I
A
E
S
I
n
t
e
r
n
at
i
on
al
Jou
r
n
al
o
f
A
r
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
ge
n
c
e
(
I
J
-
AI
)
V
ol
. 15, N
o. 1, F
e
br
ua
r
y 2026
, pp.
681
~
694
I
S
S
N
:
2252
-
8938
,
D
O
I
:
10.11591/
i
j
a
i
.v
15
.i
1
.pp
681
-
694
681
Jou
r
n
al
h
om
e
page
:
h
t
t
p
:
/
/
i
j
a
i
.i
ae
s
c
or
e
.c
om
C
om
p
ar
i
s
on
b
e
t
w
e
e
n
e
n
s
e
m
b
l
e
an
d
l
i
n
e
ar
m
e
t
h
od
s
f
or
w
e
b
s
i
t
e
p
h
i
s
h
i
n
g
d
e
t
e
c
t
i
on
S
ab
a
H
u
s
s
e
i
n
R
as
h
i
d
,
S
ab
a
A
l
aa
A
b
d
u
l
w
ah
h
ab
,
F
ar
ah
A
m
e
r
A
b
d
u
l
az
i
z
D
e
pa
r
t
m
e
nt
of
C
om
pu
t
e
r
S
c
i
e
nc
e
,
C
ol
l
e
ge
of
C
om
put
e
r
S
c
i
e
nc
e
a
nd
M
a
t
he
m
a
t
i
c
s
,
T
i
k
r
i
t
U
ni
ve
r
s
i
t
y,
T
i
k
r
i
t
,
I
r
a
q
A
r
t
i
c
l
e
I
n
f
o
A
B
S
T
R
A
C
T
A
r
t
i
c
l
e
h
i
s
t
o
r
y
:
R
e
c
e
i
ve
d
O
c
t
11, 2024
R
e
vi
s
e
d
J
a
n 4, 2026
A
c
c
e
pt
e
d
J
a
n 22, 2026
In
th
e
c
u
rre
nt
digi
t
alized
world,
th
e
n
o
t
io
n
of
cybersec
u
ri
t
y
h
as
beco
m
e
cru
cial
in
everyday
life
,
a
n
d
th
e
iss
u
e
of
privacy
t
akes
th
e
leadi
n
g
role
in
th
e
t
ec
hn
ological
age
n
da
of
th
e
global
co
mmun
i
t
y.
O
n
e
s
u
c
h
social
e
n
gi
n
eeri
n
g
att
ack
th
a
t
is
c
u
rre
nt
ly
prevale
nt
is
p
h
is
h
i
n
g,
w
h
ic
h
is
a
co
mm
o
n
t
ec
hn
iq
u
e
u
sed
by
cybercri
m
i
n
als
to
i
nt
ercep
t
se
n
si
t
ive
da
t
a.
Despi
t
e
th
e
prese
n
ce
of
cert
ai
n
li
m
i
t
a
t
io
n
s
w
h
ic
h
ca
n
res
t
ric
t
i
t
s
u
sef
u
l
n
ess,
m
ac
h
i
n
e
lear
n
i
n
g
(ML)
h
as
evolved
i
nt
o
an
i
nt
eres
t
i
n
g
approac
h
to
ide
nt
ify
p
h
is
h
i
n
g
a
tt
acks
.
Clo
u
d
ML
is
an
effec
t
ive
sol
ut
io
n
th
a
t
u
ses
clo
u
d
co
m
p
ut
i
n
g
sol
ut
io
n
s
to
crea
t
e
,
t
rai
n,
a
n
d
deploy
m
odels
th
a
t
provide
a
fas
t
er
a
n
d
m
ore
acc
u
ra
t
e
res
u
l
t
as
well
as
s
u
ppor
t
large
da
t
ase
t
s.
T
h
is
paper
co
m
pares
th
e
e
n
se
m
ble
m
e
th
od
of
Am
azo
n
SageMa
ker’s
A
ut
oML
t
ool,
A
ut
oGl
u
o
n,
wi
th
th
e
li
n
ear
m
e
th
od
of
SageMa
ker’s
lin
ear
lear
n
er
algori
thm
for
websi
t
e
p
h
is
h
i
n
g
de
t
ec
t
io
n.
Key
fact
ors
exa
m
i
n
ed
i
n
cl
u
de
t
rai
n
i
n
g
tec
hn
iq
u
es,
t
rai
n
i
n
g
t
i
m
e,
ba
t
c
h
t
ra
n
sfor
m
t
i
m
e
,
e
n
dpoi
nt
predic
t
io
n
t
i
m
e,
a
n
d
m
odel
acc
u
racy
.
T
h
e
res
u
l
t
s
de
m
o
n
s
t
ra
t
e
th
a
t
w
h
ile
A
ut
oGl
u
o
n
o
ut
perfor
m
s
li
n
ear
lear
n
er
in
t
er
m
s
of
acc
u
racy
a
n
d
predict
io
n
speed
,
li
n
ear
lear
n
er
is
fas
t
er
in
t
rai
n
i
n
g
a
n
d
ba
t
c
h
t
ra
n
sfor
m
processes.
K
e
y
w
or
d
s
:
A
ut
oG
l
uon
C
l
oud M
L
E
ns
e
m
bl
e
m
ode
l
L
i
ne
a
r
l
e
a
r
ne
r
W
e
bs
i
t
e
ph
i
s
h
i
ng
This
is
an
open
access
article
under
the
CC
BY
-
SA
license
.
C
or
r
e
s
pon
di
n
g
A
u
t
h
or
:
S
a
ba
H
us
s
e
i
n
R
a
s
hi
d
D
e
pa
r
t
m
e
n
t
of
C
om
put
e
r
S
c
i
e
nc
e
,
C
ol
l
e
g
e
of
C
om
put
e
r
S
c
i
e
n
c
e
a
nd
M
a
t
he
m
a
t
i
c
s
,
T
i
kr
i
t
U
ni
ve
r
s
i
t
y
S
t
r
e
e
t
of
T
i
kr
i
t
-
M
os
i
l
,
Al
-
Q
a
di
s
s
i
ya
h
Q
ua
r
t
e
r
009642
,
T
i
k
r
i
t
,
S
a
l
a
he
dd
i
n,
I
r
a
q
E
m
a
i
l
:
s
a
ba
hus
s
e
i
n88@
t
u.e
du.
i
q
1.
I
N
T
R
O
D
U
C
T
I
O
N
W
i
t
h
p
r
i
va
c
y
as
a
m
a
j
o
r
i
s
s
ue
,
c
yb
e
r
s
e
c
u
r
i
t
y
is
e
s
s
e
nt
i
a
l
in
t
oda
y
’
s
e
nv
i
r
onm
e
nt
[
1]
.
C
ybe
r
c
r
i
m
i
n
a
l
s
us
e
phi
s
hi
ng,
a
w
e
l
l
-
know
n
s
oc
i
a
l
e
ngi
ne
e
r
i
ng
a
t
t
a
c
k,
to
obt
a
i
n
pe
r
s
ona
l
i
nf
or
m
a
t
i
on
[
2]
,
as
a
t
t
a
c
ke
r
s
u
t
i
l
i
z
e
e
-
c
om
m
e
r
c
e
w
e
bs
i
t
e
s
,
ba
nks
,
a
nd
c
r
e
di
t
c
a
r
d
c
om
pa
n
i
e
s
as
t
he
i
r
gu
i
s
e
s
to
t
r
i
c
k
p
e
opl
e
i
nt
o
s
e
ndi
ng
s
e
ns
i
t
i
ve
da
t
a
[
3]
.
P
hi
s
h
i
ng
is
de
f
i
ne
d
as
“a
c
r
i
m
i
na
l
m
e
c
ha
ni
s
m
us
i
ng
bo
t
h
t
e
c
hn
i
c
a
l
s
ub
t
e
r
f
ug
e
as
w
e
l
l
as
s
oc
i
a
l
e
ngi
ne
e
r
i
ng
f
or
s
t
e
a
l
i
ng
pe
r
s
ona
l
i
de
n
t
i
t
y
da
t
a
a
nd
f
i
na
nc
i
a
l
a
c
c
oun
t
c
r
e
de
n
t
i
a
l
s
of
c
ons
u
m
e
r
s
”
by
A
nt
i
-
P
hi
s
h
i
ng
W
or
k
i
ng
G
r
oup
(
A
P
W
G
)
,
w
i
t
h
1,003,924
phi
s
hi
ng
a
t
t
a
c
ks
r
e
por
t
e
d
gl
ob
a
l
l
y
in
t
he
f
i
r
s
t
qua
r
t
e
r
of
2025.
N
um
e
r
ous
t
e
c
hni
que
s
w
e
r
e
pu
t
f
or
t
h
to
i
de
n
t
i
f
y
a
nd
c
l
a
s
s
i
f
y
ph
i
s
h
i
ng
a
t
t
e
m
pt
s
,
s
uc
h
as
bl
a
c
kl
i
s
t
m
e
t
hod
a
nd
he
ur
i
s
t
i
c
a
na
l
ys
i
s
of
t
he
s
our
c
e
c
ode
.
Y
e
t
,
t
he
he
ur
i
s
t
i
c
a
pp
r
oa
c
h
is
d
i
f
f
i
c
ul
t
to
us
e
,
c
l
a
s
s
i
f
i
e
r
de
ve
l
opm
e
n
t
t
a
ke
s
t
i
m
e
,
a
nd
t
he
bl
a
c
k
l
i
s
t
ha
s
t
r
oubl
e
i
d
e
nt
i
f
yi
ng
i
n
t
e
r
m
i
t
t
e
n
t
phi
s
h
i
ng
s
i
t
e
s
[
4]
.
T
hr
ough
e
xa
m
i
n
i
ng
w
e
bs
i
t
e
pa
t
t
e
r
ns
a
nd
f
e
a
t
ur
e
s
,
m
a
c
hi
ne
l
e
a
r
n
i
ng
(
M
L
)
is
good
in
i
de
n
t
i
f
yi
ng
phi
s
h
i
ng
a
t
t
a
c
ks
a
nd
he
l
ps
di
s
t
i
ngui
s
h
be
t
w
e
e
n
m
a
l
i
c
i
ous
a
nd
t
r
us
t
w
or
t
hy
w
e
bs
i
t
e
s
.
C
om
pa
r
e
d
to
c
onve
nt
i
ona
l
r
ul
e
-
ba
s
e
d
a
l
gor
i
t
hm
s
,
it
c
oul
d
de
t
e
c
t
m
i
no
r
phi
s
hi
ng
i
ndi
c
a
t
or
s
s
uc
h
as
s
us
p
i
c
i
ous
uni
f
or
m
r
e
s
our
c
e
l
o
c
a
t
or
s
(
U
R
L
s
)
a
nd
dom
a
i
n
a
ge
w
i
t
h
m
o
r
e
a
c
c
ur
a
c
y.
T
he
s
e
m
e
t
hods
m
i
ght
be
ha
m
pe
r
e
d
by
i
s
s
ue
s
in
ha
nd
l
i
ng
unba
l
a
nc
e
d
da
t
a
s
e
t
s
,
a
d
j
us
t
i
ng
to
nove
l
a
t
t
a
c
k
s
t
r
a
t
e
gi
e
s
,
a
nd
gua
r
a
n
t
e
e
i
ng
m
ode
l
ge
n
e
r
a
l
i
z
a
t
i
on
a
c
r
os
s
m
a
ny
w
e
bs
i
t
e
s
[
5]
.
To
de
ve
l
op,
t
r
a
i
n,
a
nd
de
p
l
oy
m
ode
l
s
f
o
r
f
a
s
t
,
a
c
c
ur
a
t
e
r
e
s
ul
t
s
a
nd
m
a
na
g
i
ng
huge
da
t
a
,
c
l
oud
ML
m
a
ke
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
t
i
f
I
nt
e
l
l
,
V
ol
. 15, N
o. 1, F
e
br
ua
r
y 2026
:
681
-
694
682
a
dva
nt
a
ge
of
c
l
oud
c
om
put
i
ng
s
e
r
v
i
c
e
s
.
T
hi
s
ki
nd
of
s
e
r
vi
c
e
ha
s
num
e
r
ou
s
be
ne
f
i
t
s
i
nc
l
ud
i
ng
s
c
a
l
a
b
i
l
i
t
y,
c
os
t
-
e
f
f
e
c
t
i
ve
ne
s
s
,
a
nd
s
i
m
p
l
i
c
i
t
y
of
us
e
.
It
m
a
ke
s
us
e
of
e
ns
e
m
bl
e
as
w
e
l
l
as
l
i
ne
a
r
ML
t
e
c
hn
i
que
s
,
w
hi
c
h
c
om
bi
ne
s
e
ve
r
a
l
m
od
e
l
s
to
c
r
e
a
t
e
m
or
e
r
e
l
i
a
b
l
e
pr
e
d
i
c
t
i
on
m
ode
l
.
T
he
di
ve
r
s
i
t
y
of
m
ode
l
s
is
l
e
ve
r
a
ge
d
by
e
ns
e
m
bl
e
t
e
c
hn
i
que
s
to
i
m
p
r
ove
e
r
r
or
r
e
duc
t
i
on
as
w
e
l
l
as
ge
ne
r
a
l
i
z
a
t
i
on
ove
r
s
i
ng
l
e
m
od
e
l
s
[
6]
.
A
c
l
oud
ML
m
ode
l
c
a
l
l
e
d
A
ut
oM
L
s
t
r
e
a
m
l
i
ne
s
t
he
pr
oc
e
s
s
of
c
hoos
i
ng,
c
r
e
a
t
i
ng,
a
nd
i
m
p
r
ovi
ng
ML
m
ode
l
s
a
ut
o
m
a
t
i
c
a
l
l
y,
m
a
k
i
ng
it
e
a
s
i
e
r
f
or
non
-
e
xpe
r
t
s
to
c
om
p
l
e
t
e
a
c
t
i
vi
t
i
e
s
,
s
uc
h
as
f
e
a
t
ur
e
e
ngi
ne
e
r
i
ng,
da
t
a
pr
e
p
r
oc
e
s
s
i
ng,
hype
r
p
a
r
a
m
e
t
e
r
t
un
i
ng,
a
l
gor
i
t
hm
s
e
l
e
c
t
i
on,
a
nd
m
ode
l
e
va
l
ua
t
i
on,
w
hi
c
h
in
t
ur
n
c
ut
s
dow
n
on
t
he
a
m
ount
of
t
i
m
e
s
pe
c
i
a
l
i
s
t
s
ne
e
d
to
c
ons
t
r
uc
t
h
i
gh
-
qua
l
i
t
y
m
ode
l
s
[
7
]
.
