I
AE
S In
t
er
na
t
io
na
l J
o
urna
l o
f
Art
if
icia
l In
t
ellig
ence
(
I
J
-
AI
)
Vo
l.
15
,
No
.
5
,
Octo
b
er
2
0
2
6
,
p
p
.
4
3
0
4
~
4
3
1
4
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
5
.
p
p
4
3
0
4
-
4
3
1
4
4304
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
H
y
brid mo
del
f
o
r
driver dro
wsi
nes
s predic
tion usin
g
neuro
-
fuzzy a
nd rand
o
m forest
M
o
uh
ci
ne
L
a
k
hlo
ufi,
E
l Me
hd
i Mel
lo
uli
La
b
o
r
a
t
o
r
y
o
f
E
n
g
i
n
e
e
r
i
n
g
,
S
y
s
t
e
m
a
n
d
A
p
p
l
i
c
a
t
i
o
n
,
N
a
t
i
o
n
a
l
S
c
h
o
o
l
o
f
A
p
p
l
i
e
d
S
c
i
e
n
c
e
s,
U
n
i
v
e
r
s
i
t
y
o
f
S
i
d
i
M
o
h
a
m
e
d
B
e
n
A
b
d
e
l
l
a
h
,
F
e
z
,
M
o
r
o
c
c
o
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
an
2
1
,
2
0
2
5
R
ev
is
ed
J
u
l 2
2
,
2
0
2
6
Acc
ep
ted
Au
g
1
2
,
2
0
2
6
In
th
is
re
se
a
rc
h
,
a
n
e
w
h
y
b
ri
d
m
o
d
e
l
d
e
sig
n
e
d
wa
s
in
tr
o
d
u
c
e
d
to
id
e
n
ti
fy
d
riv
e
r
d
ro
ws
in
e
ss
b
y
i
n
teg
ra
ti
n
g
a
n
a
d
a
p
ti
v
e
n
e
u
ro
-
f
u
z
z
y
in
fe
re
n
c
e
sy
ste
m
(AN
F
IS
)
wit
h
t
h
e
ra
n
d
o
m
fo
re
st
(RF
)
m
e
th
o
d
.
Driv
e
r
d
r
o
ws
in
e
ss
is
o
n
e
o
f
th
e
m
a
jo
r
ris
k
fa
c
to
rs
f
o
r
r
o
a
d
a
c
c
id
e
n
ts
,
a
n
d
ti
m
e
ly
,
a
c
c
u
ra
te
id
e
n
ti
fica
ti
o
n
o
f
it
is
v
i
tal
fo
r
imp
r
o
v
in
g
traffic
sa
fe
ty
.
Th
e
m
a
in
e
m
p
h
a
sis
o
f
o
u
r
wo
rk
is
o
n
th
e
c
o
m
b
i
n
a
ti
o
n
o
f
AN
F
I
S
wi
th
RF
.
AN
F
I
S
u
se
s
fu
z
z
y
lo
g
ic
to
re
p
re
se
n
t
in
tri
c
a
te
sy
ste
m
s
with
u
n
c
e
rtain
ti
e
s
a
n
d
n
o
n
-
l
in
e
a
r
re
latio
n
sh
i
p
s,
o
n
th
e
o
t
h
e
r
h
a
n
d
,
RF
o
ffe
rs
p
o
we
rfu
l
c
las
sifi
c
a
ti
o
n
b
y
it
s
e
n
se
m
b
le
lea
rn
in
g
t
e
c
h
n
iq
u
e
,
p
e
rfo
rm
in
g
we
ll
in
h
ig
h
-
d
ime
n
sio
n
a
l
fe
a
tu
re
sp
a
c
e
s.
Th
is
c
o
m
b
in
a
ti
o
n
c
a
p
it
a
li
z
e
s
o
n
e
a
c
h
m
o
d
e
l'
s
stre
n
g
th
s:
AN
F
IS
f
o
r
it
s
d
y
n
a
m
ic
re
sp
o
n
d
i
n
g
t
o
u
n
c
e
rtain
d
a
ta
a
n
d
RF
f
o
r
i
ts
h
i
g
h
c
las
sifica
ti
o
n
a
c
c
u
ra
c
y
,
r
o
b
u
stn
e
ss
,
a
n
d
re
sista
n
c
e
to
o
v
e
rfit
ti
n
g
.
Th
e
re
su
lt
s
fr
o
m
th
e
d
iffere
n
t
e
x
p
e
rime
n
ts
c
o
n
firme
d
th
a
t
th
e
h
y
b
ri
d
A
NFIS
-
RF
o
u
tp
e
rf
o
rm
e
d
it
s
c
o
m
p
o
n
e
n
ts
c
o
n
sid
e
ra
b
l
y
.
It
wa
s
a
b
le
to
re
t
u
rn
a
n
e
n
h
a
n
c
e
m
e
n
t
in
a
c
c
u
ra
c
y
b
y
2
.
7
0
%
c
o
m
p
a
re
d
to
AN
F
IS
,
a
n
d
0
.
1
6
%
c
o
m
p
a
re
d
t
o
RF
,
w
h
il
e
a
t
th
e
s
a
m
e
ti
m
e
,
d
ro
p
p
i
n
g
t
h
e
m
e
a
n
sq
u
a
re
d
e
rro
r
(M
S
E)
b
y
3
3
.
4
0
%
c
o
m
p
a
re
d
to
A
NFIS
a
n
d
2
.
8
5
%
c
o
m
p
a
re
d
to
RF
.
Kn
o
w
in
g
t
h
a
t
RF
is
a
lrea
d
y
b
e
tt
e
r
t
h
a
n
o
t
h
e
r
trad
it
io
n
a
l
m
a
c
h
in
e
lea
rn
in
g
m
e
th
o
d
s
i
n
c
las
sifica
ti
o
n
,
t
h
e
se
re
su
lt
s
e
m
p
h
a
siz
e
it
s m
a
jo
r
ro
le i
n
t
h
e
e
ffe
c
ti
v
e
n
e
ss
o
f
th
e
h
y
b
ri
d
m
o
d
e
l
.
K
ey
w
o
r
d
s
:
Ad
ap
tiv
e
n
e
u
r
o
-
f
u
zz
y
in
f
er
e
n
ce
s
y
s
tem
Dr
o
wsi
n
ess
d
etec
tio
n
Facial
lan
d
m
ar
k
s
Neu
r
o
-
f
u
zz
y
s
y
s
tem
s
Per
ce
n
tag
e
o
f
e
y
e
clo
s
u
r
e
R
an
d
o
m
f
o
r
est
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Mo
u
h
cin
e
L
a
k
h
lo
u
f
i
L
ab
o
r
ato
r
y
o
f
E
n
g
i
n
ee
r
in
g
,
Sy
s
tem
an
d
Ap
p
licatio
n
,
Natio
n
a
l Sch
o
o
l o
f
A
p
p
lied
Scien
ce
s
Un
iv
er
s
ity
o
f
Sid
i M
o
h
a
m
ed
B
en
Ab
d
ellah
Fez,
Mo
r
o
cc
o
E
m
ail:
m
o
u
h
cin
e
.
lak
h
lo
u
f
i@
u
s
m
b
a.
ac
.
m
a
1.
I
NT
RO
D
UCT
I
O
N
D
r
i
v
e
r
d
r
o
w
s
i
n
e
s
s
i
s
t
h
e
to
p
c
o
n
tr
i
b
u
t
i
n
g
f
a
c
to
r
i
n
f
a
t
a
l
r
o
a
d
a
c
c
i
d
e
n
t
s
.
