ga
ther
ing
da
t
a
f
r
om
a
da
taba
s
e
,
s
e
lec
ti
ng
ne
c
e
s
s
a
r
y
inf
or
mation
,
p
r
e
pa
r
ing
the
c
hos
e
n
da
ta,
the
s
a
mpl
ing
pr
oc
e
s
s
,
a
nd
da
ta
c
onve
r
s
ion.
De
a
li
ng
with
m
is
s
ing
number
s
a
nd
e
li
mi
na
ti
ng
nois
e
a
nd
outl
ier
s
f
r
om
the
da
ta
may
be
ne
c
e
s
s
a
r
y
to
a
c
hieve
thi
s
.
I
t
may
be
c
ha
ll
e
nging
f
o
r
M
L
a
lgor
it
hms
to
pr
oc
e
s
s
incoming
da
ta
if
ther
e
a
r
e
mi
s
s
ing
va
lue
s
.
C
ons
e
que
ntl
y,
be
f
or
e
us
ing
a
ny
a
ppr
oa
c
h,
the
da
ta
mus
t
be
c
onve
r
ted
int
o
a
s
tr
uc
tu
r
e
d
f
o
r
mat.
Da
ta
pr
e
pa
r
a
ti
on
is
c
omm
only
r
e
f
e
r
r
e
d
to
a
s
e
xtr
a
c
t
,
tr
a
ns
f
o
r
m,
a
nd
load
(
ETL
)
.
T
he
dis
tr
i
buti
on
of
da
ta
is
c
r
uc
ial
f
o
r
pr
e
di
c
ti
ve
modeling.
T
a
ble
3
,
s
hows
the
e
xpe
c
ted
d
is
tr
i
buti
on
of
a
tt
r
ibut
e
c
las
s
e
s
f
or
the
two
da
tas
e
ts
us
e
d.
T
his
d
e
mons
tr
a
tes
that
the
dis
tr
ibut
ion
of
the
tar
ge
t
a
tt
r
i
bute
f
or
both
of
thes
e
da
tas
e
ts
is
e
qua
l,
whic
h
he
lps
a
v
o
id
the
ove
r
f
it
ti
ng
is
s
ue
.
I
n
both
da
tas
e
ts
,
ther
e
we
r
e
no
mi
s
s
ing
va
lues
f
ound.
F
or
the
tar
ge
t
c
las
s
,
ther
e
a
r
e
f
ive
c
las
s
labe
ls
in
the
or
igi
na
l
C
leve
land
da
tas
e
t,
e
a
c
h
with
a
n
int
e
ge
r
va
lue
be
twe
e
n
0
a
nd
4
.
T
he
C
leve
land
da
tas
e
t
mainly
a
tt
e
mpt
e
d
to
dis
c
r
im
inate
be
t
we
e
n
the
e
xis
tenc
e
of
HD
with
a
tar
ge
t
pos
s
e
s
s
ing
v
a
lues
r
a
nging
f
r
om
1,
2
,
3,
a
nd
4
,
a
nd
a
n
a
bs
e
nc
e
of
HD
with
a
va
lue
of
0.
Ac
c
or
ding
to
the
r
e
s
e
a
r
c
he
r
s
,
the
f
ive
c
las
s
f
e
a
tur
e
s
of
the
tar
ge
t
a
tt
r
ibut
e
f
or
th
is
da
tas
e
t
c
a
n
be
s
im
pli
f
ied
to
two
c
las
s
e
s
i.
e
.
0
a
nd
1.
As
a
r
e
s
ult
,
the
mul
ti
c
las
s
number
s
f
or
it
s
tar
ge
t
a
t
tr
ib
ute
we
r
e
tr
a
ns
f
or
med
int
o
binar
y
number
s
by
s
e
tt
ing
e
ve
r
y
number
f
r
om
2
to
4
to
1.
T
hus
,
the
f
inal
da
tas
e
t's
diagnos
ti
c
va
lues
a
r
e
s
im
ply
0
a
nd
1,
whe
r
e
0
de
notes
the
a
bs
e
nc
e
o
f
HD
a
nd
1
de
notes
it
s
pr
e
s
e
nc
e
.
F
ur
ther
mor
e
,
a
f
il
ter
ing
method
known
a
s
c
las
s
ba
lanc
e
r
wa
s
us
e
d
to
e
ns
ur
e
e
ve
r
y
ins
tanc
e
in
the
da
tas
e
t
got
e
qua
l
w
e
ight
.
T
a
ble
3.
Dis
tr
ibut
ion
of
da
ta
in
both
da
tas
e
ts
D
a
ta
s
e
t
(
I
ns
ta
nc
e
s
)
P
a
ti
e
nt
s
ha
vi
ng H
D
(
%
)
H
e
a
lt
hy
pe
r
s
ons
(
%
)
C
le
ve
la
nd (
303)
45.8
54.1
S
ta
tl
og (
270)
44.4
55.5
3.
3.
Cr
os
s
-
vali
d
at
ion
C
r
os
s
-
va
li
da
ti
on
r
e
duc
e
s
ove
r
f
it
ti
ng
by
e
va
luating
a
n
M
L
model's
pe
r
f
or
manc
e
us
ing
uns
e
e
n
da
ta.