A
m
a
z
on
c
r
e
a
t
e
d
t
he
ope
n
-
s
ou
r
c
e
A
ut
oM
L
t
ool
k
i
t
A
ut
oG
l
uon,
w
hi
c
h
is
bui
l
t
on
P
yt
hon.
A
ut
om
a
t
i
c
f
e
a
t
u
r
e
da
t
a
t
ype
a
na
l
y
s
i
s
,
l
ow
pr
e
di
c
t
i
ve
a
t
t
r
i
but
e
di
s
c
a
r
di
ng,
ha
ndl
i
ng
of
m
i
s
s
i
ng
va
l
ue
s
,
r
a
w
da
t
a
p
r
e
pr
oc
e
s
s
i
ng,
a
nd
da
t
a
s
e
pa
r
a
t
i
on
i
nt
o
t
r
a
i
ni
ng
a
nd
va
l
i
da
t
i
on
s
e
t
s
a
r
e
a
l
l
done
by
i
t
.
W
i
t
h
t
he
us
e
of
r
e
pe
a
t
e
d
k
-
f
ol
d
ba
ggi
ng
f
or
pr
e
ve
n
t
i
ng
ove
r
f
i
t
t
i
ng,
it
t
r
a
i
ns
a
va
r
i
e
t
y
of
m
od
e
l
s
,
i
nc
l
ud
i
ng
k
-
ne
a
r
e
s
t
ne
i
ghbo
r
s
(
KNN
),
e
xt
r
e
m
e
g
r
a
di
e
n
t
boos
t
i
ng
(
X
G
B
oos
t
)
,
r
a
ndom
f
o
r
e
s
t
s
(
R
F
)
,
c
a
t
e
gor
i
c
a
l
boo
s
t
i
ng
(
C
a
t
B
oos
t
)
,
l
i
ght
g
r
a
di
e
n
t
boos
t
i
ng
m
a
c
hi
ne
(
L
i
ght
G
B
M
)
,
e
xt
r
e
m
e
l
y
r
a
ndo
m
i
z
e
d
t
r
e
e
s
(
E
xt
r
a
T
r
e
e
s
)
,
a
nd
ne
u
r
a
l
ne
t
w
or
ks
[
8
]
.
In
c
l
oud
M
L
,
l
i
ne
a
r
t
e
c
hni
que
s
de
pi
c
t
t
he
r
e
l
a
t
i
on
b
e
t
w
e
e
n
i
npu
t
pa
r
a
m
e
t
e
r
s
a
nd
p
r
e
di
c
t
e
d
out
pu
t
s
us
i
ng
a
l
i
ne
a
r
f
unc
t
i
on.
T
hos
e
m
e
t
hods
can
ha
nd
l
e
huge
da
t
a
s
e
t
s
as
w
e
l
l
as
r
e
a
l
-
t
i
m
e
a
pp
l
i
c
a
t
i
ons
w
i
t
h
hi
gh
-
di
m
e
ns
i
ona
l
da
t
a
s
i
nc
e
t
he
y
pr
e
s
um
e
t
he
t
a
r
ge
t
pa
r
a
m
e
t
e
r
is
a
l
i
ne
a
r
c
om
b
i
na
t
i
on
of
i
nput
f
e
a
t
ur
e
s
[
9]
.
W
i
t
h
t
h
e
us
e
of
a
l
i
ne
a
r
f
unc
t
i
on
f
or
r
e
p
r
e
s
e
n
t
i
ng
i
nput
pa
r
a
m
e
t
e
r
s
as
w
e
l
l
as
t
a
r
ge
t
va
r
i
a
bl
e
s
in
da
t
a
s
e
t
s
w
i
t
h
l
i
ne
a
r
r
e
l
a
t
i
ons
,
A
m
a
z
on
Web
S
e
r
vi
c
e
s
(
A
W
S
)
S
a
ge
M
a
ke
r
pr
ovi
de
s
l
i
ne
a
r
l
e
a
r
ne
r
,
a
s
upe
r
v
i
s
e
d
ML
t
e
c
hni
que
a
ppr
op
r
i
a
t
e
f
or
c
l
a
s
s
i
f
i
c
a
t
i
on
a
nd
r
e
gr
e
s
s
i
on
a
ppl
i
c
a
t
i
ons
[
10]
.
A
ut
oG
l
uon
a
nd
S
a
ge
M
a
ke
r
’
s
l
i
n
e
a
r
l
e
a
r
ne
r
ha
ve
be
e
n
e
xt
e
ns
i
ve
l
y
s
t
udi
e
d
in
t
h
e
pa
s
t
f
or
a
va
r
i
e
t
y
of
ML
a
ppl
i
c
a
t
i
ons
,
i
nc
l
udi
ng
t
i
m
e
s
e
r
i
e
s
f
or
e
c
a
s
t
i
ng,
c
l
a
s
s
i
f
i
c
a
t
i
on,
a
nd
r
e
gr
e
s
s
i
on
[
11]
−
[
21]
.
Y
e
t
,
t
he
r
e
is
a
c
l
e
a
r
l
a
c
k
of
r
e
s
e
a
r
c
h
on
t
he
a
pp
l
i
c
a
t
i
on
of
s
uc
h
m
e
t
hods
pa
r
t
i
c
ul
a
r
l
y
in
on
l
i
ne
w
e
bs
i
t
e
ph
i
s
hi
ng
de
t
e
c
t
i
on.
H
ow
e
ve
r
,
t
he
a
l
gor
i
t
h
m
s
i
nc
o
r
por
a
t
e
d
w
i
t
h
i
n
t
he
A
ut
oG
l
uon
a
r
c
h
i
t
e
c
t
ur
e
ha
ve
be
e
n
s
e
pa
r
a
t
e
l
y
i
nve
s
t
i
ga
t
e
d
f
or
w
e
bs
i
t
e
phi
s
h
i
ng
de
t
e
c
t
i
on,
e
s
t
a
bl
i
s
hi
ng
a
r
obus
t
ba
s
e
l
i
n
e
f
or
j
us
t
i
f
y
i
ng
t
he
c
ur
r
e
n
t
c
om
pa
r
i
s
on
[
22]
−
[
29]
.
In
t
hi
s
s
t
udy,
A
ut
oG
l
uon
is
c
om
pa
r
e
d
w
i
t
h
l
i
ne
a
r
l
e
a
r
ne
r
,
e
m
pha
s
i
z
i
ng
f
unda
m
e
nt
a
l
e
l
e
m
e
nt
s
s
u
c
h
as
t
r
a
i
ni
ng
t
i
m
e
,
t
r
a
i
ni
ng
m
e
t
hods
,
e
ndpoi
n
t
pr
e
di
c
t
i
on
t
i
m
e
,
a
nd
a
c
c
u
r
a
c
y.
M
or
e
ove
r
,
a
l
a
r
ge
da
t
a
s
e
t
of
11,430
pr
e
pr
oc
e
s
s
e
d
U
R
L
s
a
m
pl
e
s
m
a
i
nt
a
i
ne
d
on
A
m
a
z
on
S
i
m
p
l
e
S
t
or
a
ge
S
e
r
vi
c
e
(
S
3
)
is
us
e
d
to
m
a
ke
t
he
t
r
a
de
-
of
f
be
t
w
e
e
n
a
c
c
ur
a
c
y
a
nd
t
he
ne
e
d
f
or
s
pe
e
dy,
e
f
f
i
c
i
e
n
t
pr
oc
e
s
s
i
ng,
w
hi
c
h
a
l
l
ow
s
f
or
t
he
s
e
l
e
c
t
i
on
of
one
of
s
uc
h
m
ode
l
s
.
A
m
a
z
on
C
l
oudW
a
t
c
h
is
us
e
d
to
f
o
l
l
ow
up
t
he
pe
r
f
or
m
a
nc
e
,
w
hi
l
e
A
m
a
z
on
S
a
ge
M
a
ke
r
is
us
e
d
to
bui
l
d,
t
r
a
i
n,
a
nd
de
pl
oy
t
he
m
ode
l
.
T
he
vi
r
t
ua
l
r
e
s
our
c
e
s
r
e
qui
r
e
d
f
or
m
ode
l
de
ve
l
op
m
e
nt
a
r
e
a
c
qui
r
e
d
t
hr
ough
A
m
a
z
on
E
l
a
s
t
i
c
C
om
put
e
C
l
oud
(
E
C
2)
i
ns
t
a
nc
e
s
.
2.
M
E
T
H
O
D
T
hi
s
s
t
udy
e
xhi
bi
t
s
a
c
om
pa
r
i
s
on
of
t
w
o
A
W
S
S
a
ge
M
a
ke
r
f
r
a
m
e
w
or
ks
de
s
i
gne
d
to
de
t
e
c
t
w
e
bs
i
t
e
phi
s
hi
ng
by
e
m
p
l
oyi
ng
t
w
o
di
s
s
i
m
i
l
a
r
ye
t
c
om
pl
e
m
e
n
t
a
r
y
ML
c
onc
e
pt
s
w
i
t
hi
n
t
he
S
a
ge
M
a
k
e
r
c
l
oud
pl
a
t
f
o
r
m
.
T
he
p
r
oc
e
s
s
is
c
o
m
pr
i
s
e
d
of
t
he
s
ubs
e
que
n
t
s
t
a
ge
s
:
i
n
t
h
e
f
i
r
s
t
s
t
a
ge
,
t
he
A
W
S
S
a
ge
M
a
ke
r
s
e
t
t
i
ngs
w
a
s
c
onf
i
gu
r
e
d
a
nd
s
e
t
a
c
c
or
d
i
ng
to
t
h
e
r
e
qu
i
r
e
m
e
nt
s
of
t
he
pr
opos
e
d
m
ode
l
,
in
t
he
s
e
c
ond
s
t
a
ge
,
t
he
da
t
a
s
e
t
w
a
s
dow
nl
oa
de
d
f
r
om
t
he
s
our
c
e
,
m
a
nua
l
l
y
c
onf
i
gur
e
d,
a
nd
upl
oa
d
e
d
to
t
he
S3
buc
ke
t
f
o
r
f
ur
t
he
r
pr
oc
e
s
s
i
ng,
t
he
t
h
i
r
d
s
t
a
g
e
i
nvol
ve
d
da
t
a
s
e
t
p
r
e
pr
oc
e
s
s
i
ng,
w
hi
c
h
w
a
s
c
onduc
t
e
d
us
i
ng
a
S
a
ge
M
a
ke
r
P
y
t
hon
S
c
r
i
pt
s
pe
c
i
f
i
c
a
l
l
y
w
r
i
t
t
e
n
f
o
r
t
h
e
da
t
a
s
e
t
a
nd
i
nt
e
g
r
a
t
e
d
in
t
he
no
t
e
book,
t
he
r
e
s
ul
t
s
of
t
hi
s
s
t
a
g
e
w
a
s
s
a
ve
d
i
nt
o
S
3.
In
t
he
f
our
t
h
s
t
a
ge
,
t
he
t
r
a
i
ni
ng
of
bo
t
h
A
ut
oG
l
uon
a
nd
l
i
ne
a
r
l
e
a
r
ne
r
w
a
s
c
onduc
t
e
d
on
t
he
pr
e
pr
o
c
e
s
s
e
d
t
r
a
i
n
i
ng
da
t
a
w
he
r
e
A
ut
oG
l
uon
w
a
s
t
r
a
i
ne
d
us
i
ng
a
S
a
ge
M
a
ke
r
P
y
t
hon
S
c
r
i
pt
,
c
a
l
l
e
d
l
a
t
e
r
in
t
he
not
e
book,
w
hi
l
e
l
i
ne
a
r
l
e
a
r
ne
r
w
a
s
di
r
e
c
t
l
y
i
n
t
e
gr
a
t
e
d
i
nt
o
t
he
no
t
e
book,
t
he
t
r
a
i
ni
ng
t
i
m
e
w
a
s
r
e
c
or
de
d
us
i
ng
A
W
S
C
l
oudW
a
t
c
h
a
nd
t
he
r
e
s
ul
t
s
w
e
r
e
s
a
v
e
d
in
t
he
S3
buc
ke
t
,
in
t
h
e
f
i
f
t
h
s
t
a
ge
,
an
of
f
l
i
ne
ba
t
c
h
t
r
a
ns
f
o
r
m
pr
e
di
c
t
i
on
w
a
s
c
onduc
t
e
d
on
t
he
t
r
a
i
n
i
ng
r
e
s
u
l
t
s
to
m
e
a
s
u
r
e
t
he
a
c
c
ur
a
c
y
of
t
he
m
ode
l
s
a
nd
t
h
e
t
i
m
e
ne
e
de
d
to
obt
a
i
n
t
he
r
e
s
ul
t
s
,
t
he
s
i
xt
h
s
t
a
ge
i
nc
l
ude
d
t
he
de
pl
oy
m
e
nt
of
t
he
p
r
opos
e
d
m
ode
l
s
us
i
ng
t
w
o
s
e
pa
r
a
t
e
s
i
ngl
e
-
m
ode
l
e
ndpoi
n
t
s
to
m
e
a
s
ur
e
t
he
t
i
m
e
r
e
qui
r
e
d
f
o
r
t
h
e
pr
e
d
i
c
t
i
on
to
be
c
onduc
t
e
d
in
a
r
e
a
l
-
t
i
m
e
s
c
e
na
r
i
o.
T
he
c
onc
l
ud
i
ng
pha
s
e
i
nvol
ve
s
a
c
qui
r
i
ng
t
he
e
ndpo
i
nt
p
r
e
di
c
t
i
on
r
e
s
ul
t
s
a
nd
e
v
a
l
ua
t
i
ng
t
he
m
a
ga
i
ns
t
t
he
t
r
a
i
n
i
ng
a
nd
ba
t
c
h
t
r
a
n
s
f
or
m
out
c
o
m
e
s
.