I
t
i
s
t
h
er
e
f
o
r
e
v
e
r
y
r
i
s
k
y
f
o
r
d
r
i
v
er
s
a
s
we
l
l
a
s
o
t
h
e
r
r
o
a
d
u
s
e
r
s
.
T
h
e
W
o
r
l
d
H
e
a
l
th
O
r
g
a
n
i
z
a
t
io
n
h
a
s
r
ec
o
g
n
i
z
ed
t
h
a
t
a
p
p
r
o
x
im
a
t
e
l
y
2
0
%
o
f
t
r
a
f
f
i
c
a
c
c
id
en
t
s
r
e
s
u
l
t
f
r
o
m
s
l
e
e
p
y
d
r
iv
i
n
g
a
n
d
,
th
e
r
e
f
o
r
e,
t
h
ey
h
av
e
m
ad
e
i
t
a
p
r
i
o
r
i
t
y
to
d
ev
e
l
o
p
d
r
o
w
s
y
s
t
a
t
e
d
e
t
ec
t
i
o
n
s
y
s
t
em
s
t
h
a
t
w
a
r
n
d
r
iv
e
r
s
b
e
f
o
r
e
th
e
y
c
a
n
l
o
s
e
c
o
n
t
r
o
l
.
N
e
x
t
g
e
n
e
r
a
t
io
n
i
n
-
v
e
h
i
c
l
e
d
r
o
w
s
in
e
s
s
m
o
n
i
t
o
r
i
n
g
s
y
s
t
em
s
a
r
e
c
ap
a
b
le
o
f
d
e
t
ec
t
i
n
g
s
u
b
t
l
e
f
e
a
t
u
r
e
s
o
f
d
r
o
w
s
i
n
e
s
s
t
o
c
a
u
t
io
n
th
e
d
r
i
v
e
r
i
n
t
i
m
e
.
Dr
o
wsi
n
ess
d
etec
tio
n
m
eth
o
d
s
ar
e
ess
en
tially
th
o
s
e
th
at
m
ea
s
u
r
e
p
h
y
s
io
lo
g
ical,
v
eh
icle
-
b
ased
,
an
d
b
eh
av
io
r
al
ch
an
g
es.
Ph
y
s
io
lo
g
ical
m
eth
o
d
s
r
ef
er
to
m
o
n
it
o
r
in
g
th
e
b
r
ain
ac
tiv
ity
(
elec
t
r
o
en
ce
p
h
alo
g
r
a
p
h
y
(
E
E
G)
)
,
h
ea
r
t
r
ate
(
elec
tr
o
ca
r
d
io
g
r
ap
h
y
(
E
C
G)
)
,
o
r
s
k
in
c
o
n
d
u
cta
n
ce
(
g
alv
an
ic
s
k
in
r
es
p
o
n
s
e
(
GSR
)
)
o
f
a
p
er
s
o
n
wh
ich
is
a
v
er
y
ac
cu
r
a
te
way
o
f
g
ath
er
in
g
d
ata
[
1
]
,
[
2
]
.
Ho
wev
e
r
,
s
u
ch
m
eth
o
d
s
a
r
e
u
s
u
ally
in
tr
u
s
iv
e
,
an
d
th
e
r
ef
o
r
e
,
it
is
n
o
t
ea
s
y
t
o
in
co
r
p
o
r
ate
th
em
i
n
ca
r
s
,
a
s
an
ex
am
p
le,
E
E
G
s
en
s
o
r
s
n
ec
ess
itate
th
at
th
e
d
r
iv
er
'
s
h
ea
d
h
as
elec
tr
o
d
es
o
n
it
wh
ich
m
ay
d
i
v
er
t
h
is
atten
tio
n
f
r
o
m
th
e
r
o
ad
[
3
]
,
[
4
]
.
Veh
icle
-
b
ase
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Hyb
r
id
mo
d
el
fo
r
d
r
iver d
r
o
w
s
in
ess
p
r
ed
ictio
n
u
s
in
g
n
eu
r
o
-
fu
z
z
y
a
n
d
r
a
n
d
o
m
fo
r
est
(
Mo
u
h
cin
e
La
kh
lo
u
fi
)
4305
m
ea
s
u
r
es
d
etec
t
th
e
s
ig
n
s
o
f
g
ettin
g
s
leep
y
wh
ile
d
r
iv
in
g
b
y
s
tu
d
y
in
g
th
e
d
r
iv
in
g
p
atter
n
s
lik
e
lan
e
d
ev
iatio
n
s
o
r
s
teer
in
g
co
r
r
ec
tio
n
s
[
5
]
,
[
6
]
.
Fin
ally
,
b
e
h
av
io
r
al
m
ea
s
u
r
es
f
o
cu
s
o
n
an
aly
zin
g
f
ac
i
al
ex
p
r
ess
io
n
s
,
ey
e
m
o
v
em
en
ts
[
7
]
,
b
lin
k
s
,
an
d
y
awn
s
to
ass
e
s
s
aler
tn
ess
.
T
h
e
tech
n
iq
u
es
ar
e
b
ased
o
n
v
is
u
a
l
p
er
ce
p
tio
n
an
d
a
ca
m
er
a,
s
o
th
e
r
e
is
n
o
r
eq
u
ir
em
en
t
f
o
r
wea
r
in
g
a
s
en
s
o
r
,
an
d
it
is
ea
s
y
to
im
p
lem
en
t
t
h
e
m
eth
o
d
s
in
th
e
r
ea
l w
o
r
ld
[
8
]
.
T
h
is
s
tu
d
y
p
r
im
ar
ily
co
n
ce
n
tr
ates
o
n
b
eh
av
io
r
al
m
ea
s
u
r
es
b
ec
au
s
e
th
ey
p
r
o
v
id
e
a
n
o
n
-
in
tr
u
s
iv
e
way
o
f
g
ettin
g
r
ea
l
-
lev
el
in
s
ig
h
ts
in
to
a
d
r
iv
e
r
'
s
wak
ef
u
ln
ess
th
r
o
u
g
h
f
ac
ial
an
d
e
y
e
m
o
v
em
e
n
t
r
ec
o
g
n
itio
n
[
9
]
.
C
o
m
p
ar
ed
to
p
h
y
s
io
lo
g
ical
m
eth
o
d
s
th
at
n
ec
ess
itate
s
en
s
o
r
s
to
b
e
attac
h
e
d
,
co
m
p
u
ter
v
i
s
io
n
m
eth
o
d
s
allo
w
d
r
iv
er
s
to
b
e
m
o
n
ito
r
ed
in
r
ea
l
-
tim
e,
th
u
s
th
ey
ar
e
t
h
e
b
est
f
it
f
o
r
r
ea
l
d
r
iv
in
g
s
itu
atio
n
s
.
S
o
m
e
o
f
t
h
e
ty
p
ical
in
d
icato
r
s
ar
e
ey
e
an
d
m
o
u
th
f
ea
tu
r
es
th
at
ca
n
s
h
o
w
if
a
p
er
s
o
n
is
s
leep
y
th
r
o
u
g
h
s
ig
n
s
lik
e
p
er
ce
n
tag
e
o
f
ey
e
clo
s
u
r
e
(
PERC
L
OS)
[
1
0
]
an
d
y
awn
in
g
.
