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
, N
o.
1
,
F
e
br
ua
r
y 2026
, pp.
695
~
706
I
S
S
N
:
2252
-
8938
,
D
O
I
:
10.11591/
ij
a
i.
v
15
.i
1
.pp
695
-
706
695
Jou
r
n
al
h
om
e
page
:
ht
tp
:
//
ij
ai
.
ia
e
s
c
or
e
.c
om
D
e
e
p
f
e
a
t
u
r
e
-
b
ase
d
m
u
l
t
i
-
c
l
ass
A
l
z
h
e
i
m
e
r
’
s d
i
se
ase
c
l
ass
i
f
i
c
at
i
on
w
i
t
h
st
at
i
st
i
c
al
p
e
r
f
o
r
m
an
c
e
e
val
u
at
i
on
M
ays
al
oon
A
b
e
d
Q
as
im
1
, M
a
r
w
a M
aw
f
aq
M
oh
am
e
d
s
h
e
e
t
A
l
-
H
at
ab
2
, L
u
b
ab
H.
A
lb
ak
1
1
T
e
c
hni
c
a
l
E
ngi
ne
e
r
i
ng C
ol
l
e
ge
f
or
C
om
put
e
r
a
nd A
r
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
ge
nc
e
, N
or
t
he
r
n T
e
c
hni
c
a
l
U
ni
ve
r
s
i
t
y, M
os
ul
, I
r
a
q
2
T
e
c
hni
c
a
l
E
ngi
ne
e
r
i
ng C
ol
l
e
ge
, N
or
t
he
r
n T
e
c
hni
c
a
l
U
ni
ve
r
s
i
t
y, M
os
ul
, I
r
a
q
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
O
c
t
8, 2024
R
e
vi
s
e
d
J
a
n 4, 2026
A
c
c
e
pt
e
d
J
a
n 22, 2026
This
study
evaluated
the
performance
of
multiple
machine
learning
classifi
ers
for
the
classifi
cation
of
Alzheimer’s
disease
(AD)
stages
using
deep
features
extracted
from
a
pre
-
trained
SqueezeN
et
model.
M
agnetic
resonance
imaging
(MRI)
sc
ans
were
pro
cessed
through
Squee
ze
Net
to
generate high
-
dimensional feature vectors, whic
h were then used as ac
hieved
an
accuracy
of
94.78%
input
to
six
classifiers:
k
-
nearest
neighbors
(
K
NN),
decision
tree
(DT)
,
support
vector
machine
(SVM),
neural
network
(NN),
naive
Bayes
(NB),
and
logistic
regression
(LR).
Models
were
assessed
using
a
70/30%
training
-
testing
split
and
5
-
,
10
-
,
and
20
-
fold
stratified
cross
-
validation. Principal component analysis (PCA) was applied to retain
99% of
variance.
On
the
original
dataset
consisting
of
6,400
images,
KN
N
has
achieved
97.48%
accuracy
and
0.998
area
under
the
curve
(AUC),
an
d
when
a
larger
dataset
of
44,000
images
was
used
it
achieved
an
accuracy
and
of
94.78%
and
an
AUC
of
0.987,
demonstrating
the
system’s
robustness
across
scales.
S
tatistical
tests,
including
paired
t
-
tests
and
Wilcoxon
signe
d
-
rank
tests,
confirmed
that
KNN
has
significantly
leveraged
from
PCA.
These
outcomes
demonstrate
that
combining
deep
feature
extraction
with
PCA
improved
the
reliability
and
efficie
ncy
of
the
classifier
for
AD
stage
prediction.
K
e
y
w
o
r
d
s
:
A
lz
he
im
e
r
'
s
di
s
e
a
s
e
K
-
n
e
a
r
e
s
t
ne
ig
hbor
s
P
r
in
c
ip
a
l
c
om
pone
nt
a
na
ly
s
is
S
que
e
z
e
N
e
t
W
il
c
oxon s
ig
ne
d
-
r
a
nk t
e
s
ts
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
:
M
a
ys
a
lo
on A
b
e
d Q
a
s
im
T
e
c
hni
c
a
l
E
ngi
ne
e
r
in
g C
ol
le
ge
f
or
C
om
put
e
r
a
nd A
r
ti
f
ic
ia
l
I
nt
e
ll
ig
e
nc
e
, N
or
th
e
r
n T
e
c
hni
c
a
l
U
ni
v
e
r
s
it
y
M
os
ul
, I
r
a
q
E
m
a
il
:
m
a
ys
lo
on.a
lh
a
s
hi
m
@
nt
u.e
du.i
q
1.
I
N
T
R
O
D
U
C
T
I
O
N
A
lz
he
im
e
r
’
s
di
s
e
a
s
e
(
A
D
)
is
c
ons
id
e
r
e
d
th
e
l
e
a
di
ng
c
a
u
s
e
of
de
m
e
nt
ia
w
or
ld
w
id
e
,
it
c
a
n
be
c
ha
r
a
c
te
r
iz
e
d
by s
ym
pt
om
s
li
ke
m
e
m
or
y
lo
s
s
, c
ogni
ti
ve
de
c
li
ne
,
a
nd
c
ha
nge
s
in
m
ood
or
p
e
r
s
ona
li
ty
[
1]
.
T
he
m
os
t
a
f
f
e
c
te
d
r
e
gi
ons
of
th
e
br
a
in
in
c
lu
de
th
e
hi
ppoc
a
m
pus
,
a
m
ygda
la
,
a
nd
ot
he
r
c
om
pone
nt
s
of
th
e
li
m
bi
c
s
ys
te
m
,
w
hi
c
h
pl
a
y
a
c
r
uc
i
a
l
r
ol
e
in
c
ogni
ti
ve
f
unc
ti
oni
ng
[
2
]
.
M
a
ny
in
di
vi
dua
ls
e
xpe
r
ie
nc
e
a
tr
a
ns
it
io
na
l
s
ta
ge
of
c
ogni
ti
ve
de
c
li
ne
c
a
ll
e
d
m
il
d
c
ogni
ti
ve
im
pa
ir
m
e
nt
(
M
C
I
)
be
f
or
e
th
e
a
ppe
a
r
a
nc
e
of
th
e
s
e
ve
r
e
A
lz
he
im
e
r
’
s
s
ym
pt
om
s
;
th
is
pe
r
io
d
r
e
pr
e
s
e
nt
s
a
n
in
te
r
m
e
di
a
te
c
ondi
ti
on
be
twe
e
n
nor
m
a
l
a
gi
ng
a
nd
A
D
[
3]
.
T
he
r
e
f
or
e
, M
C
I
i
s
c
ons
id
e
r
e
d a
s
ig
ni
f
ic
a
nt
i
ndi
c
a
to
r
f
or
e
a
r
ly
di
a
gnos
is
of
t
he
A
D
[
4]
.
P
hys
ic
ia
ns
us
e
a
gr
oup
of
te
c
hni
que
s
in
c
ol
la
bor
a
ti
on
w
it
h
ne
ur
ol
ogi
s
ts
a
nd
ne
ur
o
ps
yc
hol
ogi
s
ts
in
di
a
gnos
in
g
A
D
[
5]
.
T
he
s
e
di
a
gnos
ti
c
a
ppr
oa
c
h
e
s
in
c
lu
de
r
e
vi
e
w
in
g
c
li
ni
c
a
l
hi
s
to
r
ie
s
,
c
onduc
ti
ng
phy
s
ic
a
l
a
nd
ne
ur
ol
ogi
c
a
l
e
xa
m
in
a
ti
ons
,
pe
r
f
or
m
in
g
di
a
gnos
ti
c
te
s
t
s
,
a
nd
a
dm
in
is
te
r
in
g
c
ogni
ti
ve
a
s
s
e
s
s
m
e
nt
s
s
uc
h
a
s
th
e
m
in
i
-
m
e
nt
a
l
s
ta
te
e
xa
m
in
a
ti
o
n
(
M
M
S
E
)
[
6]
.
H
ow
e
ve
r
,
th
e
s
e
tr
a
di
ti
ona
l
m
e
th
ods
m
a
y
be
ti
m
e
-
c
ons
um
in
g
a
nd
c
a
n
le
a
d
to
in
c
ons
is
te
nt
r
e
s
ul
ts
.
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
, N
o.
1
,
F
e
br
ua
r
y 2026
:
695
-
706
696
B
e
c
a
us
e
A
D
pr
im
a
r
il
y
a
f
f
e
c
ts
th
e
gr
a
y
m
a
tt
e
r
of
th
e
br
a
in
[
7]
,
im
a
gi
ng
of
th
e
br
a
in
ha
s
be
c
om
e
a
n
e
s
s
e
nt
ia
l
to
ol
in
a
na
ly
z
in
g
th
e
f
unc
ti
ona
l
a
nd
s
tr
uc
tu
r
a
l
c
ha
n
ge
s
a
s
s
oc
i
a
te
d
w
it
h
th
e
di
s
e
a
s
e
[
8]
.
A
r
ti
f
ic
ia
l
ne
ur
a
l
ne
twor
ks
(
ANNs
)
,
w
hi
c
h
a
r
e
a
c
om
put
a
ti
ona
l
a
ppr
oa
c
h
in
s
pi
r
e
d
by
bi
ol
ogi
c
a
l
n
e
ur
a
l
s
ys
t
e
m
s
,
ha
v
e
e
xhi
bi
te
d
a
gr
e
a
t
pot
e
nt
ia
l
in
th
e
e
a
r
ly
de
te
c
ti
on
of
A
D
.
A
N
N
c
a
n
qua
nt
if
y
di
s
ti
nc
t
pa
tt
e
r
ns
a
nd
bi
om
a
r
ke
r
s
a
s
s
oc
ia
t
e
d
w
it
h
th
e
c
ondi
ti
on
by
a
na
ly
z
in
g
l
a
r
ge
da
ta
s
e
ts
of
br
a
in
im
a
ge
s
[
9]
,
[
10]
.