K
-
f
old
c
r
os
s
-
va
li
da
ti
on
s
e
pa
r
a
te
s
da
ta
int
o
k
e
qua
l
-
s
ize
d
f
olds
(
in
thi
s
c
a
s
e
,
k=
10)
a
nd
us
e
s
e
ve
r
y
s
ingl
e
f
old
a
s
a
va
li
da
ti
on
s
e
t.
T
he
model
is
tr
a
ined
a
nd
e
va
luate
d
k
ti
mes
,
a
nd
a
n
unbias
e
d
e
s
ti
mate
is
p
r
od
uc
e
d
by
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14
,
No.
4
,
Augus
t
2025
:
284
9
-
2863
2856
a
ve
r
a
ging
the
pe
r
f
o
r
manc
e
ove
r
a
ll
f
o
lds
.
T
his
wor
k
s
pli
ts
the
da
tas
e
t
int
o
s
e
ts
f
or
tr
a
ini
ng
a
nd
tes
ti
ng
u
s
ing
a
tenf
old
c
r
os
s
-
va
li
da
ti
on
tec
hnique.
3.
4.
S
e
lec
t
ion
of
t
h
e
algorit
h
m
T
h
e
c
ho
ice
o
f
the
a
lg
or
i
th
m
d
e
pe
nd
s
o
n
t
he
da
tas
e
t
a
nd
p
r
e
di
c
t
io
n
ty
pe
.
T
his
s
tu
dy
us
e
s
R
e
f
.
c
ou
n
t
,
a
v
a
r
i
a
b
le
t
r
a
c
k
in
g
th
e
f
r
e
que
nc
y
o
f
th
e
a
l
go
r
it
hms
us
e
d
i
n
pr
e
v
io
us
s
t
ud
ies
,
t
o
s
e
le
c
t
s
u
i
tab
le
a
lg
o
r
it
h
ms
f
or
a
na
lys
is
.
T
h
is
s
ub
-
s
e
c
t
io
n
e
xa
mi
ne
s
a
lg
o
r
i
th
ms
wi
th
a
R
e
f
.
c
o
un
t
e
xc
e
e
d
in
g
6
f
r
om
T
a
b
le
1
a
n
d
d
is
c
us
s
e
s
the
m
.
‒
L
R
:
L
R
is
a
n
a
ppr
oa
c
h
to
s
upe
r
vis
e
d
lea
r
ning
that
c
a
n
be
uti
li
z
e
d
f
or
c
las
s
if
ica
ti
on
a
nd
r
e
gr
e
s
s
ion.
I
t
is
c
omm
o
nly
e
mpl
oye
d
in
bina
r
y
c
las
s
if
ica
ti
on
pr
o
blems
whe
r
e
the
outcome
va
r
iable
c
a
n
be
0
or
1
.
L
R
a
na
lys
e
s
the
c
onne
c
ti
on
be
twe
e
n
indepe
nde
nt
va
r
i
a
bles
a
nd
c
a
tegor
ize
s
them
int
o
dis
ti
nc
t
c
las
s
e
s
us
ing
the
logi
s
ti
c
f
unc
ti
on
,
of
ten
r
e
f
e
r
r
e
d
to
a
s
the
s
igm
o
id
f
unc
t
ion.
‒
S
VM
:
S
VM
is
a
r
obus
t
s
upe
r
vis
e
d
lea
r
ning
a
ppr
oa
c
h
that
pe
r
f
or
ms
we
ll
in
both
r
e
gr
e
s
s
ion
a
nd
c
las
s
if
ica
ti
on
a
ppli
c
a
ti
ons
.
T
he
pr
im
a
r
y
objec
ti
ve
is
to
identif
y
the
opti
mu
m
hype
r
plane
in
a
s
pa
c
e
with
N
dim
e
ns
ions
that
c
a
n
e
f
f
icie
ntl
y
divi
de
da
ta
po
int
s
int
o
d
if
f
e
r
e
nt
c
las
s
e
s
.
T
he
hype
r
plane
's
pur
pos
e
i
s
to
maximi
z
e
the
dis
tanc
e
a
mongs
t
point
s
that
a
r
e
c
los
e
s
t
in
e
a
c
h
c
las
s
.
‒
NB
:
NB
c
las
s
if
ier
s
a
r
e
pr
oba
bil
is
ti
c
c
las
s
if
ier
s
that
us
e
B
a
ye
s
'
theor
e
m.
I
t
is
a
s
s
umed
that
the
pr
e
s
e
nc
e
of
a
s
pe
c
if
ic
a
tt
r
ibut
e
in
the
c
las
s
doe
s
not
a
f
f
e
c
t
t
he
pr
e
s
e
nc
e
of
a
nother
a
tt
r
ibu
te
in
a
s
im
il
a
r
c
las
s
.
I
t
c
omput
e
s
the
li
ke
li
hood
of
a
n
input
r
e
lating
to
a
gi
ve
n
c
las
s
,
a
s
s
umi
ng
f
e
a
tur
e
indepe
nde
nc
e
[
47]
.