R
e
c
e
n
t
s
t
u
d
i
e
s
d
e
m
o
n
s
t
r
a
te
t
h
e
p
o
t
e
n
t
i
al
o
f
m
a
c
h
in
e
l
ea
r
n
in
g
an
d
d
e
e
p
le
a
r
n
i
n
g
(
D
L
)
-
b
a
s
e
d
m
o
d
e
l
s
f
o
r
i
d
en
t
i
f
y
i
n
g
s
l
e
e
p
in
e
s
s
.
T
o
c
i
te
e
x
am
p
l
e
s
,
a
r
ea
l
-
t
i
m
e
ap
p
l
i
ca
t
i
o
n
o
f
a
r
e
cu
r
r
e
n
t
n
eu
r
a
l
n
e
t
wo
r
k
(
R
N
N
)
i
s
i
n
tr
o
d
u
c
ed
,
w
h
ic
h
c
a
n
c
o
n
s
t
a
n
t
ly
wa
t
c
h
a
d
r
i
v
er
an
d
ef
f
i
c
i
en
t
l
y
s
p
o
t
f
a
t
ig
u
e
[
1
1
]
.
A
p
ar
t
f
r
o
m
t
h
a
t
,
a
3
D
co
n
v
o
lu
t
i
o
n
a
l
n
eu
r
a
l
n
e
t
w
o
r
k
(
C
N
N
)
w
h
ic
h
i
d
en
t
i
f
i
e
s
f
a
c
ia
l
e
x
p
r
e
s
s
i
o
n
s
an
d
i
m
p
o
r
t
a
n
t
p
o
in
t
s
o
f
t
h
e
f
ac
e
h
a
s
b
ee
n
d
o
cu
m
en
te
d
a
s
p
e
r
f
o
r
m
i
n
g
b
e
t
t
e
r
in
t
er
m
s
o
f
a
c
cu
r
ac
y
f
o
r
d
r
o
w
s
i
n
e
s
s
d
e
t
e
c
t
i
o
n
[
1
2
]
.
A
l
t
h
o
u
g
h
l
e
s
s
ex
p
lo
r
e
d
i
n
th
i
s
f
i
e
ld
a
d
ap
t
i
v
e
n
e
u
r
o
-
f
u
zz
y
in
f
er
e
n
c
e
s
y
s
t
e
m
(
A
N
F
I
S)
i
s
w
e
ll
-
s
u
i
t
ed
f
o
r
h
a
n
d
l
i
n
g
c
o
m
p
l
ex
s
y
s
t
e
m
s
wi
t
h
u
n
c
e
r
ta
i
n
t
i
e
s
,
a
k
ey
a
s
p
ec
t
o
f
d
r
i
v
e
r
s
t
a
t
e
m
o
n
i
t
o
r
i
n
g
.
C
o
m
b
i
n
in
g
n
eu
r
a
l
n
e
t
w
o
r
k
s
a
n
d
f
u
z
z
y
lo
g
i
c,
AN
F
I
S
a
d
a
p
t
s
e
f
f
e
c
t
iv
e
l
y
to
d
r
i
v
i
n
g
d
a
t
a
v
a
r
i
a
t
io
n
s
,
a
s
s
e
en
i
n
[
1
3
]
ap
p
l
i
c
a
t
io
n
t
o
c
l
a
s
s
i
f
y
v
ig
i
l
a
n
ce
l
ev
e
l
s
(
a
l
e
r
t,
d
r
o
w
s
y
,
a
n
d
s
l
e
ep
)
f
r
o
m
E
E
G
d
a
ta
u
s
i
n
g
w
av
e
l
et
-
e
n
tr
o
p
y
f
e
a
t
u
r
e
s
.
A
d
d
i
t
io
n
a
l
ly
,
A
N
F
I
S
h
a
s
s
h
o
w
n
p
r
o
m
i
s
e
in
d
e
te
c
t
i
n
g
d
r
o
w
s
i
n
e
s
s
s
t
a
t
e
s
f
r
o
m
s
t
ee
r
in
g
d
a
ta
f
o
r
ad
ap
t
i
v
e
f
e
a
t
u
r
e
s
e
l
e
c
t
io
n
[
1
4
]
.
S
i
m
ila
r
l
y
,
r
an
d
o
m
f
o
r
e
s
t
(
RF
)
b
as
e
d
m
e
th
o
d
s
h
a
v
e
b
e
en
ef
f
ec
t
i
v
e
in
c
l
a
s
s
i
f
y
i
n
g
f
a
t
i
g
u
e
,
e
v
e
n
w
i
t
h
n
o
i
s
y
o
r
in
d
i
v
i
d
u
a
l
-
s
p
e
c
i
f
i
c
d
a
t
a
[
1
5
]
,
a
s
w
e
l
l
a
s
s
t
e
e
r
in
g
an
g
l
e
d
a
t
a
f
o
r
l
a
n
e
d
e
p
ar
t
u
r
e
d
e
t
e
c
t
io
n
r
e
l
a
t
ed
to
d
r
o
w
s
in
e
s
s
[
1
6
]
.
T
h
e
ex
p
er
im
en
tal
r
esu
lts
wer
e
ev
alu
ate
d
u
s
in
g
v
id
eo
s
f
r
o
m
t
h
e
NT
HU
-
DDD
d
atas
et
[
1
7
]
,
a
co
m
m
o
n
l
y
u
s
ed
r
ef
er
e
n
ce
in
d
r
o
wsi
n
ess
d
etec
tio
n
r
esear
ch
.
T
h
e
r
esu
lts
o
b
tain
ed
in
th
is
wo
r
k
wer
e
co
m
p
ar
ed
to
class
if
icatio
n
r
ep
o
r
ts
an
d
th
e
f
in
d
in
g
s
o
f
o
t
h
er
s
tu
d
ies
t
h
at
u
s
ed
th
e
s
am
e
d
ataset.
T
h
e
co
m
p
ar
is
o
n
b
etwe
en
th
ese
wo
r
k
s
h
ig
h
lig
h
ts
th
e
p
r
o
p
o
s
ed
m
o
d
el'
s
ca
p
ab
ilit
y
to
o
u
tp
er
f
o
r
m
th
e
p
r
ev
io
u
s
ap
p
r
o
ac
h
es
in
th
e
liter
atu
r
e.
T
h
is
p
ap
er
m
ain
ly
f
o
cu
s
es
o
n
an
in
n
o
v
ativ
e
h
y
b
r
i
d
m
o
d
el
wh
ich
h
as
b
ee
n
d
ev
el
o
p
ed
b
y
co
m
b
in
i
n
g
th
e
s
tr
en
g
th
s
o
f
ANFI
S
with
class
if
icatio
n
alg
o
r
ith
m
s
to
tack
le
b
eh
av
io
r
d
ata
u
n
ce
r
tai
n
ties
f
o
r
d
r
o
wsi
n
ess
d
etec
tio
n
.
T
h
e
m
eth
o
d
tak
e
s
ad
v
an
tag
e
o
f
ANFI
S'
s
ca
p
ab
ilit
y
to
r
e
p
r
esen
t
n
o
n
lin
ea
r
an
d
u
n
ce
r
tain
r
elatio
n
s
h
ip
s
an
d
at
t
h
e
s
am
e
tim
e,
p
r
o
v
i
d
in
g
r
eliab
le
d
e
cisi
o
n
s
with
th
e
h
elp
o
f
class
if
icatio
n
m
eth
o
d
s
.