T
r
a
in
in
g
th
e
s
e
ne
twor
ks
on
huge
im
a
gi
ng
da
ta
,
c
a
n
he
lp
c
li
ni
c
ia
ns
in
di
a
gnos
in
g
a
nd
m
oni
to
r
in
g
A
D
m
or
e
e
f
f
e
c
ti
ve
ly
a
nd
e
f
f
ic
ie
nt
ly
.
O
nc
e
tr
a
in
e
d,
A
N
N
s
c
a
n
c
la
s
s
if
y
ne
w
un
s
e
e
n
da
ta
a
c
c
ur
a
te
ly
.
D
e
e
p
n
e
ur
a
l
ne
twor
k
s
(
D
N
N
s
)
,
w
hi
c
h
a
r
e
a
s
ubgr
oup
of
A
N
N
s
,
c
ont
a
in
in
g
m
ul
ti
pl
e
hi
dde
n
la
ye
r
s
be
twe
e
n t
he
in
put
a
nd
out
put
th
a
t
e
na
bl
e
th
e
m
to
le
a
r
n
c
om
pl
e
x
r
e
pr
e
s
e
nt
a
ti
ons
a
nd
obt
a
in
hi
gh
le
ve
l
a
c
c
ur
a
c
y
in
di
f
f
e
r
e
nt
a
ppl
ic
a
ti
ons
[
11]
,
[
12]
.
S
ys
te
m
s
us
in
g
D
N
N
s
c
a
n
r
e
c
ogni
z
e
pa
tt
e
r
ns
,
m
a
ke
pr
e
di
c
ti
ons
,
a
nd
s
ol
ve
a
v
a
r
ie
ty
of
c
om
pl
e
x
pr
ob
le
m
s
in
di
f
f
e
r
e
nt
f
ie
ld
s
s
uc
h a
s
c
om
put
e
r
vi
s
io
n, s
pe
e
c
h r
e
c
ogni
ti
on, a
nd na
tu
r
a
l
la
ngua
ge
pr
oc
e
s
s
in
g
[
13]
–
[
15]
.
2.
L
I
T
E
R
A
T
U
R
E
R
E
V
I
E
W
R
e
c
e
nt
ly
,
th
e
u
ti
li
z
a
ti
on
of
m
a
c
hi
n
e
l
e
a
r
ni
ng
(
ML
)
a
n
d
d
e
e
p
l
e
a
r
ni
n
g
(
DL
)
te
c
h
ni
q
ue
s
in
d
ia
gn
os
is
of
A
D
ha
s
a
d
va
nc
e
d
r
e
m
a
r
ka
bl
y
, e
s
p
e
c
ia
l
ly
t
hr
o
ug
h
br
a
i
n
im
a
g
in
g da
t
a
a
n
a
l
ys
is
.
I
n
2
022
,
A
lS
a
e
e
d a
n
d O
m
a
r
[
1
6]
pr
opos
e
d
a
hybr
id
s
y
s
te
m
c
om
bi
ni
ng
c
onvolut
io
na
l
ne
ur
a
l
n
e
twor
k
(
C
N
N
)
w
it
h
tr
a
di
ti
ona
l
M
L
f
or
A
D
c
la
s
s
if
ic
a
ti
on
us
in
g
m
a
gne
ti
c
r
e
s
ona
n
c
e
im
a
gi
ng
(
M
R
I
)
da
ta
.
T
he
a
ppr
oa
c
h
us
e
d
C
N
N
a
s
f
e
a
tu
r
e
e
xt
r
a
c
ti
on
a
nd M
L
a
s
a
c
la
s
s
if
ie
r
. T
he
ir
r
e
s
ul
ts
s
how
e
d t
ha
t
c
om
bi
ni
ng f
e
a
tu
r
e
s
de
r
iv
e
d f
r
om
C
N
N
w
it
h a
lg
or
it
hm
s
s
uc
h
a
s
s
uppor
t
ve
c
to
r
m
a
c
hi
ne
s
(
S
V
M
)
a
nd
r
a
ndom f
o
r
e
s
ts
s
ig
ni
f
ic
a
nt
ly
e
nha
nc
e
d di
a
gnos
ti
c
a
c
c
ur
a
c
y, e
s
pe
c
ia
ll
y
w
he
n
us
in
g
li
m
it
e
d
da
ta
s
a
m
pl
e
s
.
I
n
2023
,
K
ha
li
d
e
t
al
.
[
17]
de
ve
lo
pe
d
im
pr
ove
d
f
e
a
tu
r
e
e
xt
r
a
c
ti
on
s
tr
a
te
gy
th
a
t
f
us
e
d
C
N
N
-
ba
s
e
d
a
nd
ha
ndc
r
a
f
te
d
te
xt
ur
e
f
e
a
tu
r
e
s
f
or
pr
e
di
c
ti
ng
of
m
ul
ti
pl
e
A
D
s
ta
ge
s
s
ta
r
ti
ng
f
r
om
nor
m
a
l
c
ogni
ti
ve
s
ta
te
s
to
M
C
I
a
nd
a
dva
n
c
e
d
A
D
.
P
ubl
ic
M
R
I
da
ta
s
e
ts
w
e
r
e
us
e
d
in
s
tu
dy. T
he
ir
s
ys
te
m
ha
s
obt
a
in
e
d
a
hi
ghe
r
r
obus
tn
e
s
s
a
nd
s
ta
g
e
-
s
pe
c
if
ic
a
c
c
ur
a
c
y
c
om
pa
r
e
d
to
s
ys
te
m
s
us
in
g
tr
a
di
ti
ona
l
C
N
N
onl
y.
I
n t
he
s
a
m
e
ye
a
r
,
C
he
r
ia
n
e
t
al
.
[
18]
s
tu
di
e
d t
he
p
ot
e
nt
i
a
l
of
di
f
f
e
r
e
nt
M
L
a
ppr
o
a
c
h
e
s
l
i
ke
, d
e
c
i
s
io
n t
r
e
e
s
(
D
T
)
,
k
-
ne
a
r
e
s
t
ne
i
ghbor
s
(
K
N
N
)
,
a
n
d
lo
gi
s
t
ic
r
e
gr
e
s
s
io
n
(
L
R
)
in
e
a
r
l
y
de
t
e
c
t
io
n of
A
D
. T
h
e
ir
s
tu
d
y
hi
ghl
ig
h
te
d
th
a
t
tr
a
di
ti
on
a
l
M
L
a
lg
or
it
hm
s
r
e
m
a
in
va
lu
a
bl
e
w
h
e
n
i
m
pl
e
m
e
nt
e
d
in
lo
w
-
c
om
pl
e
xi
ty
s
y
s
te
m
s
a
nd
c
a
n
c
om
pl
e
m
e
nt
th
e
w
or
ki
n
g of
d
e
e
p
ne
t
w
or
ks
i
n h
ybr
id
di
a
gno
s
ti
c
pi
pe
li
n
e
s
.
A
ls
o
i
n
202
3
,
V
a
s
hi
s
ht
ha
e
t
al
.
[
19]
de
s
ig
ne
d
a
hybr
id
D
L
m
ode
l
to
de
te
c
t
e
a
r
ly
s
ig
ns
of
A
D
f
r
om
M
R
I
s
c
a
ns
,
u
s
in
g
a
dva
n
c
e
d
C
N
N
a
r
c
hi
te
c
tu
r
e
s
s
u
c
h
a
s
I
nc
e
pt
io
nV
2
a
nd
R
e
s
N
e
t5
0.
T
he
s
e
m
ode
l
s
'
hybr
id
iz
a
ti
on
ha
s
le
d
to
im
pr
ove
d
f
e
a
tu
r
e
r
e
pr
e
s
e
nt
a
ti
on
a
nd
e
nha
nc
e
d
c
l
a
s
s
if
ic
a
ti
on
a
c
c
ur
a
c
y
th
r
ough
bot
h
tr
a
in
in
g
a
nd
va
li
da
ti
on
da
ta
s
e
ts
.
I
n
2024,
N
a
s
ir
e
t
al
.
[
20]
de
ve
lo
pe
d
a
hybr
id
s
ys
te
m
us
in
g
d
e
e
p
a
nd
m
e
ta
-
le
a
r
ni
ng
m
ode
ls
f
or
M
R
I
-
ba
s
e
d
A
D
c
la
s
s
if
ic
a
ti
on,
de
m
o
ns
tr
a
ti
ng
im
pr
ove
d
a
c
c
ur
a
c
y
w
he
n
c
om
bi
ni
ng
C
N
N
s
w
it
h m
e
ta
-
le
a
r
ne
r
s
. M
or
e
r
e
c
e
nt
ly
, i
n 2025, L
iu
e
t
al
.
[
2
1]
in
tr
oduc
e
d a
m
ul
ti
-
m
oda
l
tr
a
ns
f
or
m
e
r
-
ba
s
e
d
f
r
a
m
e
w
or
k
c
om
bi
ne
d
M
R
I
,
ge
not
ypi
c
.
T
hi
s
f
r
a
m
e
w
or
k
pr
e
di
c
te
d
br
a
in
a
ge
in
g,
c
ogni
ti
ve
de
c
li
ne
,
a
nd
a
m
yl
oi
d
pa
th
ol
ogy,
e
xt
e
ndi
ng
di
a
gnos
ti
c
c
a
pa
bi
li
ty
be
yond
s
im
pl
e
c
la
s
s
if
ic
a
ti
on.
I
n
th
e
s
a
m
e
ye
a
r
,
A
r
ya
e
t
al
.