S
i
x
e
va
l
ua
t
i
on
m
e
t
r
i
c
s
a
r
e
us
e
d
in
t
h
i
s
s
t
udy
:
a
c
c
ur
a
c
y
s
c
o
r
e
,
pr
e
c
i
s
i
on
,
F1
-
s
c
or
e
,
r
e
c
a
l
l
,
a
nd
r
e
c
e
i
ve
r
ope
r
a
t
i
ng
c
ha
r
a
c
t
e
r
i
s
t
i
c
(
R
O
C
)
.
It
a
i
m
s
to
c
om
pa
r
e
t
he
pe
r
f
or
m
a
nc
e
of
t
he
ba
gg
i
ng
e
ns
e
m
bl
e
s
t
r
a
t
e
gy
of
A
W
S
S
a
ge
M
a
ke
r
A
ut
oG
l
uon
w
i
t
h
t
he
s
i
m
p
l
e
l
i
ne
a
r
s
t
r
a
t
e
gy
of
A
W
S
l
i
ne
a
r
l
e
a
r
ne
r
to
gi
ve
s
om
e
i
de
a
on
t
he
t
r
a
de
-
of
f
be
t
w
e
e
n
t
he
a
ut
o
m
a
t
i
on
-
ba
s
e
d
e
n
s
e
m
b
l
e
s
c
he
m
e
a
nd
t
h
e
t
r
a
d
i
t
i
ona
l
l
i
ne
a
r
m
od
e
l
in
phi
s
hi
ng
de
t
e
c
t
i
on
s
e
t
t
i
ng.
T
he
a
na
l
ys
i
s
f
oc
u
s
e
s
on
f
our
f
und
a
m
e
n
t
a
l
a
s
pe
c
t
s
,
i
nc
l
udi
ng
de
t
e
c
t
i
on
a
c
c
ur
a
c
y,
t
r
a
i
n
i
ng
t
i
m
e
,
o
f
f
l
i
ne
p
r
e
di
c
t
i
on
t
i
m
e
,
a
nd
on
l
i
ne
i
n
f
e
r
e
nc
e
s
pe
e
d.
T
he
s
y
s
t
e
m
a
t
i
c
a
pp
r
oa
c
h
a
dopt
e
d
in
t
he
pa
pe
r
is
p
r
e
s
e
nt
e
d
in
a
s
y
s
t
e
m
a
t
i
c
w
a
y
a
nd
is
b
r
oke
n
dow
n
i
nt
o
t
he
f
ol
l
ow
i
ng
s
t
e
ps
,
as
s
how
n
in
Fi
gur
e
1
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
t
i
f
I
nt
e
l
l
I
S
S
N
:
2252
-
8938
C
om
par
i
s
on be
t
w
e
e
n e
ns
e
m
bl
e
and
l
i
ne
ar
m
e
t
hods
f
or
w
e
bs
i
t
e
ph
i
s
hi
ng d
e
t
e
c
t
i
on
(
Saba H
us
s
e
i
n R
as
h
i
d
)
683
F
i
gur
e
1.
T
he
ge
ne
r
a
l
f
l
ow
c
ha
r
t
of
t
he
pr
opo
s
e
d
m
e
t
hodol
ogy
2.1.
E
n
vi
r
on
m
e
n
t
c
on
f
i
gu
r
a
t
i
on
T
he
e
xpe
r
i
m
e
nt
de
t
a
i
l
e
d
in
t
hi
s
pa
pe
r
w
a
s
e
xe
c
ut
e
d
us
i
ng
a
l
a
pt
op
w
i
t
h
an
I
nt
e
l
C
or
e
i5
-
10210U
C
P
U
at
1.60
G
H
z
,
8
GB
of
R
A
M
,
a
nd
t
he
W
i
ndow
s
11
H
om
e
E
di
t
i
on
ope
r
a
t
i
ng
s
ys
t
e
m
.
T
he
A
W
S
S
a
ge
M
a
ke
r
S
t
udi
o
pl
a
t
f
or
m
w
a
s
e
m
pl
oye
d
to
c
r
e
a
t
e
a
nd
r
un
P
yt
hon
c
ode
s
in
J
upyt
e
r
N
ot
e
book
e
nvi
r
on
m
e
nt
.
S3
f
a
c
i
l
i
t
a
t
e
d
t
he
s
t
o
r
a
ge
a
nd
r
e
t
r
i
e
va
l
of
a
l
l
e
xpe
r
i
m
e
n
t
a
l
a
r
t
i
f
a
c
t
s
,
w
hi
l
e
t
he
“
m
l
.m
4.xl
a
r
ge
”
vi
r
t
ua
l
i
ns
t
a
nc
e
t
ype
f
r
o
m
E
C
2
pr
ov
i
de
d
f
ou
r
vi
r
t
ua
l
C
P
U
s
,
16
GB
of
vi
r
t
ua
l
m
e
m
or
y,
a
nd
h
i
gh
-
pe
r
f
or
m
a
nc
e
n
e
t
w
or
ki
ng
f
or
a
64
-
bi
t
ope
r
a
t
i
ng
s
ys
t
e
m
.
A
W
S
C
l
oudW
a
t
c
h
w
a
s
u
t
i
l
i
z
e
d
to
m
oni
t
or
t
he
pe
r
f
o
r
m
a
nc
e
of
t
he
S
a
ge
M
a
ke
r
m
ode
l
a
nd
to
l
og
t
h
e
pr
oc
e
s
s
i
ng,
t
r
a
i
n
i
ng,
a
nd
pr
e
d
i
c
t
i
on
dur
a
t
i
ons
.
2.2.
D
at
as
e
t
c
on
f
i
gu
r
a
t
i
on
T
he
“
Web
pa
ge
phi
s
hi
ng de
t
e
c
t
i
on
”
da
t
a
s
e
t
ut
i
l
i
z
e
d
in
t
h
i
s
w
or
k
is
a
va
i
l
a
b
l
e
on
M
e
nde
l
e
y
da
t
a
[
30]
.
It
ha
s
87
f
e
a
t
ur
e
s
a
nd
11,430
U
R
L
s
.
To
i
nc
r
e
a
s
e
a
c
c
ur
a
c
y
a
nd
g
a
i
n
a
de
e
pe
r
unde
r
s
t
a
nd
i
ng
of
t
he
da
t
a
s
e
t
’
s
be
ha
vi
or
,
t
he
87
c
h
a
r
a
c
t
e
r
i
s
t
i
c
s
in
t
he
da
t
a
s
e
t
a
r
e
c
l
a
s
s
i
f
i
e
d
as
s
e
e
n
in
F
i
gur
e
2:
56
f
e
a
t
ur
e
s
de
r
i
ve
d
f
r
om
U
R
L
s
ynt
a
x;
24
f
e
a
t
u
r
e
s
r
e
s
ul
t
i
ng
f
r
om
t
he
c
ont
e
n
t
of
w
e
bs
i
t
e
s
;
a
nd
7
f
e
a
t
ur
e
s
obt
a
i
ne
d
f
r
o
m
w
e
bs
i
t
e
s
e
r
v
i
c
e
s
.
2.3. D
at
as
e
t
p
r
e
p
r
oc
e
s
s
i
n
g
A
f
t
e
r
a
na
l
yz
i
ng
t
he
da
t
a
s
e
t
f
or
t
hi
s
w
or
k,
i
t
ha
s
be
e
n
c
onc
l
ude
d
t
ha
t
i
t
i
s
ba
l
a
nc
e
d,
w
i
t
h
5,715
s
a
m
pl
e
s
of
l
e
g
i
t
i
m
a
t
e
U
R
L
s
a
nd
5,715
e
xa
m
p
l
e
s
o
f
phi
s
hi
ng
U
R
L
s
,
a
l
l
of
w
hi
c
h
ha
d
no
m
i
s
s
i
ng
va
l
ue
s
.
T
he
r
e
i
s
no
ne
e
d
f
or
a
n
e
a
r
l
i
e
r
pr
e
pr
oc
e
s
s
i
ng
s
t
e
p
a
s
A
ut
oG
l
uon
c
oul
d
pr
e
p
r
oc
e
s
s
t
he
d
a
t
a
.
H
ow
e
ve
r
,
t
he
pur
pos
e
of
t
h
i
s
s
t
e
p
i
s
t
o
p
r
e
pr
oc
e
s
s
t
he
da
t
a
be
f
or
e
t
r
a
i
ni
ng
i
n
l
i
ne
a
r
l
e
a
r
n
e
r
.
A
s
a
r
e
s
ul
t
,
t
he
pr
e
pr
oc
e
s
s
i
ng
s
c
r
i
p
t
,
i
nt
e
gr
a
t
e
d
a
s
a
S
a
ge
M
a
ke
r
S
t
ud
i
o
P
yt
hon
S
c
r
i
pt
,
i
s
r
e
duc
e
d
t
o t
he
f
o
l
l
ow
i
ng f
i
ve
s
t
e
ps
,
w
hi
c
h
a
r
e
s
how
n
i
n
F
i
gur
e
3.
T
o
r
e
duc
e
dupl
i
c
a
t
i
on
a
nd
a
ny
de
t
r
i
m
e
n
t
a
l
e
f
f
e
c
t
s
on
m
ode
l
pe
r
f
or
m
a
nc
e
,
hi
ghl
y
c
or
r
e
l
a
t
e
d
f
e
a
t
u
r
e
s
w
e
r
e
f
i
r
s
t
e
xc
l
ude
d
f
r
o
m
t
he
f
e
a
t
ur
e
s
e
l
e
c
t
i
on
pr
oc
e
s
s
w
i
t
h
t
he
us
e
of
P
e
a
r
s
on
c
or
r
e
l
a
t
i
on. T
hr
e
e
f
e
a
t
ur
e
s
w
e
r
e
r
e
m
ov
e
d be
c
a
us
e
t
he
i
r
c
or
r
e
l
a
t
i
on r
a
t
e
s
w
e
r
e
h
i
ghe
r
t
ha
n 90%
. S
ubs
e
que
n
t
l
y,
t
o
i
m
pr
ov
e
m
ode
l
pe
r
f
or
m
a
nc
e
t
h
r
ough
l
ow
e
r
i
ng
no
i
s
e
,
f
e
a
t
ur
e
r
e
duc
t
i
on
w
a
s
c
a
r
r
i
e
d
out
by
r
e
m
ov
i
ng
t
he
nom
i
na
l
f
e
a
t
ur
e
“
U
R
L
,”
w
hi
c
h
r
e
pr
e
s
e
nt
e
d
t
he
na
m
e
s
of
t
he
da
t
a
s
e
t
’
s
U
R
L
s
.
L
a
be
l
e
nc
odi
ng
w
a
s
us
e
d
t
o
t
r
a
ns
f
o
r
m
c
a
t
e
gor
i
c
a
l
l
a
be
l
va
l
ue
s
i
nt
o
nu
m
e
r
i
c
va
l
ue
s
,
e
ns
u
r
i
ng
c
o
m
pa
t
i
bi
l
i
t
y
w
i
t
h
M
L
t
e
c
hni
que
s
.
T
he
e
nc
ode
d
“
s
t
a
t
us
”
f
e
a
t
ur
e
,
w
hi
c
h
de
no
t
e
s
da
t
a
s
e
t
c
l
a
s
s
i
f
i
c
a
t
i
ons
,
ha
s
“
phi
s
h
i
ng”
a
s
s
i
gn
e
d
t
o
one
(
1
)
a
nd
“
l
e
gi
t
i
m
a
t
e
”
s
e
t
t
o
z
e
r
o
(
0
)
.
A
ddi
t
i
ona
l
l
y,
f
e
a
t
u
r
e
s
c
a
l
i
ng
w
a
s
c
a
r
r
i
e
d
ou
t
t
o
r
e
c
onf
i
gur
e
da
t
a
w
i
t
h
i
n
a
gi
ve
n
r
a
nge
,
of
t
e
n
0 t
o
1,
us
i
ng
P
y
t
hon'
s
M
i
nM
a
xS
c
a
l
e
r
(
)
.
B
y
t
a
ki
ng t
hi
s
s
t
e
p,
f
e
a
t
ur
e
s
w
i
t
h
va
l
ue
s
be
t
w
e
e
n
2
a
nd
100
a
r
e
no
t
a
l
l
ow
e
d t
o
do
m
i
na
t
e
t
he
o
t
he
r
f
e
a
t
ur
e
s
.
F
ol
l
ow
i
ng t
ha
t
,
a
70
:
30
r
a
t
i
o
w
a
s
us
e
d
t
o
s
pl
i
t
t
he
da
t
a
s
e
t
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
t
i
f
I
nt
e
l
l
,
V
ol
. 15, N
o. 1, F
e
br
ua
r
y 2026
:
681
-
694
684
i
nt
o
t
r
a
i
ni
ng
a
s
w
e
l
l
a
s
t
e
s
t
i
ng
s
e
t
s
,
w
i
t
h
30
%
(
3,425
s
a
m
p
l
e
s
)
a
nd
70%
(
8,001
s
a
m
pl
e
s
)
go
i
ng
t
o
e
a
c
h
s
e
t
.
F
i
na
l
l
y, S
3 w
a
s
us
e
d
t
o upl
oa
d bo
t
h s
e
t
s
f
o
r
a
ddi
t
i
ona
l
pr
oc
e
s
s
i
ng.
F
i
gur
e
2. A
na
l
ys
i
s
of
t
he
da
t
a
s
e
t
us
e
d i
n t
h
e
pr
opos
e
d
m
e
t
hodo
l
ogy
F
i
gur
e
3.