Hen
ce
,
o
u
r
m
ain
o
b
jectiv
e
is
to
m
in
im
ize
o
u
r
p
r
e
d
ictio
n
er
r
o
r
s
,
e
n
lar
g
e
o
u
r
m
o
d
el'
s
ca
p
ab
ilit
y
to
b
e
well
g
en
er
alize
d
in
d
iv
er
s
e
d
r
i
v
in
g
en
v
ir
o
n
m
en
ts
,
an
d
e
v
en
tu
ally
co
n
tr
ib
u
te
to
th
e
b
etter
p
r
ev
en
tio
n
o
f
d
r
o
wsi
n
ess
-
r
elate
d
d
an
g
e
r
s
o
n
th
e
r
o
a
d
.
T
h
is
p
ap
er
h
as
b
ee
n
s
tr
u
ct
u
r
e
d
in
a
way
th
at:
s
ec
tio
n
1
g
iv
e
s
an
ac
co
u
n
t
o
f
t
h
e
s
ig
n
if
ica
n
c
e
o
f
d
r
iv
e
r
d
r
o
wsi
n
ess
d
etec
tio
n
,
an
d
also
it
b
r
in
g
s
a
b
r
ief
r
ev
iew
o
f
th
e
r
elate
d
r
esear
ch
.
Sectio
n
2
is
co
n
ce
r
n
e
d
with
th
e
d
etails
o
f
th
e
m
eth
o
d
o
l
o
g
y
th
at
was
u
s
ed
in
th
is
s
tu
d
y
,
a
n
d
th
e
d
ataset
th
at
was
em
p
l
o
y
ed
is
d
escr
ib
e
d
.
Sectio
n
3
is
d
ev
o
ted
to
th
e
p
r
esen
tatio
n
o
f
th
e
e
x
p
er
im
e
n
tal
r
esu
lts
an
d
an
ex
ten
s
iv
e
an
aly
s
is
is
m
ad
e
h
ig
h
lig
h
tin
g
th
e
p
er
f
o
r
m
a
n
ce
o
f
th
e
p
r
o
p
o
s
ed
h
y
b
r
id
m
o
d
e
l
in
co
m
p
ar
is
o
n
with
th
e
in
d
i
v
id
u
al
m
o
d
els
an
d
o
th
er
r
elate
d
s
tu
d
ies.
Fin
ally
,
s
ec
tio
n
4
s
u
m
s
u
p
t
h
e
p
a
p
er
t
h
r
o
u
g
h
a
d
is
cu
s
s
io
n
o
f
th
e
f
in
d
in
g
s
,
th
e
f
ir
s
t
th
r
ee
im
p
licatio
n
s
,
an
d
h
o
w
th
ey
g
u
id
e
th
e
f
u
tu
r
e
r
esear
ch
d
ir
ec
tio
n
s
.
2.
M
E
T
H
O
D
T
h
is
s
tu
d
y
p
r
o
p
o
s
es
a
m
u
lti
-
s
tep
m
eth
o
d
o
lo
g
ical
ap
p
r
o
ac
h
,
as
s
h
o
wn
in
Fig
u
r
e
1
,
wh
ich
d
etails
th
e
co
m
p
lete
p
ip
elin
e
f
o
r
d
r
iv
er
d
r
o
wsi
n
ess
d
etec
tio
n
.
T
h
e
f
r
a
m
ewo
r
k
s
tar
ts
with
f
ac
ial
lan
d
m
ar
k
ex
tr
ac
tio
n
an
d
th
e
co
m
p
u
tatio
n
o
f
b
e
h
av
io
r
al
f
ea
tu
r
es
(
ey
e
asp
ec
t
r
atio
(
E
AR
)
,
m
o
u
th
asp
ec
t
r
atio
(
MA
R
)
,
an
d
PERC
L
OS)
,
f
o
llo
wed
b
y
i
n
d
ep
e
n
d
en
t
p
r
e
d
ictio
n
s
g
en
er
ated
b
y
th
e
ANFI
S
an
d
RF
m
o
d
els.
Fin
ally
,
th
ese
p
r
ed
ictio
n
s
a
r
e
co
m
b
in
ed
th
r
o
u
g
h
a
weig
h
ted
f
u
s
io
n
s
tr
ateg
y
o
p
tim
ized
u
s
in
g
g
r
ad
ie
n
t d
escen
t.
L
ik
ewise,
ANFI
S
an
d
R
F
m
o
d
els
ca
n
ea
ch
b
e
th
o
u
g
h
t
o
f
a
s
p
er
f
o
r
m
i
n
g
th
e
f
ir
s
t
lev
el
o
f
p
r
ed
ictio
n
in
d
ep
en
d
en
tly
.
Her
e,
ANFI
S
tak
es
ad
v
an
tag
e
o
f
its
s
tr
en
g
th
in
m
o
d
ellin
g
u
n
ce
r
tain
t
ies
an
d
n
o
n
-
lin
ea
r
r
elatio
n
s
h
ip
s
th
at
ar
e
p
r
esen
t
in
th
e
d
r
o
wsi
n
ess
d
etec
tio
n
d
ata.
I
t
g
o
es
th
r
o
u
g
h
s
ev
e
r
al
s
tag
es
s
u
ch
as
f
u
zz
if
icatio
n
,
n
o
r
m
aliza
tio
n
,
a
n
d
d
ef
u
zz
if
icatio
n
to
m
ak
e
p
r
e
d
ictio
n
s
b
ased
o
n
f
u
zz
y
r
u
les.
T
h
ese
s
tag
es a
llo
w
th
e
m
o
d
el
to
u
n
d
er
s
tan
d
th
e
s
lig
h
t
ch
an
g
es
in
b
eh
a
v
io
r
th
at
in
d
icate
d
r
o
wsi
n
ess
,
an
d
ANFI
S’
p
er
f
o
r
m
a
n
ce
will b
e
ev
alu
ated
to
g
iv
e
an
i
n
itial p
r
ed
ictio
n
.
Me
an
wh
ile,
th
e
R
F
m
o
d
el
b
u
ild
s
u
p
a
c
o
llectio
n
o
f
d
e
cisi
o
n
tr
ee
s
b
y
ap
p
ly
i
n
g
t
h
e
b
o
o
ts
tr
ap
s
am
p
lin
g
an
d
f
ea
tu
r
e
s
elec
tio
n
m
eth
o
d
s
.
E
ac
h
tr
ee
will
g
iv
e
its
p
r
ed
ictio
n
,
an
d
th
en
th
ese
ar
e
co
m
b
in
ed
u
s
in
g
m
ajo
r
ity
v
o
tin
g
t
o
f
o
r
m
u
late
th
e
f
in
al
d
ec
is
io
n
.
Hen
ce
,
th
e
en
s
em
b
le
m
eth
o
d
(
E
M)
n
o
t
o
n
ly
in
cr
ea
s
es
th
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
5
,
Octo
b
er
2
0
2
6
:
4
3
0
4
-
4
3
1
4
4306
m
o
d
el'
s
s
tab
ilit
y
b
u
t
also
m
ak
es
it
m
o
r
e
e
f
f
icien
t
in
r
e
co
g
n
izin
g
d
r
o
wsi
n
ess
s
tate
s
.
R
F
'
s
p
r
ed
ictio
n
is
ev
alu
ated
as we
ll,
th
u
s
g
iv
in
g
a
co
m
p
lem
en
tar
y
o
u
tco
m
e
to
ANFI
S
'
s
.