‒
KN
N:
KN
N
is
a
non
-
pa
r
a
metr
ic
a
ppr
oa
c
h
to
s
upe
r
vis
e
d
lea
r
ning
that
c
a
n
be
e
mpl
oye
d
f
or
b
oth
c
las
s
if
ica
ti
on
a
nd
r
e
gr
e
s
s
ion
pr
oblems
.
I
t
wo
r
ks
b
y
c
ompar
ing
da
ta
point
s
to
f
ind
s
im
il
a
r
it
ies
.
T
he
l
a
be
l
a
s
s
oc
iate
d
with
ne
w
da
t
a
is
pr
e
dicte
d
by
e
va
luatin
g
the
labe
li
ng
of
the
K
c
los
e
s
t
ne
ighbor
s
in
the
tr
a
ini
ng
s
e
t.
T
he
dis
tanc
e
a
mongs
t
da
ta
point
s
is
de
ter
mi
ne
d
uti
li
z
ing
E
uc
li
de
a
n,
M
a
n
ha
tt
a
n,
o
r
M
ink
ows
ki
dis
tanc
e
s
.
‒
DT
:
D
T
is
a
non
-
pa
r
a
metr
ic
s
upe
r
vis
e
d
lea
r
ning
method
us
e
d
f
o
r
r
e
gr
e
s
s
ion
a
nd
c
las
s
if
ica
ti
on.
I
t
u
s
e
s
a
hier
a
r
c
hica
l
tr
e
e
s
tr
uc
tur
e
with
lea
f
node
s
,
int
e
r
na
l
node
s
,
br
a
nc
he
s
,
a
nd
a
r
oot
node
.
De
c
is
ions
a
r
e
made
us
ing
br
a
nc
he
s
,
int
e
r
na
l
node
s
de
s
c
r
ibe
da
tas
e
t
pr
ope
r
ti
e
s
a
nd
lea
f
node
s
dis
play
de
s
ir
e
d
outcome
s
.
DT
us
e
s
a
gr
e
e
dy
s
e
a
r
c
h
a
nd
divi
de
-
a
nd
-
c
onqu
e
r
s
tr
a
tegy
to
f
ind
opti
mal
s
pli
t
loca
ti
ons
,
r
e
pe
a
ti
ng
the
t
op
-
down
divi
ding
p
r
oc
e
s
s
unti
l
mos
t
r
e
c
or
ds
a
r
e
c
a
tegor
ize
d
unde
r
s
pe
c
if
ic
c
las
s
labe
ls
.
‒
R
F
:
R
F
is
a
n
M
L
s
tr
a
tegy
us
e
d
f
or
r
e
gr
e
s
s
ion
a
nd
c
las
s
if
ica
ti
on.
I
t
c
r
e
a
tes
DT
dur
ing
tr
a
ini
ng
,
e
a
c
h
e
va
luating
a
r
a
ndom
s
a
mpl
e
of
f
e
a
tur
e
s
.
T
his
r
a
nd
omi
z
a
t
ion
pr
e
ve
nts
ove
r
f
it
ti
ng
a
nd
im
p
r
ove
s
pr
e
di
c
ti
on
a
c
c
ur
a
c
y.
Dur
ing
pr
e
diction
,
the
a
lgo
r
it
hm
c
ombi
n
e
s
the
output
s
of
a
ll
tr
e
e
s
th
r
ough
vo
ti
ng
or
a
ve
r
a
g
ing,
r
e
pe
a
ti
ng
r
e
c
ur
s
ively
unti
l
mos
t
r
e
c
or
ds
a
r
e
c
a
tegor
ize
d
unde
r
s
pe
c
if
ic
c
las
s
labe
ls
[4
8]
.
‒
Ada
B
oos
t:
Ada
B
oo
s
t
invol
ve
s
c
ombi
ning
s
e
ve
r
a
l
we
a
k
c
las
s
if
ier
s
int
o
one
e
ns
e
mbl
e
method
to
p
r
o
duc
e
a
s
tr
onge
r
c
las
s
if
ier
.
T
his
a
lgor
it
h
m
tr
a
ins
a
nd
de
ploys
a
s
e
que
nc
e
of
tr
e
e
s
,
im
pleme
nti
ng
boos
ti
ng.
E
a
c
h
c
las
s
if
ier
im
pr
ove
s
the
c
las
s
if
ica
ti
on
of
s
a
mpl
e
s
i
nc
or
r
e
c
tl
y
c
las
s
if
ied
by
it
s
pr
e
de
c
e
s
s
or
.
B
y
c
ombi
ning
we
a
k
c
las
s
if
ier
s
,
boos
ti
ng
e
f
f
e
c
ti
ve
ly
ge
ne
r
a
tes
a
powe
r
f
ul
c
las
s
if
ier
that
c
a
tegor
ize
s
r
e
c
or
ds
unde
r
s
pe
c
if
ic
c
las
s
labe
ls
[
49]
.
3.
5.
P
r
e
d
ic
t
ion
Ada
B
oos
t,
DT
,
R
F
,
KN
N
,
NB
,
L
R
,
a
nd
S
VM
a
r
e
the
M
L
a
lgor
it
hms
s
e
lec
ted
f
r
om
T
a
ble
1.
P
r
e
dictions
a
r
e
ge
ne
r
a
ted
us
ing
thes
e
c
las
s
if
ier
s
o
n
both
da
tas
e
ts
.