S
t
e
ps
of
da
t
a
p
r
e
pr
oc
e
s
s
i
ng
s
t
e
p
of
t
he
pr
opo
s
e
d
m
e
t
hodol
ogy
56
24
7
f
e
a
t
ur
e
s
de
r
i
ve
d f
r
om
t
he
s
ynt
a
x of
t
he
U
ni
f
or
m
R
e
s
our
c
e
L
oc
a
t
or
s
(
U
R
L
)
f
e
a
t
ur
e
s
de
r
i
ve
d f
r
om
t
he
c
ont
e
nt
of
t
he
w
e
bs
i
t
e
s
f
e
a
t
ur
e
s
de
r
i
ve
d f
r
om
t
he
s
e
r
vi
c
e
s
a
s
s
oc
i
a
t
e
d w
i
t
h t
he
w
e
bs
i
t
e
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
t
i
f
I
nt
e
l
l
I
S
S
N
:
2252
-
8938
C
om
par
i
s
on be
t
w
e
e
n e
ns
e
m
bl
e
and
l
i
ne
ar
m
e
t
hods
f
or
w
e
bs
i
t
e
ph
i
s
hi
ng d
e
t
e
c
t
i
on
(
Saba H
us
s
e
i
n R
as
h
i
d
)
685
2.4.
T
r
ai
n
i
n
g
To
a
s
c
e
r
t
a
i
n
t
h
e
opt
i
m
a
l
a
ppr
oa
c
h
f
or
m
a
na
gi
ng
l
a
r
g
e
da
t
a
s
e
t
s
in
pr
a
c
t
i
c
a
l
a
pp
l
i
c
a
t
i
on
s
,
t
he
t
r
a
i
ni
ng
s
t
a
ge
c
on
t
r
a
s
t
s
S
a
ge
M
a
ke
r
’
s
e
n
s
e
m
b
l
e
m
e
t
hod
w
i
t
h
t
he
l
i
ne
a
r
m
e
t
hod.
T
w
o
a
l
gor
i
t
hm
s
,
A
ut
oM
L
A
ut
oG
l
uon
a
nd
l
i
ne
a
r
l
e
a
r
ne
r
,
a
r
e
e
m
p
l
oye
d
t
o
t
r
a
i
n
a
nd
a
s
s
e
s
s
t
he
pr
opo
s
e
d
m
ode
l
.
L
i
ne
a
r
l
e
a
r
ne
r
w
a
s
i
nvoke
d
in
t
he
S
a
ge
M
a
ke
r
no
t
e
book
v
i
a
s
pe
c
i
f
i
c
c
ode
,
as
it
is
i
nt
e
g
r
a
t
e
d
w
i
t
h
i
n
t
he
S
a
ge
M
a
ke
r
S
t
ud
i
o
N
ot
e
book
a
r
c
hi
t
e
c
t
u
r
e
,
w
hi
l
e
A
ut
oG
l
uon’
s
c
ode
ne
e
de
d
a
dd
i
t
i
ona
l
,
s
e
p
a
r
a
t
e
P
yt
hon
S
c
r
i
pt
s
to
c
onf
i
gu
r
e
t
he
t
r
a
i
ni
ng
a
nd
i
nf
e
r
e
nc
e
hype
r
pa
r
a
m
e
t
e
r
s
.
2.4.1.
A
u
t
oG
l
u
on
A
ut
oG
l
uon
can
be
de
f
i
ne
d
as
an
ope
n
-
s
our
c
e
A
ut
oM
L
f
r
a
m
e
w
or
k
de
ve
l
ope
d
by
A
W
S
w
hi
c
h
a
ut
om
a
t
e
s
t
he
c
ons
t
r
uc
t
i
on
of
e
x
a
c
t
ML
m
ode
l
s
w
i
t
h
m
i
ni
m
a
l
m
a
nua
l
e
f
f
or
t
[
31]
.
T
he
f
r
a
m
e
w
or
k
s
i
m
p
l
i
f
i
e
s
c
om
pl
i
c
a
t
e
d
p
r
oc
e
dur
e
s
s
uc
h
as
m
od
e
l
s
e
l
e
c
t
i
on,
hype
r
pa
r
a
m
e
t
e
r
t
uni
ng,
a
nd
f
e
a
t
u
r
e
e
ngi
ne
e
r
i
ng,
m
a
ki
ng
t
he
m
a
va
i
l
a
b
l
e
f
or
e
xpe
r
t
s
a
nd
non
-
e
xpe
r
t
s
,
i
m
p
r
ovi
ng
a
c
c
u
r
a
c
y
by
m
e
r
gi
ng
di
f
f
e
r
e
nt
m
ode
l
s
a
nd
m
e
t
hods
a
nd
t
hus
m
a
k
i
ng
it
us
e
f
u
l
f
or
t
a
bul
a
r
da
t
a
a
ppl
i
c
a
t
i
ons
s
u
c
h
as
c
l
a
s
s
i
f
i
c
a
t
i
on
[
32
]
.
A
ut
oG
l
uon
op
e
r
a
t
e
s
w
i
t
hi
n
P
yt
hon
S
c
r
i
pt
s
to
ge
ne
r
a
t
e
r
e
l
i
a
bl
e
pr
e
di
c
t
i
on
m
od
e
l
s
w
i
t
h
m
i
n
i
m
a
l
hum
a
n
i
nput
,
as
it
us
e
s
ba
ggi
ng
t
e
c
hni
que
s
to
m
i
ni
m
i
z
e
va
r
i
a
t
i
on
a
nd
e
nh
a
nc
e
s
t
a
bi
l
i
t
y,
by
t
r
a
i
ni
ng
m
u
l
t
i
pl
e
m
ode
l
s
on
r
a
ndom
d
a
t
a
,
a
nd
m
a
ke
pr
e
d
i
c
t
i
ons
ba
s
e
d
on
m
a
j
or
i
t
y
vo
t
e
s
[
33
]
.
A
ut
oG
l
uon
a
l
s
o
a
ut
o
m
a
t
e
s
hype
r
pa
r
a
m
e
t
e
r
t
uni
ng
by
us
i
ng
r
a
ndom
s
e
a
r
c
h
a
l
gor
i
t
hm
s
,
as
s
how
n
in
T
a
bl
e
1.
A
f
t
e
r
t
r
a
i
ni
ng,
it
a
n
a
l
yz
e
s
m
ode
l
s
vi
a
c
r
os
s
-
va
l
i
da
t
i
on
a
nd
a
ggr
e
ga
t
e
s
t
he
be
s
t
pe
r
f
o
r
m
e
r
s
f
o
r
r
e
l
i
a
b
l
e
p
r
e
di
c
t
i
ons
[
34]
.
F
o
r
t
h
i
s
pa
pe
r
,
A
ut
oG
l
uon
w
a
s
c
hos
e
n
due
to
its
a
bi
l
i
t
y
to
a
ut
om
a
t
e
m
a
j
or
opt
i
m
i
z
a
t
i
on
pr
oc
e
s
s
e
s
by
ut
i
l
i
z
i
ng
t
he
c
om
b
i
na
t
i
on
di
s
t
r
i
bu
t
e
d
t
r
a
i
ni
ng
a
m
ong
m
ul
t
i
pl
e
d
i
ve
r
s
e
l
e
a
r
ne
r
s
w
i
t
h
i
n
i
t
s
a
r
c
hi
t
e
c
t
ur
e
e
nha
nc
e
s
t
he
r
obus
t
ne
s
s
a
nd
ge
ne
r
a
l
i
z
a
t
i
on
of
t
he
m
ode
l
,
e
na
bl
i
ng
r
a
pi
d
de
p
l
oym
e
n
t
,
a
nd
m
a
k
i
ng
it
pr
e
f
e
r
a
bl
e
f
or
r
e
a
l
-
w
or
l
d
a
pp
l
i
c
a
t
i
ons
.
A
ut
oG
l
uon
u
t
i
l
i
z
e
s
t
he
ba
ggi
ng
a
ppr
oa
c
h
to
e
nha
nc
e
t
he
pe
r
f
or
m
a
n
c
e
a
nd
s
t
r
e
ngt
h
of
ML
a
l
go
r
i
t
hm
s
as
de
pi
c
t
e
d
in
f
ol
l
ow
i
ng
s
t
e
ps
:
i)
B
a
s
e
m
ode
l
s
(
l
e
ve
l
1)
−
A
ut
oG
l
uon
t
r
a
i
n
s
a
s
e
t
of
ba
s
e
m
od
e
l
s
us
i
ng
a
l
go
r
i
t
hm
s
l
i
ke
L
i
gh
t
G
B
M
,
C
a
t
B
oos
t
,
RF
,
E
xt
r
a
T
r
e
e
s
,
X
G
B
oos
t
,
KNN
,
a
nd
ne
ur
a
l
ne
t
w
or
ks
(
F
a
s
t
A
I
)
.
−
B
a
ggi
ng
is
a
ppl
i
e
d
as
a
f
or
m
of
c
r
o
s
s
va
l
i
da
t
i
on,
w
he
r
e
:
a)
T
he
t
r
a
i
ni
ng
da
t
a
is
di
vi
de
d
i
n
t
o
s
e
ve
r
a
l
boot
s
t
r
a
pp
e
d
s
a
m
pl
e
s
(
t
hi
s
r
e
s
e
a
r
c
h
us
e
d
f
i
ve
r
a
ndom
s
ubs
e
t
s
)
.
b)
E
a
c
h
ba
s
e
m
ode
l
is
t
r
a
i
ne
d
on
a
d
i
s
s
i
m
i
l
a
r
s
e
t
of
da
t
a
,
e
na
bl
i
ng
t
he
m
ode
l
e
ns
e
m
bl
e
to
l
e
a
r
n
us
i
ng
s
l
i
ght
l
y
di
s
s
i
m
i
l
a
r
da
t
a
d
i
s
t
r
i
bu
t
i
on.
c)
T
he
pr
oc
e
s
s
ge
ne
r
a
t
e
s
s
e
ve
r
a
l
va
r
i
a
t
i
ons
of
t
he
s
a
m
e
ba
s
e
m
ode
l
,
a
nd
each
one
of
t
he
m
r
e
f
l
e
c
t
s
d
i
s
t
i
nc
t
f
e
a
t
u
r
e
a
s
s
o
c
i
a
t
i
ons
a
nd
m
i
n
i
m
i
z
e
s
va
r
i
a
nc
e
in
t
he
f
i
na
l
pr
e
d
i
c
t
i
on.
E
a
c
h
t
r
a
i
ne
d
m
ode
l
t
he
n
ge
ne
r
a
t
e
s
pr
e
di
c
t
i
on
s
ba
s
e
d
on
t
he
pa
t
t
e
r
ns
l
e
a
r
n
e
d
f
r
o
m
i
t
s
r
e
s
pe
c
t
i
ve
da
t
a
s
ubs
e
t
,
a
nd
t
he
i
r
ou
t
put
s
a
r
e
l
a
t
e
r
c
o
m
bi
ne
d
to
f
o
r
m
a
r
obu
s
t
e
ns
e
m
bl
e
pr
e
d
i
c
t
i
on.
ii)
F
i
r
s
t
-
l
a
ye
r
pr
e
d
i
c
t
i
ons
−
O
nc
e
t
r
a
i
ne
d,
t
he
ba
s
e
m
ode
l
s
m
a
ke
p
r
e
di
c
t
i
ons
on
a
v
a
l
i
d
a
t
i
on
s
e
t
of
t
he
t
r
a
i
ni
ng
da
t
a
.
−
T
he
s
e
pr
e
di
c
t
i
ons
a
r
e
r
e
f
e
r
r
e
d
to
as
f
i
r
s
t
-
l
a
ye
r
pr
e
d
i
c
t
i
ons
a
nd
s
e
r
ve
as
i
nput
f
e
a
t
ur
e
s
f
or
t
he
s
e
c
ond
l
a
ye
r
of
m
ode
l
s
.
i
i
i
)
S
t
a
c
ki
ng
(
l
e
ve
l
2)
−
M
e
t
a
-
m
ode
l
s
,
or
l
e
v
e
l
2
m
od
e
l
s
,
a
r
e
c
r
e
a
t
e
d
us
i
ng
t
he
f
i
r
s
t
-
l
a
ye
r
p
r
e
di
c
t
i
on
s
as
i
npu
t
.
−
In
s
t
a
c
ki
ng,
a
m
e
t
a
-
m
ode
l
is
t
r
a
i
ne
d
to
c
om
b
i
ne
t
h
e
out
put
s
f
r
o
m
l
e
ve
l
1
m
ode
l
s
.
−
T
hi
s
m
e
t
a
-
m
ode
l
l
e
a
r
ns
to
w
e
i
gh
t
he
pr
e
d
i
c
t
i
ons
f
r
om
di
f
f
e
r
e
nt
ba
s
e
m
ode
l
s
to
m
i
ni
m
i
z
e
pr
e
di
c
t
i
on
e
r
r
or
s
.
i
v)
W
e
i
ght
e
d
e
n
s
e
m
b
l
e
(
l
e
ve
l
2
m
ode
l
)
−
T
he
“
w
e
i
ght
e
d
e
ns
e
m
bl
e
”
m
ode
l
r
e
pr
e
s
e
n
t
s
t
he
m
e
t
a
-
m
ode
l
c
om
b
i
ni
ng
pr
e
di
c
t
i
on
s
f
r
om
l
e
ve
l
1
m
ode
l
s
w
i
t
h
t
he
us
e
of
a
w
e
i
gh
t
e
d
a
ve
r
a
g
e
.
−
W
e
i
ght
s
a
r
e
ba
s
e
d
upon
pe
r
f
o
r
m
a
nc
e
of
e
ve
r
y
one
of
t
he
ba
s
e
m
ode
l
s
t
hr
oughou
t
t
he
t
r
a
i
ni
ng
—
be
t
t
e
r
-
pe
r
f
or
m
i
ng
m
od
e
l
s
ha
v
e
be
e
n
a
s
s
i
gne
d
h
i
ghe
r
w
e
i
gh
t
va
l
ue
s
,
w
hi
l
e
poor
e
r
one
s
r
e
c
e
i
ve
d
l
ow
e
r
w
e
i
ght
s
.
−
T
hi
s
t
e
c
hn
i
que
r
e
s
ul
t
s
in
t
he
i
m
pr
ove
m
e
nt
of
t
he
ov
e
r
a
l
l
a
c
c
ur
a
c
y
of
pr
e
di
c
t
i
on
by
gi
v
i
ng
a
h
i
ghe
r
l
e
ve
l
of
i
m
po
r
t
a
nc
e
to
t
h
e
s
t
r
onge
r
m
od
e
l
s
.