I
n
th
e
last
p
h
ase,
th
e
ANFI
S
a
n
d
R
F
m
o
d
el
o
u
tp
u
ts
ar
e
in
teg
r
ated
th
r
o
u
g
h
a
weig
h
ted
f
u
s
i
o
n
m
o
d
el.
T
h
is
f
u
s
io
n
tak
es
ad
v
an
tag
e
o
f
th
e
two
m
o
d
els'
f
u
n
d
am
en
tal
d
if
f
er
en
ce
s
a
n
d
ad
ju
s
ts
th
eir
w
eig
h
ts
ac
co
r
d
i
n
g
ly
b
y
m
ea
n
s
o
f
a
g
r
ad
ien
t
d
esce
n
t
alg
o
r
ith
m
.
T
h
is
m
eth
o
d
d
et
er
m
in
es
th
e
o
p
tim
al
co
m
b
in
at
io
n
o
f
ANFI
S
an
d
R
F
p
r
ed
ictio
n
s
th
at
r
esu
lts
in
m
in
im
al
er
r
o
r
.
Sy
n
er
g
is
tically
,
th
e
h
y
b
r
id
ANFI
S
-
R
F
m
o
d
el
r
esu
lts
in
in
cr
ea
s
ed
p
r
ed
ictio
n
ac
cu
r
ac
y
an
d
less
m
ea
n
s
q
u
ar
ed
er
r
o
r
(
MSE
)
b
y
ef
f
ec
tiv
ely
b
alan
cin
g
ANFI
S’s
f
lex
ib
ilit
y
d
ea
lin
g
with
th
e
u
n
ce
r
tain
n
atu
r
e
o
f
d
ata
an
d
R
F’s
p
o
wer
f
u
l
class
if
icatio
n
f
ea
tu
r
es
.
T
h
er
e
f
o
r
e,
t
h
is
f
u
s
io
n
p
r
o
v
id
es
b
etter
r
esu
lts
in
d
r
iv
er
d
r
o
wsi
n
ess
d
etec
tio
n
.
Fig
u
r
e
1
.
Ov
e
r
v
iew
o
f
th
e
p
r
o
p
o
s
ed
h
y
b
r
id
ANFI
S
–
R
F f
r
am
ewo
r
k
f
o
r
d
r
i
v
er
d
r
o
wsi
n
ess
d
etec
tio
n
2
.
1
.
Da
t
a
s
et
des
cr
iptio
n a
nd
f
ea
t
ure
ex
t
r
a
ct
io
n
T
h
e
NT
HU
-
DDD
d
ataset
ai
m
s
to
ev
alu
ate
d
r
iv
er
d
r
o
wsi
n
ess
an
d
b
eh
av
io
r
(
I
Ds
0
0
1
–
0
0
6
)
ea
ch
r
ec
o
r
d
e
d
d
u
r
in
g
1
2
s
ep
ar
ate
d
r
iv
in
g
s
ess
io
n
s
u
n
d
er
d
if
f
er
en
t
co
n
d
itio
n
s
,
in
clu
d
in
g
s
ce
n
ar
io
s
with
an
d
with
o
u
t
g
lass
es.
E
ac
h
r
ec
o
r
d
in
g
is
a
n
n
o
tated
with
t
h
e
d
r
iv
er
’
s
v
ig
ila
n
ce
s
tate
(
d
r
o
wsy
o
r
n
o
n
-
d
r
o
wsy
)
an
d
ass
o
ciate
d
b
eh
av
io
r
s
s
u
ch
as
s
leep
y
C
o
m
b
in
atio
n
,
s
lo
wB
lin
k
W
ith
No
d
d
in
g
,
an
d
y
awn
in
g
,
as
illu
s
tr
ated
in
T
ab
le
1
.
I
n
to
tal,
th
is
d
ataset
co
m
p
r
is
es
6
6
,
4
9
1
r
ec
o
r
d
in
g
s
f
r
o
m
6
in
d
iv
id
u
als.
I
t
is
d
iv
id
ed
s
o
th
at
7
0
%
is
u
s
ed
f
o
r
tr
ain
in
g
an
d
th
e
r
em
ain
in
g
3
0
%
f
o
r
test
in
g
.
B
y
an
aly
zin
g
th
is
d
ata
co
llectio
n
,
o
n
e
ca
n
g
ai
n
d
ee
p
in
s
ig
h
t
in
to
th
e
b
eh
av
io
r
al
ch
a
n
g
es
r
elate
d
to
d
r
o
wsi
n
ess
,
an
d
it
also
f
o
r
m
s
a
s
tr
o
n
g
b
asis
f
o
r
th
e
cr
ea
tio
n
o
f
p
r
ed
ictiv
e
m
o
d
els th
at
s
ee
k
to
im
p
r
o
v
e
r
o
ad
s
af
ety
.
T
h
e
f
ac
ial
la
n
d
m
ar
k
-
b
ased
d
r
o
wsi
n
ess
d
etec
tio
n
s
y
s
tem
i
s
an
ess
en
tial
elem
en
t
o
f
th
e
r
esear
ch
r
ep
o
r
ted
i
n
th
is
p
ap
er
.
T
o
lo
ca
te
th
e
f
ac
e
f
ir
s
t,
th
e
au
th
o
r
s
u
s
ed
th
e
Dlib
lib
r
ar
y
th
at
u
tili
ze
s
a
p
r
etr
ain
ed
h
is
to
g
r
am
o
f
o
r
ien
ted
g
r
ad
ien
ts
(
HOG)
d
etec
to
r
alo
n
g
with
a
lin
ea
r
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
in
e
(
SVM
)
.
Af
ter
r
ec
o
g
n
izin
g
th
e
f
ac
e
r
eg
io
n
,
6
8
f
ac
ial
lan
d
m
ar
k
s
r
ef
er
r
in
g
t
o
th
e
two
ey
es
an
d
m
o
u
th
am
o
n
g
o
th
e
r
k
ey
f
ac
ial
co
m
p
o
n
en
ts
,
wer
e
m
ap
p
ed
f
o
r
ea
ch
f
r
am
e
.
Fro
m
th
ese
lan
d
m
ar
k
s
,
it
is
p
o
s
s
ib
le
to
ca
lcu
late
th
r
ee
b
eh
av
io
r
al
c
h
ar
ac
ter
is
tics
:
E
AR
,
MA
R
,
an
d
PERC
L
OS.
T
h
e
g
r
ea
t
ad
v
an
tag
e
o
f
th
ese
f
ea
tu
r
es
is
th
at
t
h
ey
g
iv
e
d
r
i
v
er
s
'
d
r
o
wsi
n
ess
n
o
n
-
in
t
r
u
s
iv
ely
an
d
r
eliab
ly
.
T
h
e
E
AR
is
a
s
ca
lar
v
alu
e
th
at
ex
p
lain
s
h
o
w
o
p
en
o
r
clo
s
ed
th
e
ey
es a
r
e
[
1
8
]
,
[
1
9
]
.
W
h
en
a
p
er
s
o
n
is
s
leep
y
,
ey
e
E
AR
v
alu
e
s
lo
wly
b
ec
o
m
es
ze
r
o
.