T
he
tar
ge
t
va
r
iable
with
va
lue
0
i
ndica
tes
a
n
a
bs
e
nc
e
o
f
HD
a
nd
va
lue
1
indi
c
a
tes
it
s
pr
e
s
e
nc
e
.
E
a
c
h
c
las
s
if
ier
's
e
f
f
ica
c
y
is
then
e
va
luate
d
us
in
g
s
e
ve
r
a
l
pe
r
f
or
manc
e
metr
ics
.
3.
6.
P
e
r
f
or
m
an
c
e
e
valu
at
io
n
T
o
de
ter
mi
ne
how
e
f
f
e
c
ti
ve
ly
a
model
ope
r
a
te
s
,
it
is
ne
c
e
s
s
a
r
y
to
e
mpl
oy
s
e
ve
r
a
l
e
va
luation
s
tanda
r
ds
tha
t
pr
ov
ide
a
c
ompr
e
he
ns
ive
pictu
r
e
of
it
s
pe
r
f
o
r
manc
e
.
T
he
e
f
f
e
c
ti
ve
ne
s
s
of
the
c
hos
e
n
c
l
a
s
s
if
ier
s
is
e
va
luate
d
us
ing
s
e
ve
r
a
l
e
va
luation
mea
s
ur
e
s
,
c
ompr
is
ing
M
C
C
,
Ka
ppa
va
lue,
F
-
mea
s
ur
e
,
R
OC
a
r
e
a
,
a
c
c
ur
a
c
y,
pr
e
c
is
ion,
a
nd
r
e
c
a
ll
.
T
he
metr
ics
a
r
e
c
omput
e
d
ut
il
izing
the
c
onf
us
ion
matr
ix
a
s
a
ba
s
e
.
T
he
c
onf
us
ion
matr
ix
in
T
a
ble
4
s
hows
both
the
a
c
tual
a
s
we
ll
a
s
pr
e
dicte
d
c
las
s
if
ica
ti
ons
ge
ne
r
a
ted
by
a
two
-
c
las
s
c
las
s
if
ier
.
T
his
matr
ix
pr
ovides
ins
ight
s
int
o
the
pe
r
f
or
manc
e
o
f
c
las
s
if
ica
ti
on
s
ys
t
e
ms
by
inves
ti
ga
ti
ng
the
da
ta
it
c
ontains
.
T
a
ble
4.
T
he
c
on
f
us
ion
matr
ix
P
r
e
di
c
te
d H
D
pa
ti
e
nt
s
P
r
e
di
c
te
d he
a
lt
hy i
ndi
vi
dua
ls
A
c
tu
a
l
H
D
pa
ti
e
nt
s
T
r
ue
pos
it
iv
e
(
T
P
)
F
a
ls
e
ne
ga
ti
ve
(
F
N
)
A
c
tu
a
l
he
a
lt
hy i
ndi
vi
dua
ls
F
a
ls
e
pos
it
iv
e
(
F
P
)
T
r
ue
ne
ga
ti
ve
(
T
N
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
P
e
r
for
manc
e
analys
is
and
c
ompar
i
s
on
of
mac
hine
lear
ning
algor
it
hms
for
pr
e
dicting
he
ar
t
(
N
e
ha
B
h
adu)
2857
He
r
e
,
T
P
de
notes
the
tot
a
l
number
of
c
a
s
e
s
a
c
c
ur
a
tely
identif
ied
with
HD
.
F
N
s
igni
f
ies
the
tot
a
l
number
o
f
indi
viduals
ha
ving
HD
who
a
r
e
incor
r
e
c
tl
y
c
a
tegor
ize
d
a
s
he
a
lt
hy.
T
N
s
igni
f
ies
the
n
umber
o
f
a
c
c
ur
a
tely
c
las
s
if
ied
he
a
lt
hy
pa
ti
e
nts
.
F
inally,
F
P
s
ign
if
ies
the
number
o
f
he
a
lt
hy
ins
tanc
e
s
that
a
r
e
incor
r
e
c
tl
y
ident
if
ied
with
HD
.
T
a
ble
5
pr
ovi
de
s
a
n
ove
r
view
of
the
e
va
luation
metr
ics
a
nd
their
mathe
matica
l
f
or
mu
las
[
50
]
.
T
he
s
e
f
or
mul
a
s
a
r
e
u
s
e
f
ul
f
or
mea
s
ur
ing
the
pe
r
f
or
manc
e
o
f
M
L
a
lgor
i
thm
s
in
pr
e
dicting
HD
.
T
a
ble
5.
P
e
r
f
o
r
manc
e
metr
ics
a
nd
their
mathe
mati
c
a
l
f
or
mul
a
P
e
r
f
or
ma
nc
e
m
e
tr
ic
F
or
mul
a
D
e
s
c
r
ip
ti
on
A
c
c
ur
a
c
y
=
(
+
)
(
+
+
+
)
I
t
r
e
pr
e
s
e
nt
s
th
e
pr
opor
ti
on
o
f
a
c
c
ur
a
te
pr
e
di
c
ti
ons
a
mongs
t
a
ll
pr
e
di
c
ti
ons
ma
de
.