[
22
]
w
or
ke
d
on
a
s
ys
te
m
a
ti
c
r
e
vi
e
w
,
w
hi
c
h
s
um
m
a
r
iz
e
d
r
e
c
e
nt
D
L
a
nd
M
L
te
c
hni
que
s
,
in
c
lu
di
ng
C
N
N
s
,
tr
a
n
s
f
e
r
le
a
r
ni
ng,
a
nd
hybr
id
a
r
c
hi
te
c
tu
r
e
s
a
c
r
o
s
s
im
a
gi
ng
m
oda
li
ti
e
s
f
or
A
D
de
te
c
ti
on,
w
it
h
f
oc
us
in
g
on c
ha
ll
e
nge
s
s
uc
h a
s
da
ta
s
e
t
im
ba
la
nc
e
a
nd mod
e
l
in
te
r
pr
e
ta
bi
li
ty
.
3.
M
A
T
E
R
I
A
L
S
A
N
D
M
E
T
H
O
D
S
3.1. Dat
as
e
t
I
n t
hi
s
s
tu
dy, a
publi
c
ly
a
c
c
e
s
s
ib
le
AD
M
R
I
da
ta
s
e
t
f
r
om
K
a
ggl
e
ha
s
be
e
n ut
il
iz
e
d, w
hi
c
h c
ons
i
s
ts
of
s
tr
uc
tu
r
a
l
M
R
I
im
a
ge
s
gr
oups
in
to
f
our
c
li
ni
c
a
ll
y
r
e
le
va
nt
c
la
s
s
e
s
:
non
-
de
m
e
nt
e
d,
ve
r
y
m
il
d
de
m
e
nt
e
d,
m
il
d
de
m
e
nt
e
d,
a
nd
m
ode
r
a
te
de
m
e
nt
e
d
.
T
h
e
tr
a
in
in
g
-
te
s
ti
ng
s
pl
i
t
pr
ovi
de
d
on
th
e
w
e
bs
it
e
w
a
s
in
te
nt
io
na
ll
y
a
voi
de
d
be
c
a
u
s
e
it
u
s
e
d
a
s
e
le
c
ti
on
c
r
it
e
r
io
n
m
a
y
in
tr
oduc
e
hi
dde
n
bi
a
s
e
s
th
a
t
w
e
a
ke
n
r
e
pr
oduc
ib
il
it
y.
I
ns
te
a
d, a
s
ubs
e
t
of
6,400 im
a
ge
s
(
176×
208 pixe
ls
, J
P
G
f
or
m
a
t)
ha
s
be
e
n c
ol
le
c
te
d f
r
om
t
he
c
om
pl
e
te
da
ta
s
e
t,
di
s
tr
ib
ut
e
d
a
s
f
ol
lo
w
s
:
3,200
non
-
d
e
m
e
nt
e
d,
2,240
ve
r
y
m
il
d
de
m
e
nt
e
d,
896
m
il
d
de
m
e
nt
e
d,
a
nd
64
m
ode
r
a
te
de
m
e
nt
e
d
.
A
70
-
30
tr
a
in
-
te
s
t
di
vi
s
io
n
w
a
s
im
pl
e
m
e
nt
e
d,
in
or
de
r
to
e
ns
ur
e
f
a
ir
ne
s
s
a
nd
r
obus
t
e
v
a
lu
a
ti
on.
T
he
pr
obl
e
m
of
im
ba
l
a
nc
e
d
d
a
ta
,
e
s
pe
c
ia
ll
y
th
e
m
ode
r
a
te
de
m
e
nt
e
d
li
m
it
e
d
r
e
pr
e
s
e
nt
a
ti
on
of
s
a
m
pl
e
s
,
is
r
e
c
ogni
z
e
d.
T
hi
s
in
e
qua
li
ty
r
e
f
le
c
ts
th
e
r
e
a
l
c
li
ni
c
a
l
di
s
tr
ib
ut
io
n
of
A
lz
he
im
e
r
’
s
s
ta
ge
s
,
w
hi
c
h
e
nh
a
nc
e
s
th
e
e
c
ol
ogi
c
a
l
va
li
di
ty
of
th
e
da
ta
s
e
t.
I
n
a
ddi
ti
on,
m
ul
ti
pl
e
c
r
os
s
-
v
a
li
da
ti
on
e
xpe
r
im
e
nt
s
ha
ve
be
e
n
pe
r
f
or
m
e
d
to
r
e
duc
e
ove
r
f
it
ti
ng
a
nd
c
onf
ir
m
th
e
r
e
li
a
bi
li
ty
of
th
e
r
e
s
ul
ts
.
T
hi
s
m
e
th
odol
ogi
c
a
l
de
c
i
s
io
n
e
nha
n
c
e
s
our
r
e
s
ul
ts
c
r
e
di
bi
li
ty
a
nd
a
ddr
e
s
s
e
s
di
r
e
c
tl
y
c
onc
e
r
ns
r
e
ga
r
di
ng
da
ta
s
e
t
c
la
r
it
y
a
nd
c
la
s
s
di
s
tr
ib
ut
io
n
a
nd
di
s
tr
ib
ut
io
n
of
th
e
c
la
s
s
.
F
ig
ur
e
1
s
how
s
r
e
pr
e
s
e
nt
a
ti
ve
e
xa
m
pl
e
s
of
th
e
f
our
di
a
gnos
ti
c
c
a
te
gor
ie
s
[
23]
:
F
ig
ur
e
1(
a
)
m
i
ld
de
m
e
nt
e
d
,
F
ig
ur
e
1(
b
)
m
ode
r
a
te
de
m
e
nt
e
d
,
F
i
g
ur
e
1(
c
)
non
-
de
m
e
nt
e
d
,
a
nd
F
ig
u
r
e
1(
d)
ve
r
y
m
il
d de
m
e
nt
e
d
.
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
D
e
e
p f
e
at
ur
e
-
bas
e
d m
ul
ti
-
c
la
s
s
A
lz
he
ime
r
’
s
di
s
e
as
e
c
la
s
s
if
ic
at
i
on w
it
h
…
(
M
ay
s
al
oon A
be
d Q
as
im
)
697
(
a
)
(
b)
(
c
)
(
d)
F
ig
ur
e
1. C
la
s
s
di
s
tr
ib
ut
io
n i
n t
he
da
ta
s
e
t:
(
a
)
m
il
d de
m
e
nt
e
d, (
b)
m
ode
r
a
te
de
m
e
nt
e
d, (
c
)
non
-
de
m
e
nt
e
d, a
nd
(
d)
ve
r
y m
il
d de
m
e
nt
e
d
3.2. De
e
p
f
e
at
u
r
e
e
xt
r
a
c
t
io
n
A
pr
e
-
tr
a
in
e
d
S
que
e
z
e
N
e
t
m
ode
l
w
a
s
a
ppl
ie
d
f
or
de
e
p
f
e
a
tu
r
e
e
xt
r
a
c
ti
on.
I
m
a
ge
s
w
e
r
e
s
e
nt
to
th
e
ne
twor
k
th
r
ough
a
de
di
c
a
te
d
f
e
a
tu
r
e
la
ye
r
,
th
e
n
th
e
r
e
s
ul
ti
ng
f
e
a
tu
r
e
ve
c
to
r
s
w
e
r
e
us
e
in
th
e
a
na
ly
s
is
a
nd
c
la
s
s
if
ic
a
ti
on.
S
que
e
z
e
N
e
t
i
s
a
c
om
pa
c
t
C
N
N
a
r
c
hi
te
c
tu
r
e
,
w
hi
c
h
de
s
ig
ne
d
to
pr
ovi
de
hi
gh
a
c
c
ur
a
c
y
w
it
h
s
ig
ni
f
ic
a
nt
ly
f
e
w
e
r
pa
r
a
m
e
te
r
s
,
m
a
ki
ng
it
s
ui
ta
bl
e
f
or
m
e
m
or
y
-
c
ons
tr
a
in
e
d
e
nvi
r
onm
e
nt
s
a
nd
f
or
e
f
f
ic
ie
nt
m
ode
l
de
pl
oym
e
nt
. I
ts
de
s
ig
n de
pe
nd
s
on 1×
1 a
nd 3
×
3 c
onvolu
ti
ons
t
o ke
e
p s
tr
ong pe
r
f
or
m
a
nc
e
[
24]
.
T
he
a
r
c
hi
te
c
tu
r
e
is
bui
lt
a
r
ound
th
e
f
ir
e
m
odul
e
,
w
hi
c
h
c
ons
is
ts
of
a
s
que
e
z
e
la
ye
r
w
it
h
1×
1
c
onvolut
io
ns
a
nd
a
n
e
xpa
nd
la
ye
r
in
te
gr
a
ti
ng
1×
1
a
nd
3×
3
f
il
te
r
s
to
c
a
pt
ur
e
m
ul
ti
-
s
c
a
le
s
pa
ti
a
l
f
e
a
tu
r
e
s
.
I
m
a
ge
s
w
e
r
e
pr
oc
e
s
s
e
d
th
r
ough
c
onv1
a
nd
f
ir
e
2
–
f
ir
e
9,
m
a
x
-
pool
in
g
w
a
s
a
ppl
ie
d
a
f
te
r
c
onv1,
f
ir
e
4,
f
ir
e
8,
a
nd
c
onv10
to
gr
a
dua
ll
y
d
e
c
r
e
a
s
e
s
p
a
ti
a
l
r
e
s
ol
ut
io
n
a
nd
m
a
in
ta
in
e
s
s
e
nt
ia
l
in
f
or
m
a
ti
on
[
24]
.
N
onl
in
e
a
r
f
e
a
tu
r
e
r
e
pr
e
s
e
nt
a
ti
on w
a
s
s
uppor
te
d by
r
e
c
ti
f
ie
d l
in
e
a
r
un
it
(
R
e
L
U
)
a
c
ti
va
ti
on, while
ove
r
f
it
ti
ng w
a
s
r
e
duc
e
d
by
us
in
g
dr
opout
a
f
te
r
f
ir
e
9.