T
h
e
E
AR
v
al
u
e
f
o
r
a
n
ey
e
is
m
ea
s
u
r
e
d
two
tim
es
f
o
r
b
o
th
th
e
r
ig
h
t a
n
d
lef
t e
y
e
b
y
u
s
in
g
E
u
c
lid
ea
n
d
is
tan
ce
s
b
etwe
en
p
ar
ti
cu
lar
ey
e
lan
d
m
ar
k
s
as p
er
(
1
)
.
_
=
‖
38
−
42
‖
+
‖
39
−
41
‖
2
‖
37
−
40
‖
an
d
_
=
‖
44
−
48
‖
+
‖
45
−
47
‖
2
‖
43
−
46
‖
(
1
)
T
h
e
f
in
al
E
AR
v
alu
e
is
o
b
tain
ed
b
y
a
v
er
ag
in
g
th
e
r
ig
h
t a
n
d
l
ef
t e
y
e
r
atio
s
as d
ef
in
e
d
in
(
2
)
.
=
_
+
_
2
(
2
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Hyb
r
id
mo
d
el
fo
r
d
r
iver d
r
o
w
s
in
ess
p
r
ed
ictio
n
u
s
in
g
n
eu
r
o
-
fu
z
z
y
a
n
d
r
a
n
d
o
m
fo
r
est
(
Mo
u
h
cin
e
La
kh
lo
u
fi
)
4307
T
h
e
MA
R
aim
s
to
ass
ess
m
o
u
th
o
p
en
in
g
an
d
is
p
ar
tic
u
lar
ly
ef
f
ec
tiv
e
f
o
r
y
aw
n
in
g
d
etec
tio
n
[
2
0
]
.
MA
R
is
co
m
p
u
ted
u
s
in
g
d
is
tan
ce
s
b
et
wee
n
s
elec
ted
m
o
u
th
lan
d
m
ar
k
s
as (
3
)
.
=
‖
62
−
68
‖
+
‖
63
−
67
‖
+
‖
64
−
66
‖
3
‖
61
−
65
‖
(
3
)
T
ab
le
1
.
Sam
p
les f
r
o
m
th
e
NT
HU
-
DDD
d
ataset
illu
s
tr
at
in
g
d
r
o
wsi
n
ess
b
eh
av
io
r
s
ID
S
t
a
t
u
s
C
o
n
d
i
t
i
o
n
B
e
h
a
v
i
o
r
B
e
h
a
v
i
o
r
a
l
s
a
m
p
l
e
s
D
r
o
w
sy
G
l
a
sses
S
l
e
e
p
y
c
o
m
b
i
n
a
t
i
o
n
S
l
o
w
b
l
i
n
k
w
i
t
h
n
o
d
d
i
n
g
Y
a
w
n
i
n
g
N
o
g
l
a
s
ses
S
l
e
e
p
y
c
o
m
b
i
n
a
t
i
o
n
S
l
o
w
b
l
i
n
k
w
i
t
h
n
o
d
d
i
n
g
Y
a
w
n
i
n
g
N
o
d
r
o
w
s
y
G
l
a
sses
S
l
e
e
p
y
c
o
m
b
i
n
a
t
i
o
n
S
l
o
w
b
l
i
n
k
w
i
t
h
n
o
d
d
i
n
g
Y
a
w
n
i
n
g
N
o
g
l
a
s
ses
S
l
e
e
p
y
c
o
m
b
i
n
a
t
i
o
n
S
l
o
w
b
l
i
n
k
w
i
t
h
n
o
d
d
i
n
g
Y
a
w
n
i
n
g
I
n
a
d
d
i
t
i
o
n
,
P
E
R
C
L
OS
is
o
b
t
a
in
e
d
b
y
c
a
l
c
u
l
a
ti
n
g
t
h
e
r
a
ti
o
o
f
f
r
a
m
e
s
i
n
w
h
i
c
h
t
h
e
E
A
R
r
e
m
ai
n
s
b
e
l
o
w
a
p
r
e
d
e
f
i
n
e
d
e
y
e
-
c
l
o
s
u
r
e
t
h
r
esh
o
l
d
o
v
e
r
a
s
p
e
c
i
f
i
e
d
o
b
s
e
r
v
a
ti
o
n
i
n
t
e
r
v
a
l
,
i
n
d
i
c
at
i
n
g
p
a
r
t
i
a
l
o
r
f
u
l
l
e
y
e
c
l
o
s
u
r
e
[
2
1
]
.
P
E
R
C
L
OS
p
r
o
v
i
d
e
s
a
r
o
b
u
s
t
t
e
m
p
o
r
a
l
m
e
a
s
u
r
e
o
f
p
r
o
l
o
n
g
e
d
e
y
e
c
l
o
s
u
r
e
a
n
d
r
e
d
u
c
e
d
a
l
e
r
t
n
es
s
.
S
p
e
c
i
f
i
ca
l
l
y
,
P
E
R
C
L
O
S is
o
b
ta
i
n
e
d
b
y
a
p
p
l
y
i
n
g
a
f
i
x
e
d
t
h
r
e
s
h
o
l
d
T
=
0
.
2
0
t
o
t
h
e
f
r
a
m
e
-
l
e
v
e
l
E
AR
s
i
g
n
a
l (
E
AR
<
T
i
n
d
i
c
a
ti
n
g
a
c
l
o
s
e
d
e
y
e
)
,
co
n
s
i
s
t
e
n
t
wi
t
h
t
h
e
E
A
R
f
o
r
m
u
l
a
t
i
o
n
o
f
S
o
u
k
u
p
o
v
á
[
1
8
]
[
1
9
]
,
a
n
d
c
o
m
p
u
t
i
n
g
t
h
e
p
r
o
p
o
r
t
i
o
n
o
f
c
l
o
s
e
d
-
e
y
e
f
r
a
m
es
wi
t
h
i
n
a
s
l
i
d
i
n
g
w
i
n
d
o
w
o
f
N
=
3
0
c
o
n
s
e
c
u
t
i
v
e
f
r
a
m
e
s
(
a
p
p
r
o
x
i
m
a
t
e
l
y
1
s
e
c
o
n
d
a
t
t
h
e
d
a
t
as
e
t'
s
n
a
ti
v
e
f
r
a
m
e
r
a
te
)
,
a
d
v
a
n
c
e
d
f
r
a
m
e
b
y
f
r
a
m
e
t
o
p
r
o
d
u
c
e
a
c
o
n
t
i
n
u
o
u
s
s
i
g
n
a
l
.
T
h
e
NT
HU
-
DDD
d
ataset
u
s
e
d
in
t
h
is
s
tu
d
y
is
cu
r
ated
a
n
d
an
n
o
tated
.
T
h
er
e
f
o
r
e,
n
o
m
is
s
in
g
v
alu
es
wer
e
o
b
s
er
v
ed
in
th
e
ex
tr
ac
t
ed
f
ea
tu
r
es.
T
o
m
itig
ate
n
o
is
e
in
tr
o
d
u
ce
d
b
y
f
ac
ial
lan
d
m
ar
k
d
etec
tio
n
an
d
in
ter
-
s
u
b
ject
v
ar
iab
ilit
y
,
all
f
ea
tu
r
es
(
E
AR
,
MA
R
,
an
d
PER
C
L
OS)
wer
e
n
o
r
m
alize
d
u
s
in
g
m
in
–
m
ax
s
ca
lin
g
to
m
ap
th
eir
v
alu
es
in
t
o
th
e
r
a
n
g
e
[
0
,
1
]
.