P
r
e
c
is
io
n
=
+
I
t
me
a
s
ur
e
s
t
he
a
c
c
ur
a
c
y of
pos
it
iv
e
pr
e
di
c
ti
ons
.
R
e
c
a
ll
or
S
e
ns
it
iv
it
y
=
+
T
he
a
c
c
ur
a
c
y
of
th
e
mode
l
in
id
e
nt
if
yi
ng
pos
it
iv
e
c
a
s
e
s
a
mong
a
ll
of
th
e
a
c
tu
a
l
pos
it
iv
e
in
s
ta
nc
e
s
in
th
e
da
ta
s
e
t.
S
pe
c
if
ic
it
y
=
+
T
he
a
c
c
ur
a
c
y
of
th
e
mode
l
in
id
e
nt
if
yi
ng
ne
ga
ti
ve
c
a
s
e
s
a
mong
a
ll
of
th
e
a
c
tu
a
l
ne
ga
ti
v
e
in
s
ta
nc
e
s
in
th
e
da
ta
s
e
t.
FP
r
a
te
=
+
I
t
r
e
f
le
c
ts
th
e
numbe
r
o
f
c
a
s
e
s
in
th
e
da
ta
s
e
t
th
a
t
a
r
e
in
c
or
r
e
c
tl
y
c
a
te
gor
iz
e
d
a
s
pos
it
iv
e
w
he
n
th
e
y
a
r
e
ne
ga
ti
ve
.
F
-
me
a
s
ur
e
−
=
2
×
×
+
I
t
is
a
me
a
s
ur
e
of
s
ta
ti
s
ti
c
a
l
s
ig
ni
f
ic
a
nc
e
th
a
t
us
e
s
a
w
e
ig
ht
e
d a
ve
r
a
ge
t
o c
ombi
ne
r
e
c
a
ll
a
nd pr
e
c
is
io
n.
M
C
C
(
×
)
−
(
×
)
√
(
+
)
×
(
+
)
×
(
+
)
×
(
+
)
I
t
me
a
s
ur
e
s
th
e
pr
e
di
c
ti
ve
c
a
p
a
c
it
y
of
a
c
l
a
s
s
if
ie
r
a
nd i
s
r
e
pr
e
s
e
nt
e
d by va
lu
e
s
be
tw
e
e
n
-
1 a
nd +
1.
K
a
ppa
s
ta
ti
s
ti
c
2
×
(
(
×
)
−
(
×
)
)
(
+
)
×
(
+
)
+
(
+
)
×
(
+
)
I
t
is
a
me
a
s
ur
e
th
a
t
c
ompa
r
e
s
th
e
obs
e
r
ve
d
a
c
c
ur
a
c
y
to
th
e
e
xpe
c
te
d
a
c
c
ur
a
c
y,
w
hi
c
h
i
s
ba
s
e
d
on r
a
ndom c
ha
nc
e
.
AUC
1
2
(
+
+
+
)
I
t
gr
a
phi
c
a
ll
y
de
pi
c
ts
th
e
r
a
ti
o
of
tr
ue
pos
it
iv
e
s
vs
f
a
ls
e
pos
it
iv
e
s
,
w
it
h
th
e
r
e
gi
on
lo
c
a
te
d
unde
r
th
e
R
O
C
c
ur
ve
.
F
ur
ther
,
the
pe
r
f
or
manc
e
of
the
c
las
s
if
ier
s
is
c
he
c
ke
d
us
ing
e
r
r
or
r
a
te
a
na
lys
is
.
F
o
r
c
omput
ing
the
pr
e
diction
e
r
r
o
r
s
,
di
f
f
e
r
e
nt
e
r
r
o
r
r
a
tes
li
ke
mea
n
a
bs
olut
e
e
r
r
or
(
M
AE
)
,
r
e
lative
a
bs
olut
e
e
r
r
o
r
(
R
AE
)
,
r
oot
mea
n
s
qua
r
e
e
r
r
o
r
(
R
M
S
E
)
,
a
nd
r
oot
r
e
lative
s
qu
a
r
e
e
r
r
o
r
(
R
R
S
E
)
a
r
e
c
a
lcula
ted
[
51]
.
T
a
ble
6
outl
ines
dif
f
e
r
e
nt
e
r
r
or
r
a
tes
a
long
with
their
de
s
c
r
ipt
ion.
T
a
ble
6.
E
r
r
or
r
a
te
met
r
ics
a
nd
their
de
s
c
r
ipt
ion
E
r
r
or
r
a
te
me
tr
ic
D
e
s
c
r
ip
ti
on
M
A
E
I
t
is
de
f
in
e
d a
s
t
he
me
a
n of
a
da
ta
s
e
t'
s
e
s
ti
ma
te
d a
nd a
c
tu
a
l
va
lu
e
s
.
R
M
S
E
I
t
is
t
he
ba
s
ic
s
ta
ti
s
ti
c
a
l
me
tr
ic
c
a
lc
ul
a
te
d by t
a
ki
ng t
he
s
qua
r
e
r
oot
of
th
e
a
ve
r
a
ge
s
qua
r
e
d di
f
f
e
r
e
nc
e
be
tw
e
e
n e
xpe
c
te
d a
nd obs
e
r
ve
d t
a
r
ge
t
va
lu
e
s
in
a
da
ta
s
e
t.