T
hr
ough
th
e
s
tr
a
te
gi
c
dow
ns
a
m
pl
in
g
a
nd
m
odul
a
r
s
tr
uc
tu
r
e
o
f
S
que
e
z
e
N
e
t,
a
c
om
pa
c
t
a
nd
e
xpr
e
s
s
iv
e
m
ode
l
w
a
s
a
c
hi
e
ve
d
,
w
hi
c
h
is
a
bl
e
to
le
a
r
n
c
om
pl
e
x
f
e
a
tu
r
e
s
e
f
f
ic
ie
nt
ly
[
25]
.
F
ig
ur
e
2 pr
e
s
e
nt
s
t
he
de
ta
il
e
d a
r
c
hi
te
c
tu
r
e
.
F
ig
ur
e
2. S
que
e
z
e
N
e
t
a
r
c
hi
te
c
tu
r
e
w
it
h a
s
qu
e
e
z
e
a
nd e
xpa
nd p
r
oc
e
s
s
3.3. M
ac
h
in
e
l
e
ar
n
in
g an
d
m
od
e
l
op
t
im
iz
at
io
n
T
he
s
tu
dy
ut
il
iz
e
d
di
f
f
e
r
e
nt
M
L
a
lg
or
it
h
m
s
to
c
la
s
s
if
y
th
e
s
ta
ge
s
of
A
D
us
in
g
M
R
I
da
ta
f
e
a
tu
r
e
s
e
xt
r
a
c
te
d f
r
om
C
N
N
t
o i
m
pr
ove
t
he
di
a
gnos
ti
c
a
c
c
ur
a
c
y
[
26]
, [
27]
.
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
, N
o.
1
,
F
e
br
ua
r
y 2026
:
695
-
706
698
i)
KNN
:
K
N
N
id
e
nt
if
y
th
e
c
la
s
s
de
pe
ndi
ng
on
th
e
m
a
jo
r
it
y
a
m
ong
k
ne
a
r
e
s
t
s
a
m
pl
e
s
,
w
hi
c
h
m
e
a
s
ur
e
d
us
in
g E
uc
li
de
a
n di
s
ta
nc
e
[
27]
:
(
,
)
=
√
∑
(
−
)
2
=
1
(
1)
ii)
NB
:
N
B
us
e
s
B
a
y
e
s
’
t
he
or
e
m
t
o e
s
ti
m
a
te
t
he
po
s
te
r
io
r
pr
oba
bi
li
ty
of
c
la
s
s
A
gi
ve
n f
e
a
tu
r
e
s
B
[
28]
, [
29]
:
(
|
)
=
(
|
)
(
)
(
)
(
2)
iii)
LR
:
L
R
m
ode
ls
t
he
pos
it
iv
e
c
la
s
s
pr
oba
bi
li
ty
us
in
g t
he
s
ig
m
oi
d
f
unc
ti
on
[
30]
:
(
)
=
1
1
+
−
(
3)
a
nd r
e
duc
e
s
t
he
c
r
o
s
s
-
e
nt
r
opy los
s
:
(
)
=
−
1
∑
[
l
o
g
(
(
)
)
+
(
1
−
)
l
o
g
(
1
−
(
)
]
=
1
(
4)
iv
)
S
V
M
:
S
V
M
e
m
pl
oys
th
e
li
ne
a
r
de
c
is
io
n
f
unc
ti
on
to
de
te
r
m
in
e
th
e
opt
im
a
l
hype
r
pl
a
ne
th
a
t
m
a
xi
m
iz
e
s
th
e
m
a
r
gi
n be
twe
e
n c
la
s
s
e
s
[
31]
.
(
)
=
.
+
(
5)
w
he
r
e
w
a
nd b
a
r
e
th
e
w
e
ig
ht
a
nd bi
a
s
t
e
r
m
s
, r
e
s
p
e
c
ti
v
e
ly
.
K
e
r
n
e
l
f
unc
ti
on
s
e
na
bl
e
non
li
ne
a
r
s
e
p
a
r
a
ti
on
.
v)
DT
:
D
T
r
e
pe
a
te
dl
y di
vi
de
s
da
t
a
i
nt
o s
ubs
e
t
s
ba
s
e
d on e
nt
r
opy
[
31]
:
E
=
−
∑
(
)
2
=
1
(
(
)
)
(
6)
N
N
w
a
s
a
ls
o
ut
il
iz
e
d
to
m
ode
l
non
-
li
ne
a
r
r
e
la
ti
ons
hi
ps
a
m
ong high
-
di
m
e
ns
io
na
l
f
e
a
tu
r
e
s
,
in
vol
vi
ng
100
hi
dde
n unit
s
w
it
h R
e
L
U
a
c
ti
va
ti
on.
3.4. M
od
e
l
op
t
im
iz
at
io
n
3.4.1. Gr
id
S
e
ar
c
h
f
or
t
r
ad
it
io
n
al
m
ac
h
in
e
l
e
ar
n
in
g m
od
e
ls
S
V
M
, K
N
N
, D
T
, L
R
, a
nd N
B
hype
r
pa
r
a
m
e
te
r
s
w
e
r
e
opt
im
iz
e
d
us
in
g
gr
id
s
e
a
r
c
h
. A
5
-
f
ol
d
s
tr
a
ti
f
ie
d
c
r
os
s
-
va
li
da
ti
on
to
a
voi
d
ove
r
f
it
ti
ng
a
nd
r
e
ta
in
ba
la
nc
e
d
r
e
pr
e
s
e
nt
a
ti
on
of
A
D
c
la
s
s
e
s
.
T
a
bl
e
1
s
um
m
a
r
iz
e
s
th
e
c
onf
ig
ur
a
ti
ons
of
s
e
le
c
te
d hype
r
pa
r
a
m
e
te
r
.
T
a
bl
e
1. H
ype
r
pa
r
a
m
e
te
r
s
a
nd gr
id
s
e
a
r
c
h
s
e
tt
in
gs
f
or
t
r
a
di
ti
ona
l
M
L
m
ode
ls
M
ode
l
K
e
y hype
r
pa
r
a
m
e
t
e
r
s
J
us
t
i
f
i
c
a
t
i
on
S
V
M
C
=1.0, ε
=0.1,
p
ol
ynom
i
a
l
ke
r
ne
l
(
de
gr
e
e
3)
B
a
l
a
nc
e
s
m
a
r
gi
n a
nd m
i
s
c
l
a
s
s
i
f
i
c
a
t
i
on;
c
a
pt
ur
e
s
non
-
l
i
ne
a
r
pa
t
t
e
r
ns
KNN
k=2, M
e
t
r
i
c
:
M
a
nha
t
t
a
n,
d
i
s
t
a
nc
e
w
e
i
ght
i
ng
S
e
ns
i
t
i
ve
t
o ne
a
r
e
s
t
ne
i
ghbor
s
w
hi
l
e
m
i
ni
m
i
z
i
ng i
nf
l
ue
nc
e
f
r
om
di
s
t
a
nt
poi
nt
s
DT
M
a
x de
pt
h=100, P
r
uni
ng:
≥
4 i
ns
t
a
nc
e
s
i
n l
e
a
ve
s
, ≥
5 i
nt
e
r
na
l
node
s
L
i
m
i
t
s
ove
r
f
i
t
t
i
ng w
hi
l
e
c
a
pt
ur
i
ng c
om
pl
e
x
pa
t
t
e
r
ns
LR
L
2 r
e
gul
a
r
i
z
a
t
i
on, C
=650
R
e
duc
e
s
ov
e
r
f
i
t
t
i
ng i
n hi
gh
-
di
m
e
ns
i
ona
l
f
e
a
t
ur
e
s
pa
c
e
NB
D
e
f
a
ul
t
E
f
f
e
c
t
i
ve
f
or
l
a
r
ge
, i
nde
pe
nde
nt
f
e
a
t
ur
e
s
e
t
s
3.4.2. B
aye
s
ia
n
op
t
im
iz
at
io
n
f
or
N
N
m
od
e
l
T
o
c
la
s
s
if
y
A
D
s
ta
g
e
s
,
1000
de
e
p
f
e
a
tu
r
e
s
pe
r
im
a
ge
e
xt
r
a
c
te
d
f
r
om
S
que
e
z
e
N
e
t,
w
e
r
e
pr
oc
e
s
s
e
d
us
in
g
N
N
.
B
a
ye
s
i
a
n
o
pt
im
iz
a
ti
on
w
a
s
a
ppl
ie
d
to
f
in
e
-
tu
ne
hype
r
pa
r
a
m
e
te
r
s
,
s
uc
h
a
s
th
e
s
iz
e
of
hi
dde
n
la
ye
r
a
nd
s
tr
e
ngt
h
of
r
e
gul
a
r
iz
a
ti
on,
ba
s
e
d
on
c
r
os
s
-
va
li
da
te
d
F
1
-
s
c
or
e
a
nd
a
c
c
ur
a
c
y.
T
he
NN
us
e
d
100
hi
dde
n
la
ye
r
s
w
it
h
R
e
L
U
a
c
ti
va
ti
on
a
nd
A
da
m
opt
im
iz
e
r
,
in
c
or
por
a
ti
ng
L
2
r
e
gul
a
r
iz
a
ti
on
(
α=
0.0001)
to
r
e
duc
e
ove
r
f
it
ti
ng
a
nd
li
m
i
te
d
to
100
it
e
r
a
ti
ons
f
o
r
r
e
pl
ic
a
bl
e
tr
a
in
in
g.
I
t
to
ok
1,000
de
e
p
f
e
a
tu
r
e
s
f
r
om
S
que
e
z
e
N
e
t
a
c
r
os
s
6,400
im
a
ge
s
,
in
c
lu
di
ng
m
e
ta
-
a
tt
r
ib
ut
e
s
li
ke
im
a
ge
na
m
e
,
s
iz
e
,
w
id
th
,
a
nd
he
ig
ht
.