T
h
is
n
o
r
m
aliza
tio
n
was
ap
p
lied
g
lo
b
ally
ac
r
o
s
s
th
e
p
o
o
led
r
ec
o
r
d
i
n
g
s
o
f
all
s
u
b
jects,
an
d
n
o
ad
d
itio
n
al
tem
p
o
r
al
s
m
o
o
th
i
n
g
was
ap
p
lied
to
th
e
f
r
a
m
e
-
lev
el
E
A
R
an
d
MA
R
s
ig
n
als
b
ey
o
n
d
th
is
s
ca
lin
g
,
as PE
R
C
L
OS a
lr
ea
d
y
p
r
o
v
id
es tem
p
o
r
al
ag
g
r
eg
atio
n
o
f
ey
e
-
clo
s
u
r
e
in
f
o
r
m
atio
n
.
T
h
e
n
o
r
m
alize
d
f
ea
tu
r
es
wer
e
n
ex
t
ex
am
in
e
d
f
r
am
e
b
y
f
r
am
e
an
d
d
ir
ec
tly
u
s
ed
as
in
p
u
ts
to
th
e
lear
n
in
g
m
o
d
els.
I
n
th
e
ca
s
e
o
f
th
e
ANFI
S
m
o
d
el
,
th
e
ch
ar
a
cter
is
tics
wer
e
f
u
zz
if
ied
b
y
m
ea
n
s
o
f
g
en
er
alize
d
b
ell
-
s
h
ap
ed
m
em
b
er
s
h
i
p
f
u
n
ct
io
n
s
f
o
r
th
e
r
ep
r
esen
tatio
n
o
f
u
n
ce
r
tain
ties
an
d
n
o
n
lin
ea
r
r
elatio
n
s
h
ip
s
.
On
th
e
o
th
er
h
a
n
d
,
f
o
r
th
e
RF
class
if
ier
,
th
e
s
am
e
n
o
r
m
alize
d
f
ea
t
u
r
es
wer
e
g
iv
en
as
n
u
m
er
ical
in
p
u
ts
with
o
u
t
an
y
f
u
r
th
er
tr
a
n
s
f
o
r
m
atio
n
.
T
h
is
c
o
m
m
o
n
p
r
ep
r
o
ce
s
s
in
g
tactic
g
u
ar
an
tees
n
u
m
e
r
ical
s
tab
ilit
y
d
u
r
in
g
t
r
ain
in
g
a
n
d
allo
ws f
o
r
a
f
air
c
o
m
p
ar
is
o
n
b
etwe
en
m
o
d
els.
2
.
2
.
M
o
del t
ra
ini
ng
pro
ce
du
re
T
h
e
ANFI
S
m
o
d
el
,
b
ased
o
n
J
an
g
'
s
wo
r
k
in
1
9
9
3
[
2
2
]
u
tili
ze
s
a
n
eu
r
al
n
etwo
r
k
c
o
m
b
in
e
d
with
f
u
zz
y
lo
g
ic
u
s
in
g
T
ak
ag
i
-
Su
g
en
o
-
K
an
g
(
T
SK)
"if
-
th
en
"
r
u
les
[
2
3
]
.
T
h
e
m
o
d
el'
s
in
p
u
t
lay
er
r
ec
eiv
es
th
e
th
r
ee
b
eh
av
io
r
al
d
escr
ip
to
r
s
,
E
AR
,
MA
R
,
an
d
PER
C
L
OS,
an
d
it
is
f
o
llo
wed
b
y
a
f
u
zz
if
icatio
n
lay
er
wh
er
e
ea
ch
in
p
u
t
is
ass
o
ciate
d
with
th
r
e
e
lin
g
u
is
tic
ter
m
s
(
lo
w,
m
e
d
iu
m
,
an
d
h
ig
h
)
u
s
in
g
g
en
e
r
alize
d
b
ell
-
s
h
ap
ed
m
em
b
er
s
h
ip
f
u
n
ctio
n
s
(
g
b
ell
m
f
)
,
as
d
ep
icted
in
Fig
u
r
e
2
.
T
h
e
p
ar
am
eter
s
o
f
th
ese
m
em
b
er
s
h
ip
f
u
n
ctio
n
s
ar
e
b
ein
g
o
p
tim
ized
as
tr
ain
in
g
p
r
o
g
r
ess
es
u
s
in
g
th
e
Ad
am
alg
o
r
ith
m
to
en
h
an
ce
t
h
e
ab
ilit
y
o
f
th
e
m
o
d
el
to
ad
ap
t
to
d
ata
v
ar
iab
ilit
y
.
T
h
e
f
u
zz
if
ied
in
p
u
ts
ar
e
th
en
co
m
b
in
e
d
in
th
e
r
u
le
lay
er
to
p
r
o
d
u
ce
2
7
f
u
zz
y
r
u
les
th
at
co
v
er
all
p
o
s
s
ib
le
co
m
b
in
atio
n
s
o
f
th
e
th
r
ee
lin
g
u
is
tic
s
tates
o
f
th
e
th
r
ee
in
p
u
t
v
ar
iab
les.
T
h
e
f
ir
in
g
s
tr
en
g
th
o
f
ev
er
y
r
u
le,
wh
ich
is
co
m
p
u
ted
b
y
m
u
ltip
ly
in
g
th
e
m
em
b
er
s
h
ip
d
e
g
r
ee
s
o
f
th
e
co
r
r
e
s
p
o
n
d
in
g
in
p
u
ts
,
is
p
r
esen
ted
in
(
4
)
.
Def
u
zz
if
icatio
n
o
u
t
p
u
ts
ar
e
co
m
p
u
ted
as (
5
)
.
=
(
_
)
.
(
_
)
.
(
_
)
(
4
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
5
,
Octo
b
er
2
0
2
6
:
4
3
0
4
-
4
3
1
4
4308
=
(
_
)
+
(
_
)
+
(
_
)
+
(
5
)
T
h
e
R
F
m
o
d
el
[
2
4
]
is
an
e
n
s
em
b
le
lear
n
in
g
m
eth
o
d
i
n
m
a
ch
in
e
lear
n
i
n
g
th
at
cr
ea
tes
a
m
o
d
el
b
y
ag
g
r
eg
atin
g
s
ev
er
al
d
ec
is
io
n
tr
ee
s
to
im
p
r
o
v
e
th
e
p
r
ed
icti
o
n
,
ac
cu
r
ac
y
,
an
d
r
eliab
ilit
y
o
f
th
e
m
o
d
el.
E
ac
h
d
ec
is
io
n
tr
ee
in
t
h
e
f
o
r
est
is
li
k
e
a
f
lo
wch
ar
t
with
a
well
-
o
r
g
an
ized
s
tr
u
ctu
r
e
wh
e
r
e
th
e
b
r
a
n
ch
es
r
ep
r
esen
t
th
e
p
o
s
s
ib
le
d
ec
is
io
n
s
o
r
ac
tio
n
s
t
h
at
lead
to
th
e
s
p
ec
if
ic
o
u
tc
o
m
es.
RF
is
b
ased
o
n
t
h
e
co
lle
ctiv
e
p
o
wer
o
f
m
an
y
tr
ee
s
,
with
ea
ch
tr
ee
b
ein
g
tr
ain
ed
o
n
a
d
if
f
er
en
t
s
u
b
s
et
o
f
th
e
d
ata,
wh
ich
b
r
in
g
s
ab
o
u
t
s
o
m
e
v
ar
iatio
n
am
o
n
g
th
em
an
d
th
u
s
r
ed
u
ce
s
th
e
ch
a
n
ce
o
f
o
v
er
f
itti
n
g
.