T
hos
e
s
t
e
ps
ha
v
e
be
e
n
i
l
l
us
t
r
a
t
e
d
in
F
i
gur
e
4.
T
a
bl
e
1
.
H
ype
r
pa
r
a
m
e
t
e
r
s
of
A
ut
oG
l
uon
H
ype
r
pa
r
a
m
e
t
e
r
V
a
l
ue
P
ur
pos
e
a
nd
e
f
f
e
c
t
num
_ba
g_f
ol
d
s
5
S
pe
c
i
f
i
e
s
t
he
num
be
r
of
f
ol
d
s
f
or
k
-
f
ol
d
ba
ggi
ng
to
r
e
d
uc
e
ove
r
f
i
t
t
i
ng.
num
_ba
g_s
e
t
s
1
i
nd
i
c
a
t
e
s
t
he
num
be
r
of
c
om
pl
e
t
e
ba
ggi
ng
r
ound
s
.
num
_s
t
a
c
k_l
e
ve
l
s
0
D
i
s
a
bl
e
s
a
d
d
i
t
i
ona
l
s
t
a
c
ki
ng
l
a
ye
r
s
to
s
hor
t
e
n
t
r
a
i
ni
ng
t
i
m
e
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
t
i
f
I
nt
e
l
l
,
V
ol
. 15, N
o. 1, F
e
br
ua
r
y 2026
:
681
-
694
686
F
i
gur
e
4
.
S
t
e
ps
of
A
ut
oG
l
uon
t
r
a
i
n
i
ng
2.4.2.
L
i
n
e
ar
l
e
ar
n
e
r
T
he
t
e
c
hni
que
of
l
i
ne
a
r
c
l
a
s
s
i
f
i
c
a
t
i
on
in
A
m
a
z
on
S
a
ge
M
a
ke
r
can
be
de
s
c
r
i
be
d
as
s
upe
r
v
i
s
e
d
l
e
a
r
ni
ng
a
ppr
o
a
c
h
t
ha
t
ha
s
be
e
n
de
s
i
gne
d
f
o
r
t
he
e
nha
nc
e
m
e
n
t
of
bi
na
r
y
c
l
a
s
s
i
f
i
c
a
t
i
on
t
a
s
k
s
.
It
bui
l
ds
a
l
i
ne
a
r
de
c
i
s
i
on
bounda
r
y,
w
hi
c
h
is
e
xh
i
bi
t
e
d
as
hype
r
p
l
a
ne
,
f
or
t
he
pur
pos
e
of
s
e
pa
r
a
t
i
ng
t
h
e
da
t
a
poi
n
t
s
of
di
f
f
e
r
e
nt
c
l
a
s
s
e
s
by
t
he
a
s
s
i
gn
m
e
nt
a
nd
op
t
i
m
i
z
a
t
i
on
of
t
h
e
f
e
a
t
u
r
e
w
e
i
ght
s
f
or
c
l
a
s
s
di
s
s
oc
i
a
t
i
on
[
35
]
.
T
he
m
ode
l
c
ont
i
nuous
l
y
a
dj
us
t
s
t
hos
e
w
e
i
gh
t
va
l
ue
s
t
h
r
oughout
t
he
t
r
a
i
ni
ng
f
or
m
i
ni
m
i
z
i
ng
e
r
r
o
r
s
a
nd
i
m
p
r
ovi
ng
a
c
c
ur
a
c
y.
T
hi
s
t
e
c
hn
i
que
is
a
ppr
e
c
i
a
t
e
d
s
i
nc
e
it
is
s
c
a
l
a
bl
e
,
c
om
pu
t
a
t
i
ona
l
l
y
e
f
f
i
c
i
e
n
t
,
a
nd
i
nt
e
r
pr
e
t
a
bl
e
,
w
hi
c
h
m
a
ke
s
it
s
ui
t
a
bl
e
f
or
l
a
r
ge
-
s
c
a
l
e
a
ppl
i
c
a
t
i
ons
[
36
]
.
A
c
c
or
d
i
ng
to
T
a
bl
e
2,
w
hi
c
h
pr
e
s
e
nt
e
d
t
he
hype
r
pa
r
a
m
e
t
e
r
s
us
e
d
f
or
t
h
e
a
l
gor
i
t
hm
,
S
a
ge
M
a
ke
r
’
s
l
i
ne
a
r
l
e
a
r
ne
r
e
m
p
l
oys
t
he
s
t
oc
ha
s
t
i
c
gr
a
d
i
e
nt
de
s
c
e
nt
(
S
G
D
)
f
or
t
he
pur
pos
e
of
ha
ndl
i
ng
l
a
r
ge
da
t
a
s
e
t
s
.
It
ut
i
l
i
z
e
s
a
ut
o
m
a
t
i
c
f
e
a
t
ur
e
s
c
a
l
i
ng
f
or
nor
m
a
l
i
z
i
ng
f
e
a
t
ur
e
s
of
di
f
f
e
r
e
nt
s
c
a
l
e
s
a
nd
r
e
gul
a
r
i
z
i
ng
t
he
t
e
c
hn
i
que
s
f
o
r
ove
r
f
i
t
t
i
ng
pr
e
ve
n
t
i
on.
S
G
D
opt
i
m
i
z
e
s
t
he
m
ode
l
t
hr
ough
t
he
a
d
j
us
t
m
e
nt
of
w
e
i
ght
va
l
ue
s
in
e
r
r
or
di
r
e
c
t
i
on,
gui
d
e
d
by
t
he
g
r
a
di
e
n
t
of
l
os
s
f
unc
t
i
on.
T
he
l
e
a
r
ni
ng
r
a
t
e
r
e
gul
a
t
e
s
t
h
e
s
i
z
e
of
t
hos
e
a
d
j
us
t
m
e
nt
s
,
w
hi
c
h
l
e
a
ds
to
ba
l
a
nc
i
ng
t
he
s
t
a
bi
l
i
t
y
of
t
he
m
ode
l
a
nd
s
pe
e
d
of
c
onve
r
ge
nc
e
.
T
he
pr
oc
e
s
s
of
t
r
a
i
n
i
ng
c
ont
i
nue
s
to
t
h
e
poi
nt
w
he
r
e
m
i
ni
m
a
l
i
m
pr
ov
e
m
e
n
t
s
a
r
e
a
c
hi
e
ve
d,
or
a
s
e
t
nu
m
be
r
of
i
t
e
r
a
t
i
ons
a
r
e
c
om
pl
e
t
e
d
[
37
]
.
T
he
S
a
ge
M
a
ke
r
i
nt
e
gr
a
t
e
d
a
l
go
r
i
t
h
m
w
a
s
c
hos
e
n
f
or
c
om
pa
r
i
s
on
a
ga
i
n
s
t
A
ut
oG
l
uon
in
t
hi
s
pa
pe
r
be
c
a
us
e
of
i
t
s
c
a
pa
bi
l
i
t
y
of
ha
ndl
i
ng
l
a
r
ge
da
t
a
s
e
t
s
e
f
f
e
c
t
i
ve
l
y,
of
f
e
r
i
ng
a
f
a
s
t
t
r
a
i
ni
ng
a
nd
i
nf
e
r
e
nc
e
t
i
m
e
,
w
i
t
h
e
a
s
y
i
nt
e
r
pr
e
t
a
bi
l
i
t
y
,
bui
l
t
-
in
r
e
gul
a
r
i
z
a
t
i
on
to
r
e
duc
e
ove
r
f
i
t
t
i
ng,
a
nd
l
ow
e
r
c
ons
u
m
pt
i
on
of
c
l
oud
r
e
s
ou
r
c
e
s
,
w
hi
c
h
g
i
ve
s
it
t
h
e
be
ne
f
i
t
of
f
a
s
t
de
pl
oy
m
e
nt
,
m
a
ki
ng
it
s
u
i
t
a
b
l
e
f
or
t
i
m
e
-
s
e
ns
i
t
i
ve
phi
s
hi
ng
de
t
e
c
t
i
on
s
ys
t
e
m
s
.
A
br
e
a
kdow
n
of
t
he
pr
oc
e
s
s
t
ha
t
is
i
nvol
ve
d
in
t
he
t
r
a
i
n
i
ng
of
a
l
i
ne
a
r
l
e
a
r
ne
r
m
ode
l
in
A
m
a
z
on
S
a
ge
M
a
k
e
r
ha
s
be
e
n
d
e
pi
c
t
e
d
in
F
i
gu
r
e
5
.
O
bj
e
c
t
i
ve
f
unc
t
i
on a
nd r
e
gu
l
a
r
i
z
a
t
i
on
:
i)
O
bj
e
c
t
i
ve
f
unc
t
i
on
:
l
i
ne
a
r
l
e
a
r
ne
r
us
e
s
an
obj
e
c
t
i
v
e
f
unc
t
i
on
to
m
e
a
s
ur
e
t
he
e
r
r
or
be
t
w
e
e
n
pr
e
di
c
t
e
d
a
nd
a
c
t
ua
l
l
a
be
l
s
.
F
or
bi
na
r
y
c
l
a
s
s
i
f
i
c
a
t
i
on,
l
ogi
s
t
i
c
r
e
gr
e
s
s
i
on
is
e
m
pl
oye
d,
e
s
t
i
m
a
t
i
ng
t
he
pr
oba
bi
l
i
t
y
t
ha
t
an
i
nput
be
l
ongs
to
a
s
pe
c
i
f
i
c
c
l
a
s
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
t
i
f
I
nt
e
l
l
I
S
S
N
:
2252
-
8938
C
om
par
i
s
on be
t
w
e
e
n e
ns
e
m
bl
e
and
l
i
ne
ar
m
e
t
hods
f
or
w
e
bs
i
t
e
ph
i
s
hi
ng d
e
t
e
c
t
i
on
(
Saba H
us
s
e
i
n R
as
h
i
d
)
687
ii)
R
e
gul
a
r
i
z
a
t
i
on
:
‒
L1
r
e
gul
a
r
i
z
a
t
i
on
:
e
nc
our
a
ge
s
s
pa
r
s
i
t
y
in
t
he
m
ode
l
by
s
hr
i
nki
ng
w
e
i
gh
t
s
t
ow
a
r
d
z
e
r
o,
w
hi
c
h
s
i
m
p
l
i
f
i
e
s
t
he
m
ode
l
by
i
gnor
i
ng
l
e
s
s
i
m
por
t
a
nt
f
e
a
t
u
r
e
s
.
‒
L2
r
e
gul
a
r
i
z
a
t
i
on
:
t
hi
s
is
a
t
e
r
m
t
ha
t
is
us
e
d
to
pe
na
l
i
z
e
bi
g
w
e
i
gh
t
s
t
ha
t
gi
ve
out
s
m
oo
t
he
r
m
ode
l
s
.
B
ot
h
r
e
gul
a
r
i
z
a
t
i
on
m
e
t
hods
a
r
e
s
pe
c
i
f
i
e
d
us
i
ng
t
he
hype
r
pa
r
a
m
e
t
e
r
l
a
m
bda
t
ha
t
m
odul
a
t
e
s
t
he
s
t
r
e
ng
t
h
of
t
he
p
e
na
l
t
y
in
or
d
e
r
to
s
uppor
t
t
he
pr
e
ve
n
t
i
on
of
ove
r
f
i
t
t
i
ng
i
i
i
)
S
G
D
opt
i
m
i
z
a
t
i
on
:
‒
I
ni
t
i
a
l
i
z
a
t
i
on:
t
he
m
ode
l
be
gi
ns
by
i
ni
t
i
a
l
i
z
i
ng
pa
r
a
m
e
t
e
r
s
,
i
nc
l
udi
ng
w
e
i
ght
s
f
or
each
f
e
a
t
u
r
e
a
nd
bi
a
s
e
s
.
‒
G
r
a
di
e
nt
c
a
l
c
u
l
a
t
i
on:
t
he
gr
a
d
i
e
nt
of
t
he
obj
e
c
t
i
v
e
f
unc
t
i
on
is
c
a
l
c
ul
a
t
e
d,
r
e
pr
e
s
e
nt
i
ng
t
he
di
r
e
c
t
i
on
a
nd
m
a
gni
t
ude
of
c
ha
nge
in
t
he
e
r
r
o
r
r
e
l
a
t
i
v
e
to
t
h
e
m
ode
l
'
s
pa
r
a
m
e
t
e
r
s
.
‒
M
i
ni
-
b
a
t
c
h
upda
t
e
s
:
i
n
s
t
e
a
d
of
c
o
m
put
i
ng
gr
a
d
i
e
nt
s
on
t
he
e
nt
i
r
e
da
t
a
s
e
t
,
t
h
e
m
ode
l
upda
t
e
s
its
pa
r
a
m
e
t
e
r
s
u
s
i
ng
s
m
a
l
l
s
ubs
e
t
s
of
t
he
da
t
a
(
m
i
ni
-
b
a
t
c
he
s
)
,
i
m
pr
ov
i
ng
c
om
pu
t
a
t
i
ona
l
e
f
f
i
c
i
e
nc
y
f
or
l
a
r
ge
da
t
a
s
e
t
s
.
‒
W
e
i
ght
a
nd
bi
a
s
upda
t
e
s
:
w
e
i
ght
s
a
nd
bi
a
s
e
s
a
r
e
i
t
e
r
a
t
i
ve
l
y
a
dj
us
t
e
d
ba
s
e
d
on
t
he
gr
a
di
e
n
t
s
to
m
i
ni
m
i
z
e
t
he
e
r
r
or
.
T
he
l
e
a
r
ni
ng
r
a
t
e
c
on
t
r
ol
s
how
l
a
r
ge
each
upda
t
e
s
t
e
p
i
s
.
T
he
pr
oc
e
s
s
c
on
t
i
nue
s
unt
i
l
e
i
t
he
r
t
he
w
e
i
gh
t
s
c
onve
r
ge
,
or
t
he
m
a
x
i
m
u
m
num
be
r
of
i
t
e
r
a
t
i
ons
is
r
e
a
c
he
d.
T
a
bl
e
2
.