T
he
m
ode
l
a
im
e
d
to
c
la
s
s
if
y t
he
A
D
s
ta
ge
c
a
t
e
gor
y.
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
D
e
e
p f
e
at
ur
e
-
bas
e
d m
ul
ti
-
c
la
s
s
A
lz
he
ime
r
’
s
di
s
e
as
e
c
la
s
s
if
ic
at
i
on w
it
h
…
(
M
ay
s
al
oon A
be
d Q
as
im
)
699
3.5.
P
r
in
c
ip
al
c
om
p
on
e
n
t
an
al
ys
is
P
r
in
c
ip
a
l
c
om
pone
nt
a
na
ly
s
is
(
P
C
A
)
is
a
di
m
e
ns
io
na
li
ty
r
e
duc
ti
on
te
c
hni
que
ba
s
e
d
on
M
L
.
It
c
onve
r
ts
a
la
r
ge
da
ta
s
e
t
to
s
m
a
ll
e
r
s
e
t
of
c
om
pone
nt
s
w
hi
le
m
a
in
ta
in
in
g
e
s
s
e
nt
ia
l
pa
tt
e
r
ns
a
nd
va
r
ia
nc
e
.
I
n
th
is
s
tu
dy
1000
f
e
a
tu
r
e
s
e
xt
r
a
c
te
d
f
r
om
th
e
S
que
e
z
e
N
e
t
de
e
p
le
a
r
ni
ng
m
ode
l
w
e
r
e
r
e
duc
e
d
to
100
pr
in
c
ip
a
l
c
om
pone
nt
s
us
in
g P
C
A
.
F
in
a
l
D
a
t
a
=
R
o
wF
e
a
t
ur
e
V
e
c
t
o
r
×
R
o
wD
a
t
a
Ad
j
us
t
e
d
(
7)
3.6.
S
t
at
is
t
ic
al
s
ig
n
if
ic
an
c
e
an
al
ys
is
T
he
pa
ir
e
d
t
-
te
s
t
a
nd
th
e
W
il
c
oxon
s
ig
n
e
d
-
r
a
nk
te
s
t
ha
v
e
b
e
e
n
ut
il
iz
e
d,
to
id
e
nt
if
y
w
he
th
e
r
th
e
obs
e
r
ve
d
e
nha
n
c
e
m
e
nt
s
in
th
e
p
e
r
f
or
m
a
nc
e
of
th
e
c
la
s
s
if
ic
a
ti
on
a
f
te
r
a
ppl
yi
ng
P
C
A
w
e
r
e
s
ta
ti
s
ti
c
a
ll
y
s
ig
ni
f
ic
a
nt
.
T
he
s
e
te
s
ts
m
a
de
a
c
om
pa
r
is
on
be
twe
e
n
a
c
c
ur
a
c
ie
s
be
f
or
e
a
nd
a
f
te
r
P
C
A
us
in
g
m
ul
ti
pl
e
c
r
os
s
-
va
li
da
ti
on
f
ol
ds
to
c
onf
ir
m
th
a
t
th
e
ga
in
s
w
e
r
e
due
to
ge
nui
ne
im
pr
ove
m
e
nt
s
a
nd
not
ju
s
t
r
a
ndom
va
r
ia
ti
on.
A
c
c
ur
a
c
ie
s
of
th
e
c
la
s
s
if
ic
a
ti
on
r
e
s
ul
te
d
f
r
om
e
a
c
h
f
ol
d
c
or
r
e
s
ponde
d
a
s
p
a
ir
e
d
s
a
m
pl
e
s
f
or
s
ta
ti
s
ti
c
a
l
c
om
pa
r
is
on.
i)
P
a
ir
e
d t
-
te
s
t:
t
he
pa
ir
e
d
t
-
te
s
t
de
te
r
m
in
e
s
t
he
m
e
a
n di
f
f
e
r
e
nc
e
be
twe
e
n t
w
o pa
ir
e
d s
a
m
pl
e
s
, a
s
s
um
in
g t
he
nor
m
a
li
ty
. T
he
s
ta
ti
s
ti
c
of
t
he
t
e
s
t
i
s
gi
ve
n
in
:
=
̅
∕
√
(
8)
w
he
r
e
,
̅
is
t
he
m
e
a
n di
f
f
e
r
e
nc
e
be
tw
e
e
n obs
e
r
va
ti
on
s
pa
ir
e
d, i
t
is
c
om
put
e
d a
s
:
̅
=
1
∑
(
−
)
1
=
1
⋅
(
9)
a
nd
r
e
pr
e
s
e
nt
s
t
he
s
ta
nda
r
d de
vi
a
ti
on of
di
f
f
e
r
e
nc
e
s
, w
hi
c
h c
a
n
be
de
te
r
m
in
e
d a
s
:
=
√
1
−
1
∑
(
(
=
1
−
)
−
̅
)
2
(
10)
w
he
r
e
n
r
e
pr
e
s
e
nt
s
th
e
num
be
r
of
obs
e
r
va
ti
ons
pa
ir
e
d
(
c
r
os
s
-
va
li
da
ti
on
f
ol
ds
)
,
a
r
e
th
e
a
c
c
ur
a
c
ie
s
be
f
or
e
a
nd
a
f
te
r
a
ppl
yi
ng
P
C
A
,
r
e
s
pe
c
ti
ve
ly
.
T
he
s
m
a
ll
va
lu
e
of
p
(
<
0.05)
,
r
e
f
le
c
ts
th
e
s
ta
ti
s
ti
c
a
ll
y
s
ig
ni
f
ic
a
nt
i
m
pr
ove
m
e
nt
s
a
f
te
r
P
C
A
[
32]
.
ii)
W
il
c
oxon
s
ig
ne
d
-
r
a
nk
te
s
t:
th
e
W
il
c
oxon
s
ig
ne
d
-
r
a
nk
te
s
t
doe
s
not
a
s
s
um
e
a
nor
m
a
l
di
s
tr
ib
ut
io
n;
th
e
r
e
f
or
e
,
it
c
ons
id
e
r
s
a
s
a
non
-
pa
r
a
m
e
tr
ic
a
lt
e
r
na
ti
ve
to
th
e
p
a
ir
e
d
t
-
te
s
t.
I
t
a
s
s
e
s
s
e
s
if
a
ve
r
a
ge
of
th
e
di
f
f
e
r
e
nc
e
s
be
twe
e
n pa
ir
e
d
s
a
m
pl
e
s
i
s
z
e
r
o. T
h
e
t
e
s
t
s
t
a
ti
s
ti
c
c
a
n be
c
om
put
e
d us
in
g
:
=
∑
+
=
1
(
11)
w
he
r
e
i
s
th
e
di
f
f
e
r
e
nc
e
be
twe
e
n
pa
ir
e
d
s
a
m
pl
e
s
(
−
)
,
a
nd
+
is
th
e
r
a
nk
of
th
e
a
bs
ol
ut
e
di
f
f
e
r
e
nc
e
s
f
or
pos
it
iv
e
di
f
f
e
r
e
nc
e
s
(
>
0)
.
A
ve
r
a
ge
r
a
nks
w
e
r
e
a
s
s
ig
ne
d t
o t
he
t
ie
s
.
W
he
n t
he
va
lu
e
of
p
is
le
s
s
th
a
n
0.05,
th
is
r
e
f
e
r
s
th
a
t
th
e
ob
s
e
r
ve
d
di
f
f
e
r
e
nc
e
s
a
r
e
s
ta
ti
s
ti
c
a
ll
y
s
ig
ni
f
ic
a
nt
,
w
hi
c
h
c
onf
ir
m
s
th
a
t
P
C
A
ha
s
c
ont
r
ib
ut
e
d t
o i
m
pr
ove
t
he
pe
r
f
or
m
a
nc
e
of
t
he
m
ode
l
[
33]
.
4.
R
E
S
U
L
T
S
A
N
D
D
I
S
C
U
S
S
I
O
N
4.1. E
val
u
at
io
n
c
r
it
e
r
ia
i
n
d
ic
at
or
s
T
he
pe
r
f
or
m
a
nc
e
of
th
e
m
ode
l
ha
s
be
e
n
e
v
a
lu
a
te
d
us
in
g
a
c
c
u
r
a
c
y,
r
e
c
a
ll
,
pr
e
c
is
io
n
,
F
1
-
s
c
or
e
,
a
nd
a
r
e
a
unde
r
th
e
c
ur
ve
(
A
U
C
)
.
A
m
ong
a
ll
m
ode
ls
,
K
N
N
ha
s
obt
a
in
e
d
th
e
be
s
t
r
e
s
ul
t
s
w
it
h
93%
a
c
c
ur
a
c
y
a
nd
a
n
A
U
C
of
0.97,
s
how
in
g
a
gr
e
a
t
di
s
c
r
im
in
a
ti
ve
a
bi
li
ty
.
I
ts
c
onf
ig
ur
a
ti
on,
w
hi
c
h
c
ons
is
ts
of
two
ne
ig
hbor
s
,
M
a
nha
tt
a
n
di
s
ta
n
c
e
,
a
nd
di
s
ta
n
c
e
-
w
e
ig
ht
e
d
vot
in
g
ha
s
ta
k
e
n
pa
r
t
in
th
is
hi
gh
pe
r
f
or
m
a
nc
e
.
I
n
c
ont
r
a
s
t,
D
T
de
m
ons
tr
a
te
d
th
e
lo
w
e
s
t
pe
r
f
or
m
a
nc
e
of
55%
a
c
c
ur
a
c
y,
A
U
C
0.64,
r
e
f
e
r
r
in
g
to
a
li
m
i
te
d
s
ta
ge
c
la
s
s
if
ic
a
ti
on.
I
ts
pe
r
f
or
m
a
nc
e
c
a
n
be
e
nha
nc
e
d
by
t
uni
ng
pr
uni
ng
th
r
e
s
hol
ds
a
nd
de
pt
h
c
ons
tr
a
in
ts
.