R
A
E
I
t
is
a
r
a
ti
o
-
ba
s
e
d s
ta
ti
s
ti
c
u
s
e
d t
o e
va
lu
a
t
e
t
he
e
f
f
ic
ie
nc
y of
a
m
ode
l
in
ma
ki
ng pr
e
di
c
ti
ons
.
R
R
S
E
I
t
is
de
f
in
e
d a
s
t
he
s
qua
r
e
r
oot
of
a
pr
e
di
c
ti
ve
mode
l'
s
t
ot
a
l
s
qua
r
e
d e
r
r
or
s
nor
ma
li
z
e
d
by t
he
t
ot
a
l
s
qua
r
e
d e
r
r
or
s
of
t
he
ba
s
ic
mode
l.
3.
7.
S
o
f
t
war
e
u
s
e
d
T
he
W
E
KA
,
is
a
publi
c
ly
a
c
c
e
s
s
ibl
e
M
L
s
of
twa
r
e
a
ppli
c
a
ti
on.
T
his
plat
f
or
m
c
ompr
om
is
e
s
a
J
a
va
pr
ogr
a
mm
ing
langua
ge
API
that
inco
r
por
a
tes
p
r
e
-
buil
t
a
lgor
it
hms
f
r
om
a
c
e
r
tain
a
r
e
a
a
nd
m
a
ke
s
the
e
xe
c
uti
on
of
dif
f
e
r
e
nt
da
ta
a
na
lys
is
methods
s
im
p
ler
.
I
t
ha
s
f
e
a
t
ur
e
s
f
or
a
s
s
oc
iation,
r
ule
mi
ning,
c
l
us
ter
ing,
r
e
gr
e
s
s
ion,
c
las
s
if
ica
ti
on,
f
e
a
tur
e
s
e
lec
ti
on,
a
nd
da
ta
vis
ua
li
z
a
ti
on
[
52]
.
I
n
thi
s
s
tudy,
W
E
KA
v3
.
9.
6
wa
s
e
mpl
oye
d
on
a
n
11
th
ge
ne
r
a
ti
on
“
I
ntel(
R
)
C
or
e
(
T
M
)
i5
-
1135G7
@
2.
40
GH
z
2.
42
GH
z
”
C
P
U
with
R
AM
of
8.
00
GB
,
ope
r
a
ti
ng
on
a
64
-
bit
ve
r
s
ion
o
f
W
indow
s
11.
4.
RE
S
UL
T
S
AN
D
DI
S
CU
S
S
I
ON
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14
,
No.
4
,
Augus
t
2025
:
284
9
-
2863
2858
T
his
a
na
lyt
ica
l
s
tudy
int
r
oduc
e
d
two
R
Qs
to
thor
oughly
a
nd
im
pa
r
ti
a
ll
y
e
va
luate
the
M
L
a
lgor
it
hms
in
pr
e
dicting
HD
.
T
o
a
dd
r
e
s
s
R
Q1,
a
c
ompr
e
he
ns
ive
e
xa
mi
na
ti
on
of
va
r
ious
M
L
pr
e
dictive
a
lgor
it
hms
is
c
a
r
r
ied
out.
T
o
a
ns
we
r
R
Q2,
a
f
r
a
mew
or
k
is
p
r
e
s
e
nted
to
de
ter
mi
ne
the
mos
t
e
f
f
e
c
ti
ve
M
L
a
lgor
it
h
m
out
of
the
c
hos
e
n
a
lgor
it
hms
f
r
o
m
R
Q1.
F
ur
the
r
,
the
s
e
lec
ted
a
lgor
it
hms
a
r
e
a
ppli
e
d
to
two
identica
l
s
tr
uc
t
ur
e
d
HD
da
tas
e
ts
a
nd
then
e
a
c
h
a
lgor
it
hm
unde
r
goe
s
a
pe
r
f
o
r
manc
e
e
va
luation
pha
s
e
.
T
he
s
tudy
c
omp
a
r
e
d
the
pe
r
f
or
manc
e
of
mul
ti
p
le
c
las
s
if
ier
s
in
pr
e
dicting
HD
,
un
li
ke
s
ome
pr
e
vious
s
tudi
e
s
that
c
ompar
e
d
only
two
M
L
c
las
s
if
ier
s
.
F
o
r
e
xpe
r
im
e
ntation
,
two
ba
lanc
e
d
a
nd
identica
l
HD
da
tas
e
t
s
a
r
e
us
e
d,
whe
r
e
a
s
s
ome
e
a
r
li
e
r
s
tudi
e
s
ha
ve
us
e
d
only
one
da
tas
e
t
.
P
r
e
vious
r
e
s
e
a
r
c
h
r
e
ve
a
led
ove
r
f
it
ti
ng
is
s
ue
s
,
but
thi
s
s
tudy
uti
li
z
e
d
c
r
os
s
-
va
li
da
ti
on
a
nd
ba
lanc
e
d
da
tas
e
t
s
to
pr
e
ve
nt
thi
s
is
s
ue
.