H
ype
r
pa
r
a
m
e
t
e
r
s
of
l
i
ne
a
r
l
e
a
r
ne
r
Hyperparam
e
t
er
Valu
e
Pu
rpose
a
n
d
effec
t
m
i
n
i_ba
t
c
h
_size
200
th
e
num
ber
of
sa
m
ples
processed
before
u
pda
t
i
n
g
m
odel
para
m
e
t
ers
to
lower
com
p
ut
a
t
io
n
t
i
m
e.
epoch
s
10
th
e
num
ber
of
co
m
ple
t
e
passes
th
ro
u
g
h
th
e
t
rai
n
i
n
g
da
t
ase
t
to
red
u
ce
overfi
tt
i
n
g.
regu
lariza
t
io
n
(L1,
L2)
Aut
o
Applies
bo
th
L1
a
n
d
L2
reg
u
lariza
t
io
n
to
red
u
ce
overfi
tt
i
n
g
.
F
i
gur
e
5
.
S
t
e
ps
of
l
i
ne
a
r
l
e
a
r
ne
r
t
r
a
i
n
i
ng
3.
R
E
S
U
L
T
S
AND
D
I
S
C
U
S
S
I
O
N
F
ol
l
ow
i
ng
t
r
a
i
n
i
ng
w
i
t
h
bot
h
A
ut
oG
l
uon
a
nd
l
i
ne
a
r
l
e
a
r
ne
r
,
t
he
t
r
a
i
ni
ng
a
r
t
i
f
a
c
t
s
w
e
r
e
upl
oa
de
d
to
t
he
S3
buc
ke
t
a
nd
ba
t
c
h
t
r
a
ns
f
o
r
m
w
a
s
t
he
n
c
a
r
r
i
e
d
out
on
t
he
r
e
s
ul
t
s
of
bot
h
t
e
c
hni
que
s
f
or
m
a
k
i
ng
of
f
l
i
ne
pr
e
di
c
t
i
ons
on
t
h
e
l
a
r
ge
d
a
t
a
s
e
t
t
hr
ough
di
vi
d
i
ng
t
he
m
i
n
t
o
ba
t
c
he
s
a
nd
upl
o
a
di
ng
t
he
r
e
s
u
l
t
s
to
S3
as
w
e
l
l
.
T
he
n,
e
ndpoi
nt
pr
e
d
i
c
t
i
on
w
a
s
c
a
r
r
i
e
d
out
by
de
pl
oy
i
ng
e
ndpoi
n
t
s
as
w
e
b
s
e
r
vi
c
e
e
ndpoi
n
t
s
w
hi
c
h
c
oul
d
r
e
c
e
i
ve
H
T
T
P
r
e
que
s
t
s
as
w
e
l
l
as
r
e
s
pons
e
s
w
i
t
h
pr
e
d
i
c
t
i
ons
in
a
l
m
os
t
r
e
a
l
t
i
m
e
.
T
he
e
va
l
ua
t
i
on
t
i
m
e
a
nd
r
e
s
ul
t
s
f
or
bo
t
h
m
e
t
hods
w
e
r
e
ob
t
a
i
ne
d
u
s
i
ng
t
he
t
e
s
t
da
t
a
s
e
t
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
t
i
f
I
nt
e
l
l
,
V
ol
. 15, N
o. 1, F
e
br
ua
r
y 2026
:
681
-
694
688
3.1.
E
val
u
at
i
on
r
e
s
u
l
t
s
F
ol
l
ow
i
ng
t
r
a
i
n
i
ng
w
i
t
h
bot
h
A
ut
oG
l
uon
a
nd
l
i
ne
a
r
l
e
a
r
ne
r
,
t
he
t
r
a
i
ni
ng
a
r
t
i
f
a
c
t
s
w
e
r
e
upl
oa
de
d
to
t
he
S3
buc
ke
t
a
nd
ba
t
c
h
t
r
a
ns
f
o
r
m
w
a
s
t
he
n
c
a
r
r
i
e
d
out
on
t
he
r
e
s
ul
t
s
of
bot
h
t
e
c
hni
que
s
f
or
m
a
k
i
ng
of
f
l
i
ne
pr
e
di
c
t
i
ons
on
t
h
e
l
a
r
ge
d
a
t
a
s
e
t
t
hr
ough
di
vi
d
i
ng
t
he
m
i
n
t
o
ba
t
c
he
s
a
nd
upl
o
a
di
ng
t
he
r
e
s
u
l
t
s
to
S3
as
w
e
l
l
.
T
he
n,
e
ndpoi
nt
pr
e
d
i
c
t
i
on
w
a
s
c
a
r
r
i
e
d
out
by
de
pl
oy
i
ng
e
ndpoi
n
t
s
as
w
e
b
s
e
r
vi
c
e
e
ndpoi
n
t
s
w
hi
c
h
c
oul
d
r
e
c
e
i
ve
H
T
T
P
r
e
que
s
t
s
as
w
e
l
l
as
r
e
s
pons
e
s
w
i
t
h
pr
e
d
i
c
t
i
ons
in
a
l
m
os
t
r
e
a
l
t
i
m
e
.
T
he
e
va
l
ua
t
i
on
t
i
m
e
a
nd
r
e
s
ul
t
s
f
or
bo
t
h
m
e
t
hods
w
e
r
e
ob
t
a
i
ne
d
u
s
i
ng
t
he
t
e
s
t
da
t
a
s
e
t
.
3.1.1.
E
val
u
at
i
on
r
e
s
u
l
t
s
of
A
u
t
o
G
l
u
on
S
i
nc
e
t
he
e
ns
e
m
b
l
e
a
r
c
hi
t
e
c
t
ur
e
of
A
ut
oG
l
uon
ge
ne
r
a
t
e
s
pr
e
di
c
t
i
ons
by
a
ggr
e
ga
t
i
ng
t
he
out
c
o
m
e
s
of
m
a
ny
ba
s
e
l
e
a
r
ne
r
s
,
T
he
r
e
f
o
r
e
,
T
a
bl
e
s
3
a
nd
4
pr
e
s
e
nt
t
he
l
e
a
de
r
boa
r
d
pr
e
di
c
t
i
ons
a
nd
e
va
l
u
a
t
i
on
m
e
t
r
i
c
s
,
r
e
s
pe
c
t
i
ve
l
y,
r
a
t
he
r
t
ha
n
a
s
i
ngu
l
a
r
c
onf
u
s
i
on
m
a
t
r
i
x.
T
he
w
e
i
ght
e
d
e
ns
e
m
bl
e
m
ode
l
ha
d
t
he
hi
gh
e
s
t
pe
r
f
or
m
a
nc
e
,
w
i
t
h
a
t
e
s
t
a
c
c
ur
a
c
y
of
97%
,
as
s
e
e
n
in
T
a
bl
e
3,
be
i
ng
t
he
l
a
s
t
l
e
ve
l
in
t
he
t
r
a
i
n
i
ng
or
de
r
.
T
hi
s
de
m
ons
t
r
a
t
e
s
t
ha
t
e
ns
e
m
bl
e
l
e
a
r
n
i
ng
in
t
he
s
e
c
ond
pr
e
di
c
t
i
on
l
e
ve
l
,
w
hi
c
h
c
o
m
bi
ne
s
t
he
out
pu
t
s
f
r
om
f
i
r
s
t
pr
e
di
c
t
i
on
l
e
v
e
l
m
ode
l
s
,
ope
r
a
t
e
s
by
i
nt
e
gr
a
t
i
ng
m
a
ny
s
t
r
a
t
e
gi
e
s
.
T
hi
s
f
i
na
l
e
ns
e
m
bl
e
m
e
r
ge
s
t
he
pr
e
di
c
t
i
ons
f
r
om
m
a
ny
f
ounda
t
i
ona
l
m
ode
l
s
,
w
hi
c
h
s
i
m
u
l
t
a
n
e
ous
l
y
r
e
duc
e
s
bot
h
b
i
a
s
a
nd
va
r
i
a
nc
e
.
In
t
he
f
i
r
s
t
l
e
ve
l
,
L
i
gh
t
G
B
M
,
C
a
t
B
oos
t
,
a
nd
E
xt
r
a
T
r
e
e
s
a
r
e
i
ns
t
a
nc
e
s
of
t
r
e
e
-
b
a
s
e
d
m
ode
l
s
t
ha
t
a
c
hi
e
v
e
d
s
e
c
ond
pl
a
c
e
w
i
t
h
an
a
c
c
u
r
a
c
y
of
96
%
a
f
t
e
r
t
r
a
i
ni
ng
in
t
he
t
hi
r
d,
f
ou
r
t
h,
a
nd
f
i
f
t
h
o
r
de
r
,
r
e
s
pe
c
t
i
ve
l
y.
V
a
r
i
e
t
y
of
U
R
L
,
c
ont
e
nt
,
a
nd
s
e
r
vi
c
e
-
ba
s
e
d
a
t
t
r
i
bu
t
e
s
.
T
he
i
r
e
xc
e
p
t
i
ona
l
pe
r
f
or
m
a
nc
e
ove
r
t
h
e
ne
ur
a
l
a
nd
d
i
s
t
a
nc
e
-
ba
s
e
d
m
ode
l
s
can
be
e
xpl
a
i
ne
d
by
t
he
f
a
c
t
t
h
a
t
t
he
y
c
oul
d
c
a
pt
u
r
e
c
om
p
l
e
x
r
e
l
a
t
i
ons
h
i
ps
be
t
w
e
e
n
f
e
a
t
ur
e
s
.
T
he
t
w
o
ne
xt
be
s
t
m
ode
l
s
w
e
r
e
RF
a
nd
X
G
B
oos
t
w
i
t
h
s
c
or
e
s
of
96%
e
ve
n
t
hough
t
he
y
h
a
d
s
c
or
e
d
s
e
c
ond
a
nd
s
e
ve
nt
h
in
t
r
a
i
ni
ng
o
r
de
r
.
T
h
i
s
w
a
s
b
e
c
a
us
e
t
h
e
y
u
s
e
d
an
e
ns
e
m
b
l
e
-
of
-
t
r
e
e
s
s
t
r
u
c
t
u
r
e
.
T
he
F
a
s
t
A
I
de
m
o
ns
t
r
a
t
e
d
95
%
a
c
c
u
r
a
c
y,
pl
a
c
i
ng
it
in
t
he
s
i
xt
h
o
r
de
r
of
t
r
a
i
ni
n
g
w
h
i
c
h
is
qu
i
t
e
g
ood
but
no
t
s
upe
r
i
or
c
om
p
a
r
e
d
to
t
he
o
t
he
r
m
ode
l
s
due
to
t
he
s
e
n
s
i
t
i
vi
t
y
of
t
he
f
e
a
t
u
r
e
s
to
s
c
a
l
i
ng
a
nd
t
he
op
t
i
m
i
z
a
t
i
on
of
pa
r
a
m
e
t
e
r
s
.
F
i
na
l
l
y,
t
he
f
i
r
s
t
in
t
he
l
i
ne
of
t
h
e
t
r
a
i
n
i
ng
m
ode
l
s
w
a
s
KNN,
w
hi
c
h
d
i
d
no
t
f
a
r
e
w
e
l
l
as
t
he
l
a
s
t
,
w
i
t
h
an
a
c
c
u
r
a
c
y
of
83
%
.
D
i
s
t
a
n
c
e
-
ba
s
e
d
l
e
a
r
n
i
ng
m
e
t
hodo
l
og
i
e
s
e
n
c
oun
t
e
r
d
i
f
f
i
c
u
l
t
i
e
s
w
i
t
h
h
i
gh
-
di
m
e
ns
i
ona
l
t
a
b
ul
a
r
da
t
a
w
he
n
s
e
v
e
r
a
l
f
e
a
t
u
r
e
s
e
x
e
r
t
m
i
ni
m
a
l
or
c
o
r
r
e
l
a
t
e
d
i
n
f
l
u
e
nc
e
.
T
a
bl
e
3
.
L
e
a
de
r
boa
r
d
p
r
e
di
c
t
i
on
t
a
bl
e
of
A
ut
oG
l
uon
a
l
go
r
i
t
hm
s
N
o.
A
l
gor
i
t
hm
A
c
c
ur
a
c
y
s
c
or
e
(
%
)
S
t
a
c
k
l
e
ve
l
T
r
a
i
ni
ng
or
d
e
r
1
W
e
i
ght
e
d
e
ns
e
m
bl
e
97
2
8
2
L
i
ght
G
B
M
96
1
3
3
C
a
t
B
oos
t
96
1
4
4
E
xt
r
a
T
r
e
e
s
96
1
5
5
RF
96
1
2
6
X
G
B
oos
t
96
1
7
7
F
a
s
t
A
I
95
1
6
8
KNN
83
1
1
T
a
bl
e
4. T
he
e
va
l
ua
t
i
on
m
e
t
r
i
c
s
o
f
A
ut
oG
l
uon a
f
t
e
r
ba
t
c
h t
r
a
ns
f
or
m
N
o.
M
e
t
r
i
c
V
a
l
ue
(
%
)
1
A
c
c
ur
a
c
y
97
2
P
r
e
c
i
s
i
on
97
3
R
e
c
a
l
l
96
4
F1
-
s
c
or
e
97
5
R
O
C
97
T
a
bl
e
4
s
um
m
a
r
i
z
e
s
t
he
e
va
l
ua
t
i
on
m
e
t
r
i
c
s
obt
a
i
ne
d
f
r
om
t
he
f
i
na
l
A
ut
oG
l
uon
e
ns
e
m
bl
e
on
t
he
t
e
s
t
da
t
a
s
e
t
.
T
he
w
e
i
gh
t
e
d
e
ns
e
m
b
l
e
a
c
hi
e
ve
d
t
he
gr
e
a
t
e
s
t
pe
r
f
or
m
a
nc
e
f
or
A
ut
oG
l
uon
on
t
he
s
e
c
ond
p
r
e
di
c
t
i
on
l
e
ve
l
,
w
i
t
h
99
%
R
O
C
,
97%
F1
-
s
c
or
e
,
97%
a
c
c
ur
a
c
y,
97%
pr
e
c
i
s
i
on,
a
nd
96
%
r
e
c
a
l
l
.