M
ode
r
a
te
A
U
C
of
0.68
lo
w
a
c
c
ur
a
c
y
43%
ha
ve
be
e
n
a
c
hi
e
ve
d
by
S
V
M
,
s
ugge
s
ti
ng
th
e
r
e
qui
r
e
m
e
nt
f
o
r
f
u
r
th
e
r
ke
r
ne
l
a
nd
r
e
gul
a
r
iz
a
ti
on
opt
im
iz
a
ti
on.
T
he
pe
r
f
o
r
m
a
nc
e
of
N
N
w
a
s
w
e
ll
w
it
h
83
%
a
c
c
ur
a
c
y
a
nd
A
U
C
0.94,
w
hi
c
h
e
na
bl
e
it
to
c
a
pt
ur
e
M
R
I
d
a
ta
pa
tt
e
r
ns
e
f
f
e
c
ti
ve
ly
th
r
ough
it
s
1
00
hi
dde
n
la
ye
r
s
,
R
e
L
U
a
c
ti
va
ti
on,
a
nd
A
da
m
opt
im
iz
e
r
.
T
he
w
e
a
k
pe
r
f
or
m
a
nc
e
w
a
s
r
e
c
or
de
d
us
in
g
N
B
w
it
h
42%
a
c
c
ur
a
c
y,
A
U
C
0.67,
th
is
is
be
c
a
us
e
of
it
s
i
nde
pe
nde
nc
e
a
s
s
um
pt
io
n m
is
m
a
t
c
h.
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
, N
o.
1
,
F
e
br
ua
r
y 2026
:
695
-
706
700
4.1.1. Conf
u
s
io
n
m
at
r
ix
e
val
u
at
io
n
F
ig
ur
e
3
il
lu
s
tr
a
te
s
th
e
c
onf
us
io
n
m
a
tr
ix
f
or
th
e
K
N
N
m
ode
l,
r
e
ve
a
li
ng
th
e
c
la
s
s
-
w
i
s
e
pr
e
di
c
ti
ve
pe
r
f
or
m
a
nc
e
a
s
f
ol
lo
w
s
:
i)
M
il
d
de
m
e
nt
e
d:
t
he
pe
r
c
e
nt
a
g
e
of
c
or
r
e
c
tl
y c
la
s
s
if
ie
d w
e
r
e
92.4
%
w
it
h l
it
tl
e
m
is
c
la
s
s
if
ic
a
ti
ons
.
ii)
M
ode
r
a
te
de
m
e
nt
e
d
:
w
it
h 92.4%
c
or
r
e
c
tl
y c
la
s
s
if
ie
d, a
nd l
ow
f
a
ls
e
pos
it
iv
e
s
.
iii)
N
on
-
de
m
e
nt
e
d
:
w
it
h
95
.0%
c
or
r
e
c
tl
y
i
d
e
n
ti
f
i
e
d
,
a
n
d l
it
tl
e
m
i
s
c
la
s
s
if
i
c
a
ti
on
s
(
3.
3%
m
il
d
, 6
.6%
v
e
r
y
m
i
ld
)
.
iv
)
V
e
r
y
m
il
d
de
m
e
nt
e
d
:
r
e
a
c
h
e
d
91.7%
c
or
r
e
c
tl
y
c
la
s
s
if
ie
d,
a
n
d
li
m
it
e
d
ove
r
la
p
(
4.1%
non
-
de
m
e
nt
e
d,
3.9%
m
il
d)
.
T
h
e
s
e
r
e
s
u
lt
s
c
o
n
f
i
r
m
e
d
K
N
N
r
o
b
us
t
ne
s
s
t
h
r
o
u
g
h
a
l
l
A
D
s
t
a
ge
s
,
e
s
pe
c
ia
l
l
y
f
o
r
t
h
e
i
de
n
t
i
f
i
c
a
t
i
on
o
f
e
a
r
l
y
de
m
e
n
t
i
a
.
F
ig
ur
e
3. C
onf
us
io
n m
a
tr
ix
f
or
K
N
N
m
ode
l
4.1.2. ROC c
u
r
ve
an
al
ys
is
F
ig
ur
e
4 de
pi
c
ts
th
e
r
e
c
e
iv
e
r
ope
r
a
ti
ng
c
h
a
r
a
c
t
e
r
i
s
ti
c
(
R
O
C
)
a
n
a
ly
s
e
s
th
e
M
L
m
od
e
ls
a
c
r
os
s
s
ta
g
e
s
of
AD
m
il
d
de
m
e
nt
e
d
a
s
in
F
ig
ur
e
4(
a
)
,
m
od
e
r
a
t
e
de
m
e
nt
e
d
a
s
in
F
ig
ur
e
4(
b)
,
non
-
d
e
m
e
nt
e
d
a
s
in
F
i
gur
e
4(
c
)
,
a
nd
ve
r
y
m
il
d
de
m
e
nt
e
d
a
s
in
F
ig
ur
e
4
(
d)
.
F
a
ls
e
pos
it
iv
e
s
a
nd
ne
ga
ti
ve
s
w
e
r
e
e
qua
ll
y
w
e
ig
ht
e
d
(
500
e
a
c
h)
,
w
it
h
c
la
s
s
pr
oba
bi
li
ti
e
s
of
14%
,
1%
,
50%
,
a
nd
35%
.
I
n
e
a
r
ly
de
te
c
ti
on
in
th
e
m
il
d
a
nd
ve
r
y
m
il
d
de
m
e
nt
e
d
s
ta
ge
s
,
hi
gh
s
e
ns
it
iv
it
y
w
a
s
c
r
it
ic
a
l,
w
hi
le
f
a
ls
e
di
a
gno
s
e
s
ha
v
e
be
e
n
pr
e
ve
nt
e
d
in
th
e
non
-
de
m
e
nt
e
d
gr
oup
w
it
h
a
c
c
ur
a
te
c
la
s
s
if
ic
a
ti
on.
th
e
m
od
e
r
a
te
de
m
e
nt
e
d
c
la
s
s
r
e
q
ui
r
e
d
th
e
id
e
nt
if
ic
a
ti
on
of
r
a
r
e
in
s
ta
nc
e
s
w
it
h
li
tt
le
f
a
ls
e
pos
it
iv
e
s
.
O
ve
r
a
ll
,
R
O
C
r
e
s
ul
ts
hi
ghl
ig
ht
th
e
tr
a
de
-
of
f
be
twe
e
n
s
e
ns
it
iv
it
y
a
nd
s
pe
c
if
ic
it
y
a
c
r
os
s
c
la
s
s
e
s
. K
N
N
d
e
m
ons
tr
a
te
s
t
he
b
e
s
t
c
la
s
s
if
ic
a
ti
on f
or
m
il
d a
nd ve
r
y m
il
d de
m
e
nt
e
d
s
ta
ge
s
.
4.2. P
C
A
d
im
e
n
s
io
n
al
it
y r
e
d
u
c
t
io
n
P
C
A
te
c
hni
que
w
a
s
a
ppl
ie
d
to
de
e
p
f
e
a
tu
r
e
s
e
xt
r
a
c
te
d
f
r
om
pr
e
-
tr
a
in
e
d
S
que
e
z
e
N
e
t
to
r
e
duc
e
di
m
e
ns
io
na
li
ty
w
hi
le
r
e
ta
in
in
g
th
e
m
os
t
in
f
or
m
a
ti
ve
c
om
pone
nt
s
.
D
if
f
e
r
e
nt
va
lu
e
s
of
va
r
ia
nc
e
th
r
e
s
hol
ds
of
90%
,
95%
,
96
%
,
a
nd
99%
w
e
r
e
te
s
te
d.
V
a
r
ia
nc
e
of
96%
ha
s
r
e
ta
in
e
d
(
~
89
c
om
pone
nt
s
)
of
f
e
r
in
g
th
e
opt
im
a
l
ba
la
nc
e
be
twe
e
n a
c
c
ur
a
c
y of
t
he
c
la
s
s
if
ic
a
ti
on a
nd c
om
put
a
ti
on
a
l
e
f
f
ic
ie
nc
y.
4.2.1. P
e
r
f
or
m
an
c
e
c
o
m
p
ar
is
on
ac
r
o
s
s
P
C
A
var
ia
n
c
e
t
h
r
e
s
h
ol
d
s
F
ig
ur
e
5
il
lu
s
tr
a
te
s
th
e
c
um
ul
a
ti
ve
e
xpl
a
in
e
d
va
r
ia
nc
e
f
or
th
e
c
om
pone
nt
s
of
P
C
A
,
s
how
in
g
th
e
a
bi
li
ty
of
th
e
f
ir
s
t
pr
in
c
ip
a
l
c
om
pone
nt
s
to
pr
og
r
e
s
s
iv
e
ly
c
a
pt
ur
e
da
ta
va
r
ia
bi
li
ty
.
S
pe
c
if
ic
a
ll
y,
F
ig
u
r
e
5(
a
)
c
or
r
e
s
ponds
to
a
90%
va
r
ia
nc
e
th
r
e
s
hol
d,
F
ig
ur
e
5(
b)
to
95%
,
F
ig
ur
e
5(
c
)
to
96
%
,
a
nd
F
ig
ur
e
5(
d
)
to
99%
.
T
a
bl
e
2 s
how
s
t
he
e
v
a
lu
a
ti
on me
tr
ic
s
of
M
L
m
ode
ls
a
f
te
r
a
ppl
y
in
g P
C
A
.