S
ome
e
a
r
l
ier
s
tudi
e
s
us
e
d
f
e
w
pe
r
f
o
r
manc
e
metr
ics
f
or
e
va
luation
a
nd
did
not
c
omput
e
the
e
r
r
o
r
r
a
tes
.
W
hil
e
a
c
c
ur
a
c
y
is
c
r
uc
ial,
it
's
a
ls
o
vit
a
l
to
take
int
o
a
c
c
ount
other
c
r
uc
ial
metr
ics
int
o
c
ons
ider
a
ti
on
.
T
his
s
tudy
e
mp
loyed
s
e
ve
r
a
l
metr
ics
including
M
C
C
,
ka
ppa
va
lue,
F
-
mea
s
ur
e
,
R
OC
a
r
e
a
,
a
c
c
ur
a
c
y,
p
r
e
c
is
ion,
r
e
c
a
ll
,
a
nd
di
f
f
e
r
e
nt
e
r
r
o
r
r
a
tes
li
ke
M
AE
,
R
AE
,
R
M
S
E
,
a
nd
R
R
S
E
.
T
his
s
tudy
va
li
da
tes
models
u
s
ing
the
R
OC
c
ur
ve
,
c
ompar
ing
it
to
s
ome
pr
e
vious
s
tudi
e
s
that
did
not.
T
his
s
tudy
c
a
lc
ulate
s
the
ti
me
take
n
in
pr
e
diction,
unli
ke
pr
e
vious
s
tudi
e
s
whic
h
did
not
c
ons
ider
ti
me
c
ompl
e
xit
y.
T
he
pe
r
f
or
manc
e
e
va
luation
f
indi
ngs
f
or
the
M
L
c
las
s
i
f
ier
s
a
r
e
s
hown
in
T
a
bles
7
a
nd
8
on
the
r
e
s
pe
c
ti
ve
da
tas
e
ts
.
T
he
highl
ight
e
d
text
indi
c
a
tes
the
be
s
t
outcome
s
.
T
a
ble
7.
P
e
r
f
o
r
manc
e
a
na
lys
is
of
C
leve
land
da
tas
e
t
M
L
A
lg
or
it
hm
A
c
c
ur
a
c
y (
%
)
FP
r
a
te
P
r
e
c
is
io
n
R
e
c
a
ll
F
-
m
e
a
s
ur
e
M
C
C
R
O
C
a
r
e
a
K
a
ppa
v
a
lu
e
LR
88.7
0.150
0.888
0.888
0.888
0.738
0.956
0.7378
KNN
87.7
0.154
0.879
0.878
0.878
0.717
0.925
0.7172
S
V
M
89.4
0.129
0.896
0.894
0.895
0.756
0.882
0.7561
NB
87.4
0.162
0.875
0.875
0.875
0.709
0.946
0.7087
DT
93.7
0.079
0.938
0.937
0.938
0.855
0.967
0.8548
RF
94.0
0.075
0.941
0.941
0.941
0.861
0.984
0.8612
A
da
B
oos
t
85.4
0.136
0.869
0.855
0.858
0.687
0.918
0.6795
T
a
ble
8.
P
e
r
f
o
r
manc
e
a
na
lys
is
of
S
tatlog
da
tas
e
t
M
L
A
lg
or
it
hm
A
c
c
ur
a
c
y (
%
)
FP
r
a
te
P
r
e
c
is
io
n
R
e
c
a
ll
F
-
m
e
a
s
ur
e
M
C
C
R
O
C
a
r
e
a
K
a
ppa
v
a
lu
e
LR
88.1
0.143
0.885
0.881
0.883
0.725
0.955
0.7237
KNN
84.0
0.195
0.846
0.841
0.843
0.631
0.866
0.6299
S
V
M
89.2
0.152
0.892
0.893
0.892
0.743
0.870
0.7434
NB
85.9
0.216
0.857
0.859
0.858
0.659
0.943
0.6577
DT
91.8
0.103
0.919
0.919
0.919
0.806
0.953
0.806
RF
90
0.149
0.899
0.900
0.899
0.760
0.975
0.7594
A
da
B
oos
t
85.9
0.113
0.885
0.859
0.864
0.709
0.907
0.6931
T
he
s
tudy
dis
c
ove
r
e
d
that
f
or
the
C
leve
land
da
tas
e
t,
R
F
e
xc
e
e
ds
other
c
las
s
if
ier
s
with
a
n
a
c
c
ur
a
c
y
s
c
or
e
of
94
.
0%
in
T
a
ble
7
a
nd
it
s
e
xpe
r
im
e
ntal
r
e
s
ult
s
on
W
E
KA
v3.
9
.
6
a
r
e
s
hown
in
F
igu
r
e
2
.
W
i
t
h
a
lm
os
t
the
s
a
me
a
c
c
ur
a
c
y
o
f
93.
7%
,
DT
pe
r
f
or
ms
be
tt
e
r
a
f
ter
R
F
.
T
he
r
e
f
o
r
e
,
it
c
a
n
be
c
onc
luded
that,
in
t
e
r
ms
of
a
c
c
ur
a
c
y,
c
ons
ider
ing
the
C
leve
land
da
tas
e
t,
R
F
a
nd
DT
a
r
e
be
tt
e
r
c
hoice
s
il
lus
tr
a
ted
in
F
igu
r
e
3
.