T
hi
s
i
m
p
l
i
e
s
t
ha
t
t
he
r
e
is
a
good
a
bi
l
i
t
y
to
di
f
f
e
r
e
n
t
i
a
t
e
be
t
w
e
e
n
phi
s
hi
ng
U
R
L
s
a
nd
l
e
gi
t
i
m
a
t
e
U
R
L
s
.
In
a
ddi
t
i
on,
i
nc
o
r
por
a
t
i
ng
bot
h
f
a
l
s
e
po
s
i
t
i
ve
s
a
nd
f
a
l
s
e
n
e
ga
t
i
ve
s
i
nt
o
a
s
i
ng
l
e
,
ba
l
a
nc
e
d
m
e
a
s
u
r
e
m
e
n
t
.
F
ur
t
he
r
m
or
e
,
t
he
e
xc
e
l
l
e
n
t
s
c
or
e
s
a
c
r
os
s
a
l
l
t
h
e
m
e
a
s
ur
e
s
i
ndi
c
a
t
e
s
t
r
ong
di
s
c
r
i
m
i
na
t
i
on
be
t
w
e
e
n
phi
s
h
i
ng
a
nd
l
e
g
i
t
i
m
a
t
e
l
a
be
l
s
a
nd
a
r
e
i
ndi
c
a
t
i
ve
of
t
he
f
a
c
t
t
ha
t
t
h
e
pe
r
f
or
m
a
nc
e
of
A
ut
oG
l
uon
’s
e
ns
e
m
bl
e
l
e
a
r
ni
ng
a
ppr
oa
c
h
w
a
s
e
xc
e
p
t
i
ona
l
l
y
w
e
l
l
on
t
h
i
s
da
t
a
s
e
t
,
as
t
he
us
a
ge
of
va
r
i
ous
a
l
gor
i
t
hm
s
is
he
l
pf
u
l
in
c
a
pt
u
r
i
ng
d
i
ve
r
s
e
da
t
a
pa
r
t
s
,
r
e
s
u
l
t
i
ng
in
s
t
r
ong
pe
r
f
or
m
a
nc
e
.
A
ut
oG
l
uon
’
s
a
u
t
om
a
t
e
d
f
e
a
t
ur
e
s
e
l
e
c
t
i
on
a
nd
hype
r
pa
r
a
m
e
t
e
r
t
uni
ng
he
l
pe
d
to
i
m
pr
ove
t
he
m
ode
l
'
s
pe
r
f
or
m
a
n
c
e
w
i
t
hout
ove
r
f
i
t
t
i
ng,
s
i
nc
e
d
i
ve
r
s
e
m
ode
l
s
in
t
he
e
ns
e
m
bl
e
m
a
y
c
ont
r
i
but
e
to
s
u
c
c
e
s
s
.
T
r
e
e
-
e
ns
e
m
bl
e
d
a
l
go
r
i
t
h
m
s
a
r
e
w
e
l
l
-
s
ui
t
e
d
to
s
t
r
uc
t
ur
e
d
d
a
t
a
,
a
nd
t
he
i
r
c
om
b
i
na
t
i
on
c
e
r
t
a
i
nl
y
i
m
pr
ove
d
ove
r
a
l
l
pe
r
f
or
m
a
nc
e
,
in
a
ddi
t
i
on
to
t
he
s
t
a
c
k
i
ng
t
e
c
hn
i
que
w
hi
c
h
m
i
ni
m
i
z
e
d
m
ode
l
va
r
i
a
nc
e
a
nd
i
nc
r
e
a
s
e
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
t
i
f
I
nt
e
l
l
I
S
S
N
:
2252
-
8938
C
om
par
i
s
on be
t
w
e
e
n e
ns
e
m
bl
e
and
l
i
ne
ar
m
e
t
hods
f
or
w
e
bs
i
t
e
ph
i
s
hi
ng d
e
t
e
c
t
i
on
(
Saba H
us
s
e
i
n R
as
h
i
d
)
689
ge
ne
r
a
l
i
z
a
t
i
on,
a
l
l
ow
i
ng
t
he
m
ode
l
to
pe
r
f
o
r
m
w
e
l
l
in
phi
s
hi
ng
c
l
a
s
s
e
s
,
a
l
ongs
i
de
t
he
ba
l
a
nc
e
d
da
t
a
s
e
t
a
nd
t
he
e
f
f
e
c
t
i
ve
r
e
gul
a
r
i
z
a
t
i
on
as
t
he
y
he
l
p
e
d
s
us
t
a
i
n
hi
gh
r
e
s
u
l
t
s
in
bot
h
a
c
c
ur
a
c
y
a
nd
r
e
c
a
l
l
.
3.1.2. E
val
u
at
i
on
r
e
s
u
l
t
s
of
l
i
n
e
ar
l
e
ar
n
e
r
B
a
s
e
d
on
t
he
c
onf
us
i
on
m
a
t
r
i
x
of
l
i
ne
a
r
l
e
a
r
n
e
r
di
s
p
l
a
ye
d
i
n
F
i
gur
e
6,
i
t
c
a
n
be
de
duc
e
d
t
ha
t
1,616
s
a
m
p
l
e
s
w
e
r
e
c
or
r
e
c
t
l
y
c
l
a
s
s
i
f
i
e
d
a
s
“
l
e
gi
t
i
m
a
t
e
,”
w
he
r
e
a
s
1,619
s
a
m
pl
e
s
w
e
r
e
a
pp
r
opr
i
a
t
e
l
y
c
l
a
s
s
i
f
i
e
d
a
s
“
phi
s
hi
ng.”
120
s
a
m
p
l
e
s
a
r
e
w
r
ongl
y
c
l
a
s
s
i
f
i
e
d
a
s
“
phi
s
hi
ng,
”
w
he
r
e
a
s
onl
y
74
s
a
m
pl
e
s
a
r
e
i
nc
o
r
r
e
c
t
l
y
c
l
a
s
s
i
f
i
e
d
a
s
“
l
e
g
i
t
i
m
a
t
e
.”
S
how
i
ng
t
ha
t
l
i
ne
a
r
l
e
a
r
ne
r
ha
s
a
l
ow
i
nc
i
de
n
c
e
of
f
a
l
s
e
po
s
i
t
i
ve
s
,
a
s
i
t
i
s
a
bl
e
t
o
a
c
c
ur
a
t
e
l
y
c
a
t
e
gor
i
z
e
a
l
a
r
ge
pr
opor
t
i
on
o
f
i
ns
t
a
nc
e
s
t
ha
t
a
r
e
pos
i
t
i
ve
.
T
h
i
s
i
nd
i
c
a
t
e
s
t
ha
t
l
i
ne
a
r
l
e
a
r
ne
r
i
s
e
f
f
e
c
t
i
ve
a
t
m
a
na
g
i
ng l
a
r
ge
da
t
a
s
e
t
s
a
nd pr
oduc
i
ng
a
c
c
ur
a
t
e
pr
e
d
i
c
t
i
ons
w
i
t
h a
l
ow
a
m
ount
o
f
m
i
s
l
a
b
e
l
i
ng. A
s
s
how
n
i
n
T
a
bl
e
5,
l
i
ne
a
r
l
e
a
r
n
e
r
ob
t
a
i
ne
d
a
n
a
c
c
u
r
a
c
y
of
94%
w
i
t
h
c
om
p
a
r
a
bl
y
hi
gh
p
r
e
c
i
s
i
on
o
f
95%
,
F1
-
s
c
or
e
of
94
%
,
R
O
C
of
94%
,
a
nd
a
r
e
c
a
l
l
of
93
%
.
T
he
f
i
nd
i
ngs
de
m
ons
t
r
a
t
e
t
ha
t
t
he
m
ode
l
i
s
e
f
f
e
c
t
i
ve
a
t
r
e
c
ogni
z
i
ng
phi
s
hi
ng
w
e
bs
i
t
e
s
w
hi
l
e
a
l
s
o
a
voi
di
ng
f
a
l
s
e
pos
i
t
i
ve
s
.
T
he
m
ode
s
t
va
r
i
a
t
i
on
be
t
w
e
e
n
a
c
c
ur
a
c
y
a
nd
r
e
c
a
l
l
i
m
pl
i
e
s
t
ha
t
t
he
m
ode
l
s
l
i
ps
up
on
t
he
s
i
de
of
c
a
ut
i
on,
f
oc
us
i
ng
on
l
ow
e
r
i
ng
f
a
l
s
e
pos
i
t
i
ve
s
w
hi
l
e
m
i
s
s
i
ng
a
f
e
w
phi
s
hi
ng
w
e
bs
i
t
e
s
.
T
he
h
i
gh
va
l
ue
of
R
O
C
i
s
a
n i
ndi
c
a
t
i
on
of
t
he
f
a
c
t
t
ha
t
t
he
m
ode
l
h
a
s
g
r
e
a
t
c
a
pa
c
i
t
y f
o
r
di
s
t
i
ngu
i
s
hi
ng be
t
w
e
e
n t
he
phi
s
h
i
ng a
nd t
he
l
e
ga
l
w
e
bs
i
t
e
s
, w
hi
c
h ha
s
be
e
n c
ons
i
de
r
e
d c
r
i
t
i
c
a
l
i
n
t
hi
s
e
nvi
r
onm
e
n
t
w
he
r
e
t
he
m
i
s
c
l
a
s
s
i
f
i
c
a
t
i
ons
m
i
ght
l
e
a
d t
o
s
e
ve
r
e
c
ons
e
que
nc
e
s
.
T
a
ki
ng
i
nt
o
c
ons
i
de
r
a
t
i
on
i
t
s
hi
gh l
e
ve
l
s
o
f
a
c
c
u
r
a
c
y,
pr
e
c
i
s
i
on,
a
nd r
e
c
a
l
l
, t
he
l
i
ne
a
r
l
e
a
r
ne
r
m
ode
l
w
oul
d
be
hi
gh
l
y
de
pe
nda
b
l
e
f
or
t
he
de
t
e
c
t
i
on of
r
e
a
l
-
w
or
l
d phi
s
h
i
ng, w
hi
c
h m
a
ke
s
i
t
a
n i
nva
l
ua
bl
e
t
ool
f
or
a
pp
l
i
c
a
t
i
on
s
of
c
ybe
r
s
e
c
ur
i
t
y.
F
i
gur
e
6
.
T
he
c
onf
us
i
on m
a
t
r
i
x of
l
i
ne
a
r
l
e
a
r
n
e
r
pr
e
d
i
c
t
i
on a
f
t
e
r
ba
t
c
h t
r
a
ns
f
o
r
m
T
a
bl
e
5
.
T
he
e
va
l
ua
t
i
on
m
e
t
r
i
c
s
o
f
l
i
n
e
a
r
l
e
a
r
ne
r
a
f
t
e
r
ba
t
c
h t
r
a
n
s
f
or
m
N
o.
M
e
t
r
i
c
V
a
l
ue
(%)
1
A
c
c
ur
a
c
y
94
2
P
r
e
c
i
s
i
on
95
3
R
e
c
a
l
l
93
4
F1
-
s
c
or
e
94
5
R
O
C
94
3.1.3.
C
om
p
ar
i
s
on
of
e
val
u
a
t
i
on
r
e
s
u
l
t
s
F
i
gur
e
7
i
l
l
us
t
r
a
t
e
s
t
he
c
o
m
pa
r
i
s
on
of
t
he
r
e
s
u
l
t
s
of
t
he
e
v
a
l
ua
t
i
on
of
bo
t
h
A
ut
oG
l
uon
a
nd
l
i
ne
a
r
l
e
a
r
ne
r
.
T
he
gr
a
ph
a
l
l
ow
s
v
i
s
ua
l
i
z
i
ng
A
ut
oG
l
uon
a
nd
l
i
ne
a
r
l
e
a
r
ne
r
in
t
e
r
m
s
of
t
he
m
a
i
n
a
s
s
e
s
s
m
e
nt
pa
r
a
m
e
t
e
r
s
.
It
w
a
s
f
ound
t
h
a
t
A
ut
oG
l
uon
in
a
l
l
f
i
ve
m
e
a
s
ur
e
s
,
a
c
c
ur
a
c
y,
p
r
e
c
i
s
i
on,
r
e
c
a
l
l
,
F1
-
s
c
or
e
,
a
nd
R
O
C
,
ha
s
a
l
w
a
ys
r
e
por
t
e
d
h
i
ghe
r
va
l
ue
s
w
he
n
c
om
pa
r
e
d
to
l
i
ne
a
r
l
e
a
r
ne
r
w
ho
ha
s
r
e
po
r
t
e
d
s
l
i
ght
l
y
l
ow
e
r
but
e
qua
l
l
y
r
e
l
i
a
bl
e
r
e
s
ul
t
s
,
w
hi
c
h
s
how
s
t
h
a
t
t
h
e
e
ns
e
m
b
l
e
m
od
e
l
ha
s
a
s
t
r
ong
ba
l
a
nc
e
be
t
w
e
e
n
t
he
de
t
e
c
t
i
on
of
phi
s
hi
ng
w
e
bs
i
t
e
s
a
nd
f
a
l
s
e
a
l
a
r
m
s
.
T
h
i
s
can
be
e
xp
l
a
i
n
e
d
by
t
he
f
a
c
t
t
ha
t
A
ut
oG
l
uon
h
a
s
a
m
e
c
ha
ni
s
m
of
c
a
pt
ur
i
ng
non
-
l
i
n
e
a
r
,
a
nd
c
om
p
l
i
c
a
t
e
d
pa
t
t
e
r
ns
in
t
he
da
t
a
.
T
he
s
t
a
c
k
i
ng
t
e
c
hn
i
que
f
ur
t
he
r
e
nha
nc
e
s
m
ode
l
ge
ne
r
a
l
i
z
a
t
i
on
a
nd
r
e
duc
e
s
va
r
i
a
nc
e
,
r
e
s
ul
t
i
ng
in
r
obus
t
p
r
e
di
c
t
i
ve
be
ha
v
i
or
a
c
r
os
s
bot
h
ph
i
s
hi
ng
a
nd
l
e
gi
t
i
m
a
t
e
c
l
a
s
s
e
s
.