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
D
e
e
p f
e
at
ur
e
-
bas
e
d m
ul
ti
-
c
la
s
s
A
lz
he
ime
r
’
s
di
s
e
as
e
c
la
s
s
if
ic
at
i
on w
it
h
…
(
M
ay
s
al
oon A
be
d Q
as
im
)
701
(
a
)
(
b)
(
c
)
(
d)
F
ig
ur
e
4. R
O
C
c
ur
ve
s
f
or
M
L
m
ode
ls
(
a
)
m
il
d de
m
e
nt
e
d, (
b)
m
ode
r
a
te
de
m
e
nt
e
d, (
c
)
non
-
de
m
e
nt
e
d, a
nd
(
d)
ve
r
y m
il
d de
m
e
nt
e
d
(
a
)
(
b)
(
c
)
(
d)
F
ig
ur
e
5. C
um
ul
a
ti
ve
e
xpl
a
in
e
d va
r
ia
nc
e
a
t
di
f
f
e
r
e
nt
t
hr
e
s
hol
ds
:
(
a
)
90%
, (
b)
95
%
, (
c
)
96
%
, a
nd (
d)
99%
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
, N
o.
1
,
F
e
br
ua
r
y 2026
:
695
-
706
702
T
a
bl
e
2. E
va
lu
a
ti
on
m
e
tr
ic
s
of
M
L
m
ode
ls
a
f
te
r
P
C
A
P
C
A
va
r
i
a
nc
e
M
ode
l
A
c
c
ur
a
c
y (
%
)
R
e
c
a
l
l
(
%
)
P
r
e
c
i
s
i
on (
%
)
F1
-
s
c
or
e
(
%
)
A
U
C
(
%
)
90%
(
~
43 c
om
pone
nt
s
)
KNN
54
54
54
54
63
NN
72
72
72
72
87
LR
60
60
59
58
76
NB
57
57
57
57
72
S
V
M
47
47
56
47
70
DT
54
54
54
54
54
95%
(
~
76 c
om
pone
nt
s
)
KNN
92
92
92
92
98
NN
77
77
77
77
91
LR
61
61
60
60
77
NB
55
55
57
55
72
S
V
M
47
47
56
47
69
DT
54
54
54
54
64
96%
(
~
89 c
om
pone
nt
s
)
KNN
93
93
93
93
99
NN
79
78
78
78
91
LR
62
62
61
61
78
NB
54
54
57
55
72
S
V
M
48
48
57
48
70
DT
53
53
53
53
62
99%
(
~
100 c
om
pone
nt
s
)
KNN
93
93
93
93
99
NN
79
79
78
78
92
LR
62
61
62
61
79
NB
54
54
56
57
72
S
V
M
47
47
57
45
70
DT
53
53
53
53
62
4.3. Cr
os
s
-
val
id
at
io
n
p
e
r
f
or
m
an
c
e
T
he
p
e
r
f
or
m
a
nc
e
of
m
ul
ti
pl
e
M
L
c
la
s
s
if
ie
r
s
w
a
s
a
s
s
e
s
s
e
d
us
in
g
s
tr
a
ti
f
ie
d
5
-
,
10
-
,
a
nd
20
-
f
ol
d
c
r
os
s
-
va
li
da
ti
on, without
a
ppl
yi
ng P
C
A
a
nd a
f
te
r
a
ppl
yi
ng i
t
to
m
a
in
ta
in
96%
t
he
va
r
ia
nc
e
.
W
it
hout
P
C
A
, t
he
hi
ghe
s
t
pe
r
f
or
m
a
nc
e
w
a
s
a
c
hi
e
ve
d
by
K
N
N
,
w
it
h
a
c
c
ur
a
c
y
of
97.05%
a
nd
A
U
C
of
98.99%
,
N
N
f
ol
lo
w
e
d
w
it
h
a
c
c
ur
a
c
y
of
86.83%
a
nd
A
U
C
of
96.24%
.
S
V
M
a
nd
N
B
de
m
ons
tr
a
te
d
poor
pe
r
f
or
m
a
nc
e
w
it
h
a
c
c
ur
a
c
y
a
r
ound 44%
a
nd A
U
C
a
r
ound 67%
, w
hi
le
D
T
a
nd L
R
s
h
a
r
e
d t
he
m
ode
r
a
te
pe
r
f
or
m
a
nc
e
.
A
f
te
r
a
ppl
yi
ng
P
C
A
,
th
e
pe
r
f
o
r
m
a
nc
e
of
K
N
N
im
pr
ove
d,
r
e
a
c
hi
ng
97.48%
a
c
c
ur
a
c
y
a
nd
99.84%
A
U
C
,
w
hi
c
h
in
di
c
a
te
s
th
a
t
r
e
m
ovi
ng
r
e
dunda
nt
f
e
a
tu
r
e
s
e
nha
nc
e
d
ge
ne
r
a
li
z
a
ti
on.
N
B
s
how
e
d
a
r
e
m
a
r
ka
bl
e
im
pr
ove
m
e
nt
,
w
hi
le
th
e
pe
r
f
or
m
a
nc
e
of
N
N
,
L
R
a
nd
D
T
s
l
ig
ht
ly
r
e
duc
e
d
due
to
th
e
lo
s
s
of
s
ubt
le
but
in
f
or
m
a
ti
ve
f
e
a
tu
r
e
s
dur
in
g
di
m
e
ns
io
na
li
ty
r
e
duc
ti
on.
T
a
bl
e
3
s
um
m
a
r
iz
e
s
th
e
e
va
lu
a
ti
on
m
e
tr
ic
s
a
c
r
os
s
a
ll
m
ode
ls
a
nd f
ol
ds
.
T
a
bl
e
3. E
va
lu
a
ti
on me
tr
ic
s
a
c
r
o
s
s
5
-
, 10
-
, a
nd 20
-
f
ol
d s
tr
a
ti
f
ie
d c
r
os
s
-
va
li
da
ti
on (
w
it
h a
nd w
it
hout
P
C
A
)
M
ode
l
F
ol
d
A
U
C
%
(
no
P
C
A
)
A
U
C
%
(
w
i
t
h
P
C
A
)
A
c
c
ur
a
c
y
%
(
no
P
C
A
)
A
c
c
ur
a
c
y
%
(
w
i
t
h
P
C
A
)
F1
-
s
c
or
e
%
(
no
P
C
A
)
F1
-
s
c
or
e
%
(
w
i
t
h
P
C
A
)
P
r
e
c
i
s
i
on
%
(
no
P
C
A
)
P
r
e
c
i
s
i
on
%
(
w
i
t
h
P
C
A
)
R
e
c
a
l
l
%
(
no
P
C
A
)
R
e
c
a
l
l
%
(
w
i
t
h
P
C
A
)
KNN
5
98.03
99.32
95.38
94.89
95.37
94.88
95.38
94.89
95.38
94.89
10
99.72
99.72
96.80
96.80
96.80
96.80
96.80
96.80
96.80
96.80
20
98.99
99.84
97.05
97.48
97.05
97.48
97.05
97.49
97.05
97.48
DT
5
64.19
61.11
56.38
52.33
56.33
52.27
56.29
52.27
56.38
52.33
10
62.34
62.34
52.70
52.70
52.74
52.74
52.84
52.84
52.70
52.70
20
63.81
62.34
56.72
53.08
56.59
52.88
56.52
52.74
56.72
53.08
S
V
M
5
66.73
70.83
43.25
48.30
44.63
47.16
52.05
57.20
43.25
48.30
10
71.01
71.01
47.84
47.84
46.54
46.54
57.40
57.40
47.84
47.84
20
67.51
70.10
44.06
47.59
45.67
46.29
54.10
57.29
44.06
47.59
NN
5
95.07
92.84
84.45
80.64
84.41
80.49
84.41
80.55
84.45
80.64
10
93.46
93.46
81.83
81.83
81.76
81.76
81.80
81.80
81.83
81.83
20
96.24
93.82
86.83
82.11
86.79
82.02
86.78
82.05
86.83
82.11
NB
5
68.26
72.79
43.09
55.06
45.02
56.03
55.06
57.97
43.09
55.06
10
71.01
73.59
47.84
56.33
46.54
57.06
57.40
58.76
47.84
56.33
20
68.65
73.90
44.09
57.03
45.82
57.61
55.16
59.13
44.09
57.03
LR
5
82.91
78.39
67.38
62.06
67.28
61.42
67.21
61.29
67.38
62.06
10
78.39
78.39
61.98
61.98
61.37
61.37
61.29
61.29
61.98
61.98
20
83.57
78.41
68.41
61.97
68.29
61.33
68.20
61.24
68.41
61.97
4.4. S
t
at
is
t
ic
al
s
ig
n
if
ic
an
c
e
an
al
ys
is
T
w
o
te
s
ts
w
e
r
e
pe
r
f
or
m
e
d:
P
a
ir
e
d
t
-
te
s
ts
a
nd
W
il
c
oxon
s
ig
ne
d
-
r
a
nk.
M
os
t
m
ode
ls
ha
ve
r
e
c
or
de
d
s
ta
ti
s
ti
c
a
ll
y
in
s
ig
ni
f
ic
a
nt
di
f
f
e
r
e
nc
e
s
be
f
or
e
a
nd
a
f
te
r
P
C
A
(
p>
0.05)
,
e
xc
e
pt
N
B
,
s
how
in
g s
ig
ni
f
ic
a
nt
in
c
r
e
a
s
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
D
e
e
p f
e
at
ur
e
-
bas
e
d m
ul
ti
-
c
la
s
s
A
lz
he
ime
r
’
s
di
s
e
as
e
c
la
s
s
if
ic
at
i
on w
it
h
…
(
M
ay
s
al
oon A
be
d Q
as
im
)
703
in
a
c
c
ur
a
c
y
(
p=
0.0144)
in
th
e
t
-
te
s
t.
T
hi
s
in
di
c
a
te
s
th
a
t
P
C
A
is
pa
r
ti
c
ul
a
r
ly
us
e
f
ul
f
or
pr
oba
bi
li
s
ti
c
c
la
s
s
if
ie
r
s
.