F
or
the
s
tatlog
da
tas
e
t,
the
outcome
s
r
e
ve
a
led
that
DT
e
x
c
e
e
ds
other
c
las
s
if
ier
s
,
with
a
n
a
c
c
ur
a
c
y
s
c
or
e
of
91.
8%
in
T
a
ble
8
.
W
it
h
a
n
a
lm
os
t
identica
l
a
c
c
ur
a
c
y
of
90
%
a
s
DT
,
R
F
wor
ks
be
tt
e
r
a
f
ter
it
.
T
he
f
unda
me
ntal
a
nd
pr
a
c
ti
c
a
l
e
va
luation
metr
ic
is
a
c
c
ur
a
c
y;
howe
ve
r
,
i
t
mi
ght
not
be
e
nough
in
da
tas
e
ts
that
a
r
e
im
ba
lanc
e
d
a
nd
ha
ve
a
pr
e
domi
na
nc
e
of
one
c
las
s
ove
r
the
o
ther
.
S
ince
both
of
the
da
tas
e
ts
us
e
d
in
thi
s
r
e
s
e
a
r
c
h
a
r
e
e
ve
nly
dis
tr
ibut
e
d
a
nd
ba
lanc
e
d,
ther
e
f
or
e
,
D
T
,
a
nd
R
F
c
a
n
be
c
ons
ider
e
d
a
s
a
ppr
op
r
iate
c
las
s
if
ier
s
in
ter
ms
of
a
c
c
ur
a
c
y
metr
ics
f
or
bo
th
of
the
da
tas
e
ts
.
I
n
s
it
ua
ti
ons
whe
r
e
mi
nim
izing
f
a
ls
e
pos
it
ives
is
o
f
utm
os
t
im
por
tanc
e
,
s
uc
h
a
s
in
HD
pr
e
diction,
p
r
e
c
is
ion
plays
a
vit
a
l
r
ole
.
F
a
ls
e
pos
it
ives
mi
ght
c
a
us
e
wor
r
y
o
r
unne
e
de
d
medic
a
l
pr
oc
e
dur
e
s
.
A
higher
leve
l
of
pr
e
c
is
ion
s
igni
f
ies
a
r
e
duc
e
d
oc
c
ur
r
e
nc
e
of
f
a
ls
e
pos
it
ives
.
F
or
the
C
leve
land
da
tas
e
t,
R
F
ha
s
a
c
hieve
d
the
hi
ghe
s
t
pr
e
c
is
ion
of
0
.
941,
f
oll
owe
d
by
DT
with
a
pr
e
c
is
ion
of
0
.
938.
W
it
h
a
pr
e
c
is
ion
o
f
0
.
919
f
or
the
s
tatlog
da
tas
e
t,
DT
of
f
e
r
s
the
highes
t
pr
e
c
is
ion,
f
o
ll
ow
e
d
by
R
F
with
0.
899.
I
n
p
r
e
diction,
s
e
ns
it
ivi
ty
(
r
e
c
a
ll
)
plays
a
c
r
it
ica
l
r
ole
in
mi
nim
izing
f
a
ls
e
ne
ga
ti
ve
s
to
e
ns
ur
e
that
indi
viduals
with
HD
a
r
e
a
c
c
ur
a
tely
identif
ied.
I
n
the
C
leve
land
da
tas
e
t,
R
F
de
mons
tr
a
ted
the
highes
t
s
e
ns
it
ivi
ty
of
0.
941,
while
DT
f
oll
owe
d
c
los
e
ly
be
hind
with
a
s
e
ns
it
ivi
ty
of
0
.
937.
C
onve
r
s
e
l
y,
in
th
e
s
tatlog
da
tas
e
t,
DT
e
xhibi
ted
the
highes
t
s
e
ns
it
ivi
ty
of
0.
9
19,
with
R
F
tr
a
i
li
ng
s
li
ghtl
y
a
t
a
s
e
ns
it
ivi
ty
o
f
0
.
90
0.
M
C
C
e
xa
mi
ne
s
the
r
e
lations
hip
be
twe
e
n
a
c
tual
a
nd
pr
e
dicte
d
va
lues
.
A
s
tr
ong
c
or
r
e
lation
lea
ds
to
a
c
c
ur
a
te
pr
e
dictions
.
T
he
M
C
C
va
lue
of
a
pe
r
f
e
c
t
pr
e
dicti
on
is
+
1,
whe
r
e
a
s
the
M
C
C
va
lue
of
a
c
ompl
e
tel
y
wr
ong
pr
e
diction
is
-
1.
R
a
ndom
pr
e
dictions
a
r
e
im
pl
ied
b
y
a
va
lue
c
los
e
to
0
.
R
F
ha
d
the
highes
t
M
C
C
s
c
or
e
f
or
the
C
leve
land
da
tas
e
t,
a
t
0
.
861,
whic
h
wa
s
f
oll
owe
d
b
y
DT
,
whic
h
ha
d
0.
8
55
.
W
it
h
a
n
M
C
C
va
lue
of
0
.
806,
D
T
Evaluation Warning : The document was created with Spire.PDF for Python.