L
i
ne
a
r
l
e
a
r
ne
r
,
by
c
om
pa
r
i
s
on,
c
onf
i
r
m
s
t
ha
t
t
he
m
ode
l
is
c
om
pe
t
e
nt
at
di
s
t
i
ngu
i
s
hi
ng
be
t
w
e
e
n
l
e
g
i
t
i
m
a
t
e
a
nd
m
a
l
i
c
i
ous
w
e
bs
i
t
e
s
bu
t
r
e
m
a
i
ns
l
i
m
i
t
e
d
by
i
t
s
l
i
ne
a
r
de
c
i
s
i
on
bounda
r
y,
w
hi
c
h
c
a
nnot
f
ul
l
y
r
e
p
r
e
s
e
n
t
t
he
non
-
l
i
n
e
a
r
f
e
a
t
ur
e
r
e
l
a
t
i
ons
hi
ps
t
yp
i
c
a
l
of
phi
s
hi
ng
da
t
a
.
T
he
s
l
i
ght
l
y
hi
gh
e
r
pr
e
c
i
s
i
on
r
e
l
a
t
i
ve
to
r
e
c
a
l
l
s
ugge
s
t
s
t
ha
t
l
i
n
e
a
r
l
e
a
r
ne
r
is
m
or
e
c
ons
e
r
va
t
i
ve
in
c
l
a
s
s
i
f
i
c
a
t
i
on,
pr
oduc
i
ng
f
e
w
e
r
f
a
l
s
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
t
i
f
I
nt
e
l
l
,
V
ol
. 15, N
o. 1, F
e
br
ua
r
y 2026
:
681
-
694
690
pos
i
t
i
ve
s
w
hi
l
e
m
i
s
s
i
ng
a
s
m
a
l
l
po
r
t
i
on
of
phi
s
h
i
ng
s
a
m
p
l
e
s
.
D
e
s
pi
t
e
t
he
s
e
di
f
f
e
r
e
nc
e
s
,
bot
h
a
l
go
r
i
t
hm
s
de
m
ons
t
r
a
t
e
s
t
r
ong
c
a
p
a
bi
l
i
t
y,
w
i
t
h
A
ut
oG
l
uon
f
a
vor
e
d
f
or
a
c
c
ur
a
c
y
-
c
r
i
t
i
c
a
l
a
ppl
i
c
a
t
i
ons
a
nd
l
i
ne
a
r
l
e
a
r
ne
r
m
or
e
s
u
i
t
a
b
l
e
f
or
t
i
m
e
-
s
e
ns
i
t
i
ve
or
r
e
s
ou
r
c
e
-
l
i
m
i
t
e
d
e
nvi
r
on
m
e
nt
s
.
D
ue
t
o
t
he
l
a
c
k
of
s
t
udi
e
s
ut
i
l
i
z
i
ng
S
a
ge
M
a
ke
r
’
s
A
ut
oG
l
uon
a
nd l
i
ne
a
r
l
e
a
r
ne
r
f
or
w
e
bs
i
t
e
ph
i
s
hi
ng
de
t
e
c
t
i
on a
s
m
e
n
t
i
one
d be
f
or
e
, T
a
b
l
e
6 pr
e
s
e
n
t
s
a
c
om
p
a
r
i
s
on be
t
w
e
e
n t
he
e
v
a
l
ua
t
i
on r
e
s
ul
t
s
of
A
ut
oG
l
uon i
n
t
hi
s
s
t
udy
a
nd
t
he
r
e
s
ul
t
s
o
f
t
he
i
nd
i
vi
dua
l
a
l
go
r
i
t
hm
s
i
nt
e
gr
a
t
e
d
w
i
t
h
i
n
i
t
s
a
r
c
hi
t
e
c
t
u
r
e
.
I
t
a
l
s
o
i
nc
l
ude
s
t
he
r
e
s
ul
t
s
of
l
i
ne
a
r
l
e
a
r
ne
r
a
nd a
s
i
m
i
l
a
r
a
l
gor
i
t
hm
i
n c
ons
t
r
uc
t
i
on a
nd pe
r
f
o
r
m
a
nc
e
, a
l
l
ob
t
a
i
ne
d f
r
om
t
he
r
e
l
a
t
e
d
w
or
k
m
e
nt
i
one
d
e
a
r
l
i
e
r
i
n
t
he
s
e
c
t
i
on
1
.
T
he
r
e
s
ul
t
s
de
m
ons
t
r
a
t
e
d
t
ha
t
t
he
pr
opos
e
d
m
ode
l
out
pe
r
f
o
r
m
e
d
i
n
t
e
r
m
s
of
a
c
c
u
r
a
c
y,
c
onf
i
r
m
i
ng
t
he
t
r
e
nd
t
ha
t
m
ode
l
di
ve
r
s
i
t
y
i
m
pr
ove
s
ph
i
s
hi
ng
-
de
t
e
c
t
i
on
r
obus
t
n
e
s
s
,
a
nd
va
l
i
da
t
i
ng i
t
s
e
f
f
i
c
i
e
nc
y
f
or
l
a
r
ge
w
e
bs
i
t
e
phi
s
h
i
ng us
e
c
a
s
e
s
.
F
i
gur
e
7
.
C
om
pa
r
i
s
on
of
e
v
a
l
ua
t
i
on
r
e
s
ul
t
s
f
o
r
bot
h
m
e
t
hods
a
f
t
e
r
ba
t
c
h
t
r
a
ns
f
or
m
T
a
bl
e
6
.
C
om
pa
r
i
s
on
of
e
va
l
ua
t
i
on r
e
s
u
l
t
s
w
i
t
h r
e
l
a
t
e
d w
or
ks
M
od
e
l
D
a
t
a
s
e
t
A
c
c
ur
a
c
y
(%)
E
nvi
r
onm
e
nt
R
e
f
e
r
e
nc
e
C
a
t
B
oos
t
C
I
C
-
B
e
l
l
-
DNS
90
G
oogl
e
C
ol
a
b
[
22]
F
a
s
t
A
I
I
S
C
X
-
U
R
L
96
G
oogl
e
C
ol
a
b
[
23]
L
i
ght
G
B
M
U
C
I
phi
s
hi
ng
d
om
a
i
ns
d
a
t
a
s
e
t
95
G
oogl
e
C
ol
a
b
[
24]
RF
P
hi
s
hT
a
nk
93
G
oogl
e
C
ol
a
b
[
25]
X
G
B
oos
t
P
hi
s
hO
F
E
94
J
upyt
e
r
N
ot
e
book
[
26]
E
xt
r
a
T
r
e
e
s
U
C
I
phi
s
hi
ng
d
om
a
i
ns
d
a
t
a
s
e
t
96
G
oogl
e
C
ol
a
b
[
27]
L
ogi
s
t
i
c
r
e
gr
e
s
s
i
on
C
om
bi
ne
d
of
m
ul
t
i
pl
e
K
a
ggl
e
d
a
t
a
s
e
t
s
93
A
W
S
S
a
ge
M
a
ke
r
[
28]
KNN
Web
pa
ge
phi
s
h
i
ng
d
e
t
e
c
t
i
on
83
A
W
S
S
a
ge
M
a
ke
r
[
29]
A
ut
oG
l
uon
Web
pa
ge
phi
s
h
i
ng
d
e
t
e
c
t
i
on
97
A
W
S
S
a
ge
M
a
ke
r
T
hi
s
s
t
ud
y
L
i
ne
a
r
l
e
a
r
ne
r
Web
pa
ge
phi
s
h
i
ng
d
e
t
e
c
t
i
on
94
A
W
S
S
a
ge
M
a
ke
r
T
hi
s
s
t
ud
y
3.2.
T
i
m
e
as
s
e
s
s
m
e
n
t
T
he
pa
pe
r
e
s
t
i
m
a
t
e
d
t
r
a
i
n
i
ng,
ba
t
c
h
t
r
a
ns
f
or
m
,
a
nd
pr
e
di
c
t
i
on
t
i
m
e
s
of
e
a
c
h
a
l
go
r
i
t
h
m
on
an
A
m
a
z
on
c
l
oudw
a
t
c
h
l
og,
w
hi
c
h
pr
ovi
d
e
s
va
l
ua
b
l
e
i
nf
o
r
m
a
t
i
on
on
t
he
e
f
f
e
c
t
i
ve
ne
s
s
of
pa
r
t
i
c
ul
a
r
a
l
gor
i
t
h
m
s
.
F
i
gur
e
8
gi
ve
s
r
e
s
u
l
t
s
of
t
he
t
i
m
e
e
v
a
l
ua
t
i
on.
A
ut
oG
l
uon
us
e
d
m
or
e
m
e
m
or
y
a
nd
pr
o
c
e
s
s
i
ng
a
nd
m
e
m
o
r
y
r
e
s
ou
r
c
e
s
t
ha
n
l
i
ne
a
r
l
e
a
r
ne
r
b
e
c
a
us
e
it
us
e
d
t
he
S
a
ge
M
a
ke
r
m
l
.
m
4.xl
a
r
ge
i
ns
t
a
n
c
e
,
w
hi
c
h
u
s
e
d
a
bou
t
1.3
t
i
m
e
s
m
or
e
m
e
m
or
y
a
nd
r
e
qui
r
e
d
m
o
r
e
t
i
m
e
to
t
r
a
i
n
be
c
a
u
s
e
it
is
a
m
ul
t
i
-
m
ode
l
e
ns
e
m
bl
e
a
r
c
h
i
t
e
c
t
ur
e
a
nd
op
t
i
m
i
z
e
s
t
he
r
e
s
ul
t
i
ng
a
r
c
hi
t
e
c
t
ur
e
.
T
h
i
s
e
x
t
r
a
c
om
pu
t
a
t
i
on
c
os
t
,
t
hough,
l
e
d
to
i
m
pr
ove
d
p
r
e
di
c
t
i
ve
a
c
c
ur
a
c
y
a
nd
i
nc
r
e
a
s
e
d
ge
ne
r
a
l
i
z
a
t
i
on,
w
hi
c
h
s
how
s
t
h
a
t
A
ut
oG
l
uon
is
s
u
i
t
a
b
l
e
w
he
n
l
a
r
ge
s
c
a
l
e
or
a
c
c
u
r
a
c
y
-
s
e
ns
i
t
i
ve
e
nvi
r
onm
e
n
t
s
a
r
e
ne
e
de
d.
On
t
he
ot
he
r
ha
nd,
l
i
ne
a
r
l
e
a
r
ne
r
t
r
a
i
ne
d
a
nd
i
nf
e
r
r
e
d
f
a
s
t
e
r
us
i
ng
l
e
s
s
r
e
s
ou
r
c
e
s
w
hi
c
h
va
l
i
da
t
e
d.
i
t
s
e
f
f
e
c
t
i
ve
ne
s
s
in
t
i
m
e
-
c
r
i
t
i
c
a
l
or
r
e
s
our
c
e
-
l
i
m
i
t
e
d
a
ppl
i
c
a
t
i
ons
.
T
he
s
e
r
e
s
ul
t
s
e
m
pha
s
i
z
e
t
ha
t
w
hi
l
e
A
ut
oG
l
uon
of
f
e
r
s
s
upe
r
i
or
a
c
c
u
r
a
c
y,
l
i
ne
a
r
l
e
a
r
ne
r
r
e
m
a
i
ns
m
or
e
e
f
f
i
c
i
e
n
t
a
nd
pr
a
c
t
i
c
a
l
w
he
r
e
c
om
put
a
t
i
ona
l
c
ons
t
r
a
i
n
t
s
a
r
e
a
pr
i
o
r
i
t
y.
3.2.1 T
r
ai
n
i
n
g t
i
m
e
A
s
F
i
gur
e
8
s
how
s
,
t
he
t
r
a
i
n
i
ng
t
i
m
e
of
A
ut
oG
l
uon
i
s
412.8
s
e
c
ond
s
,
w
hi
c
h
i
s
not
s
ur
p
r
i
s
i
ng
gi
ve
n
t
he
c
om
p
l
e
xi
t
y
of
t
he
A
ut
oG
l
uon f
r
a
m
e
w
or
k.
S
e
ve
r
a
l
s
t
e
ps
a
r
e
i
n
c
l
ude
d i
n
t
hi
s
f
r
a
m
e
w
or
k,
i
nc
l
udi
ng m
ode
l
s
e
l
e
c
t
i
on,
hype
r
p
a
r
a
m
e
t
e
r
opt
i
m
i
z
a
t
i
on,
a
nd
t
he
d
e
ve
l
op
m
e
nt
o
f
e
ns
e
m
bl
e
s
,
w
hi
c
h
c
a
n
e
xp
l
a
i
n t
he
t
i
m
e
c
os
t
.
M
ode
l
e
n
s
e
m
b
l
i
ng
i
s
on
e
of
t
he
s
i
gn
i
f
i
c
a
nt
va
r
i
a
bl
e
s
t
h
a
t
c
ont
r
i
but
e
t
o
t
he
m
or
e
t
i
m
e
s
pe
n
t
on
t
r
a
i
ni
ng.
A
ut
oG
l
uon
us
ua
l
l
y
t
r
a
i
ns
a
nu
m
be
r
of
m
od
e
l
s
a
nd
a
ggr
e
ga
t
e
s
t
he
i
r
pr
e
di
c
t
i
on
s
by
t
he
ba
ggi
ng
m
e
t
hod.
T
hi
s
a
l
gor
i
t
hm
r
e
qu
i
r
e
s
m
o
r
e
c
om
pu
t
a
t
i
ona
l
r
e
s
ou
r
c
e
s
a
nd t
i
m
e
t
ha
n t
r
a
i
n
i
ng a
s
i
ngl
e
m
ode
l
, w
hi
c
h i
s
not
t
he
c
a
s
e
97%
97%
96%
97%
97%
94%
95%
93%
94%
94%
90%
91%
92%
93%
94%
95%
96%
97%
98%
99%
100%
A
c
c
ur
a
c
y
P
r
e
c
i
s
i
on
R
e
c
a
l
l
F
-
1
R
O
C
P
e
r
c
e
n
t
a
g
e
M
e
t
r
i
c
s
A
ut
oG
l
uo
n
L
i
ne
a
r
L
e
a
r
ne
r
Evaluation Warning : The document was created with Spire.PDF for Python.