T
a
bl
e
4 i
ll
us
tr
a
te
s
t
h
e
r
e
s
ul
ts
obt
a
in
e
d w
it
h
a
nd w
it
hout
P
C
A
.
T
a
bl
e
4. S
ta
ti
s
ti
c
a
l
te
s
t
r
e
s
ul
ts
f
or
m
e
a
n a
c
c
ur
a
c
y w
it
h a
nd w
it
h
out
P
C
A
M
ode
l
M
e
a
n a
c
c
ur
a
c
y
wi
t
hout
P
C
A
(
%
)
M
e
a
n a
c
c
ur
a
c
y
w
i
t
h P
C
A
(
%
)
P
a
i
r
e
d t
s
t
a
t
i
s
t
i
c
P
a
i
r
e
d t
p
-
va
l
ue
W
i
l
c
oxon
W
i
l
c
oxon p
-
va
l
ue
KNN
96.41
96.39
-
0.075
0.9469
1.000
0.6547
DT
55.27
52.70
-
1.992
0.1847
0.000
0.1797
S
V
M
45.05
47.91
1.912
0.1960
0.000
0.1797
NN
84.37
81.53
-
1.967
0.1881
0.000
0.1797
NB
45.01
56.14
8.241
0.0144
0.000
0.2500
LR
65.92
62.00
-
1.973
0.1872
0.000
0.1797
4.5. Con
f
u
s
io
n
m
at
r
ix
an
d
c
al
ib
r
at
io
n
an
al
ys
is
K
N
N
’
s
pe
r
f
or
m
a
nc
e
a
f
te
r
P
C
A
w
it
h
(
20
-
f
ol
d
C
V
)
w
a
s
n
e
a
r
-
pe
r
f
e
c
t
w
it
h
a
c
c
ur
a
c
y=
97.48%
,
A
U
C
=
0.998.
F
ig
ur
e
6
s
how
s
th
e
c
a
li
br
a
ti
on
pl
ot
.
I
t
s
how
s
th
a
t
pr
e
di
c
te
d
pr
oba
bi
li
ti
e
s
a
li
gn
w
e
ll
w
it
h
obs
e
r
ve
d outc
om
e
s
,
c
onf
ir
m
in
g t
ha
t
th
e
m
ode
l
is
w
e
ll
-
c
a
li
br
a
te
d a
nd c
li
ni
c
a
ll
y i
nt
e
r
pr
e
ta
bl
e
.
4.6. P
e
r
f
or
m
an
c
e
c
o
m
p
ar
is
on
ac
r
o
s
s
d
at
as
e
t
s
T
o
c
he
c
k
th
e
ge
n
e
r
a
li
z
a
ti
on
of
s
ys
te
m
,
a
la
r
ge
da
t
a
s
e
t
c
ont
a
in
i
ng
44,000
s
a
m
pl
e
s
ha
s
be
e
n
a
ppl
ie
d
a
c
r
os
s
f
our
de
m
e
nt
ia
s
ta
ge
s
, K
N
N
a
c
hi
e
ve
d t
he
be
s
t
pe
r
f
or
m
a
nc
e
a
ga
in
a
s
s
how
n i
n T
a
bl
e
5
[
34]
, [
35
]
.
W
he
n
us
in
g
la
r
ge
r
da
ta
s
e
t,
K
N
N
r
e
ta
in
e
d
th
e
e
xc
e
ll
e
nt
pe
r
f
or
m
a
nc
e
w
it
h
onl
y
a
m
in
or
r
e
duc
ti
on
due
to
in
c
r
e
a
s
e
d
da
ta
di
ve
r
s
it
y.
F
ig
ur
e
7
pr
e
s
e
nt
s
th
e
p
e
r
f
or
m
a
nc
e
e
va
lu
a
ti
on
o
f
th
e
pr
opos
e
d
S
que
e
z
e
N
e
t
+
K
N
N
f
r
a
m
e
w
or
k
on
th
e
ne
w
44,000
-
im
a
ge
da
t
a
s
e
t.
T
he
c
onf
us
io
n
m
a
tr
ix
in
F
ig
ur
e
7
(
a
)
de
pi
c
ts
a
li
tt
le
ove
r
la
p
b
e
twe
e
n
m
il
d
a
nd
ve
r
y
m
il
d
de
m
e
nt
e
d
gr
ou
ps
,
w
hi
le
th
e
R
O
C
c
ur
ve
in
F
ig
ur
e
7
(
b
)
c
onf
ir
m
s
th
e
s
tr
ong
di
s
c
r
im
in
a
ti
ve
pow
e
r
a
c
r
os
s
a
ll
s
ta
ge
s
.
T
h
e
s
e
r
e
s
ul
ts
hi
ghl
ig
ht
e
d
th
a
t
th
e
pr
opos
e
d
S
que
e
z
e
N
e
t+
K
N
N
f
r
a
m
e
w
or
k
is
s
c
a
la
bl
e
a
nd r
obus
t
f
or
t
he
c
la
s
s
if
ic
a
ti
on of
A
D
.
F
ig
ur
e
6.
K
NN
pe
r
f
or
m
a
nc
e
a
f
te
r
P
C
A
:
c
a
li
br
a
ti
on plot
T
a
bl
e
5. C
om
pa
r
is
on of
be
s
t
m
ode
l
pe
r
f
or
m
a
nc
e
be
twe
e
n or
ig
i
na
l
a
nd l
a
r
ge
r
da
ta
s
e
t
D
a
t
a
s
e
t
B
e
s
t
m
ode
l
A
U
C
(
%
)
A
c
c
ur
a
c
y (
%
)
F1
-
s
c
or
e
(
%
)
P
r
e
c
i
s
i
on (
%
)
R
e
c
a
l
l
(
%
)
O
r
i
gi
na
l
da
t
a
s
e
t
(
6,400 i
m
a
ge
s
,
a
f
t
e
r
P
C
A
, 20
-
f
ol
d C
V
)
[
23]
KNN
99.84
97.48
97.48
97.49
97.48
N
e
w
da
t
a
s
e
t
(
44,000 i
m
a
ge
s
,
a
f
t
e
r
P
C
A
,
20
-
f
ol
d C
V
)
[
35]
KNN
98.74
94.78
94.75
94.79
94.78
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
, N
o.
1
,
F
e
br
ua
r
y 2026
:
695
-
706
704
(
a
)
(
b)
F
ig
ur
e
7. P
e
r
f
or
m
a
nc
e
of
K
N
N
on t
he
ne
w
44,000
-
im
a
ge
da
ta
s
e
t:
(
a
)
c
onf
us
io
n m
a
tr
ix
a
nd (
b)
R
O
C
c
ur
ve
5.
C
O
N
C
L
U
S
I
O
N
T
he
r
e
s
ul
t
s
obt
a
in
e
d
f
r
om
th
is
s
tu
dy
il
lu
s
tr
a
te
th
a
t
in
te
gr
a
ti
ng
de
e
p
f
e
a
tu
r
e
s
f
r
om
a
pr
e
-
tr
a
in
e
d
S
que
e
z
e
N
e
t
w
it
h
tr
a
di
ti
ona
l
M
L
c
la
s
s
if
ie
r
s
c
a
n
pr
e
di
c
t
A
D
s
t
a
ge
s
e
f
f
e
c
ti
ve
ly
.
K
N
N
ha
s
a
c
hi
e
ve
d
th
e
be
s
t
pe
r
f
or
m
a
nc
e
c
ons
is
te
nt
ly
,
w
it
h
97.48%
a
c
c
ur
a
c
y
a
nd
0.998
A
U
C
w
he
n
u
s
in
g
th
e
or
ig
in
a
l
da
ta
s
e
t
c
on
s
is
ti
ng
of
6,400
im
a
ge
s
,
a
nd
w
it
h
94.78%
a
c
c
ur
a
c
y
w
it
h
98.74%
A
U
C
w
he
n
us
in
g
ne
w
la
r
ge
r
da
ta
s
e
t
(
44,000
im
a
ge
s
)
.
C
onf
us
io
n
m
a
tr
ix
a
nd
R
O
C
a
na
ly
s
e
s
a
ls
o
c
onf
ir
m
e
d
gr
e
a
t
di
s
c
r
im
in
a
ti
on,
w
it
h
m
in
or
e
a
r
ly
-
s
ta
ge
m
is
c
la
s
s
if
ic
a
ti
on
s
. U
s
in
g P
C
A
i
m
pr
ove
d t
he
e
f
f
ic
ie
nc
y a
nd f
e
a
tu
r
e
di
s
c
r
im
in
a
ti
on, pa
r
ti
c
ul
a
r
ly
f
or
K
NN.
F
U
N
D
I
N
G
I
N
F
O
R
M
A
T
I
O
N
A
ut
hor
s
s
ta
te
no f
undi
ng i
nvol
ve
d.
A
U
T
H
O
R
C
O
N
T
R
I
B
U
T
I
O
N
S
S
T
A
T
E
M
E
N
T
T
hi
s
jo
ur
na
l
us
e
s
th
e
C
ont
r
ib
ut
or
R
ol
e
s
T
a
xonomy
(
C
R
e
di
T
)
to
r
e
c
ogni
z
e
in
di
vi
dua
l
a
ut
hor
c
ont
r
ib
ut
io
ns
, r
e
duc
e
a
ut
hor
s
hi
p di
s
put
e
s
,
a
nd f
a
c
il
it
a
te
c
ol
la
bo
r
a
ti
on.
N
am
e
o
f
A
u
t
h
or
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
M
a
ys
a
l
o
on
A
be
d
Q
a
s
i
m
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
M
a
r
w
a
M
a
w
f
a
q
M
o
h
a
m
e
d
s
h
e
e
t
A
l
-
H
a
t
a
b
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
L
uba
b H
.
A
lb
a
k
✓
✓
✓
✓
✓
✓
✓
Evaluation Warning : The document was created with Spire.PDF for Python.