T
h
is
d
iv
er
s
ity
en
h
an
ce
s
th
e
m
o
d
el'
s
ab
ilit
y
to
g
en
er
alize
well
o
n
n
ew
d
ata.
T
h
is
wo
r
k
im
p
lem
en
ts
an
RF
cla
s
s
if
ier
co
m
p
o
s
ed
o
f
1
0
0
tr
ee
s
,
p
r
o
v
id
i
n
g
a
g
o
o
d
tr
ad
e
-
o
f
f
b
etwe
en
co
m
p
u
tatio
n
al
ef
f
icie
n
cy
an
d
p
r
e
d
ictio
n
s
tab
ilit
y
.
Fig
u
r
e
2
.
Stru
ctu
r
e
o
f
ANFI
S with
E
AR
,
MA
R
,
an
d
PERC
L
OS in
p
u
ts
2
.
3
.
M
o
del f
us
io
n m
et
ho
do
l
o
g
y
Gr
ad
ien
t
d
escen
t
is
em
p
l
o
y
e
d
to
d
ete
r
m
in
e
th
e
o
p
tim
al
v
alu
es
o
f
th
e
weig
h
tin
g
c
o
ef
f
i
cien
ts
.
T
h
e
co
m
b
in
ed
p
r
ed
ictio
n
m
o
d
el
with
th
e
b
ias co
m
p
o
n
en
t is ex
p
r
e
s
s
ed
in
(
6
)
.
=
1
⋅
+
2
⋅
+
(
6
)
T
h
e
f
u
s
ed
p
r
ed
ictio
n
is
co
m
p
u
ted
b
y
lin
ea
r
ly
c
o
m
b
in
i
n
g
th
e
o
u
tp
u
ts
o
f
th
e
ANFI
S a
n
d
RF
m
o
d
els u
s
in
g
th
e
weig
h
tin
g
c
o
ef
f
icien
ts
1
,
an
d
2
,
with
ac
tin
g
as
th
e
b
ias
p
a
r
am
eter
.
Du
r
in
g
tr
ai
n
in
g
,
th
ese
p
a
r
am
eter
s
ar
e
iter
ativ
ely
ad
ju
s
ted
to
m
in
im
iz
e
th
e
MSE
b
etwe
en
th
e
p
r
ed
ic
ted
an
d
ac
tu
al
o
u
tp
u
ts
,
as f
o
r
m
u
lated
in
(
7
)
.
=
1
∑
(
(
)
−
(
)
)
2
=
1
(
7
)
T
h
e
weig
h
ts
1
,
2
,
an
d
th
e
b
ias
ar
e
u
p
d
ated
at
ea
ch
iter
atio
n
o
f
g
r
ad
ien
t d
escen
t a
s
(
8
)
.
1
=
1
−
1
;
2
=
2
−
2
;
=
−
(
8
)
W
h
er
e
is
th
e
lear
n
in
g
r
ate.
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
3
.
1
.
E
v
a
lua
t
ing
m
o
del per
f
o
rm
a
nce
I
n
o
r
d
er
t
o
th
o
r
o
u
g
h
ly
ass
es
s
h
o
w
well
o
u
r
m
o
d
els
ca
n
d
etec
t
d
r
iv
er
d
r
o
wsi
n
ess
,
we
em
p
lo
y
e
d
s
ev
er
al
p
er
f
o
r
m
an
ce
m
etr
ics.
MSE
r
ep
o
r
ts
th
e
m
ea
n
s
q
u
ar
ed
d
e
v
iatio
n
o
f
p
r
e
d
ictio
n
s
f
r
o
m
g
r
o
u
n
d
-
tr
u
th
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Hyb
r
id
mo
d
el
fo
r
d
r
iver d
r
o
w
s
in
ess
p
r
ed
ictio
n
u
s
in
g
n
eu
r
o
-
fu
z
z
y
a
n
d
r
a
n
d
o
m
fo
r
est
(
Mo
u
h
cin
e
La
kh
lo
u
fi
)
4309
v
alu
es
[
2
5
]
,
wh
ile
r
o
o
t
m
ea
n
s
q
u
ar
ed
er
r
o
r
(
R
MSE
)
g
iv
es
a
m
o
r
e
r
elata
b
le
m
ea
s
u
r
e
o
f
th
is
d
if
f
er
en
ce
,
an
d
m
ea
n
ab
s
o
lu
te
er
r
o
r
(
MA
E
)
m
ea
s
u
r
es
th
e
av
er
ag
e
ab
s
o
lu
t
e
d
ev
iatio
n
.
R
ec
all
an
d
F1
-
s
co
r
e
wer
e
ev
alu
ated
to
em
p
h
asize
r
ec
all
o
f
d
r
o
wsy
s
t
ates
an
d
th
e
p
r
ec
is
io
n
–
r
ec
all
b
alan
ce
,
r
esp
ec
tiv
el
y
,
wh
ile
a
r
ea
u
n
d
er
t
h
e
cu
r
v
e
(
AUC
)
ass
es
s
ed
th
e
m
o
d
els'
ab
ilit
y
to
d
is
tin
g
u
is
h
class
es
r
eg
ar
d
less
o
f
th
e
d
ec
is
io
n
th
r
esh
o
ld
;
ac
c
u
r
ac
y
q
u
an
tifie
d
th
e
o
v
er
all
p
r
o
p
o
r
t
io
n
o
f
co
r
r
ec
tly
class
if
ied
s
a
m
p
les.
T
h
ese
p
e
r
f
o
r
m
an
ce
m
etr
ics
s
h
o
u
ld
n
o
t
b
e
co
n
f
u
s
ed
with
th
e
l
o
s
s
f
u
n
ctio
n
s
u
s
ed
f
o
r
tr
ai
n
in
g
,
wh
ich
d
e
p
en
d
o
n
th
e
m
o
d
el
ty
p
e:
ANFI
S
was
tr
ain
ed
b
y
m
in
im
izin
g
MA
E
,
w
h
er
ea
s
th
e
DL
m
o
d
els
⎯
ar
tific
ial
n
eu
r
a
l
n
etwo
r
k
(
ANN
)
,
C
NN,
lo
n
g
s
h
o
r
t
-
ter
m
m
e
m
o
r
y
(
L
STM
)
,
b
id
ir
ec
tio
n
al
lo
n
g
s
h
o
r
t
-
ter
m
m
e
m
o
r
y
(
B
L
STM
)
,
g
ated
r
ec
u
r
r
en
t
u
n
it
(
G
R
U
)
⎯
wer
e
tr
ain
ed
th
r
o
u
g
h
b
in
ar
y
c
r
o
s
s
-
en
tr
o
p
y
lo
s
s
.
All
co
m
p
ar
ativ
e
r
esu
lt
s
ar
e
b
ased
ex
clu
s
iv
ely
o
n
p
er
f
o
r
m
a
n
ce
m
etr
ics
o
b
tain
ed
o
n
th
e
test
d
ata,
i
n
d
ep
en
d
e
n
t
o
f
th
e
tr
ain
i
n
g
p
r
o
ce
s
s
;
MA
E
an
d
MSE
ar
e
r
ep
o
r
ted
o
n
l
y
af
ter
th
r
esh
o
ld
in
g
th
e
p
r
o
b
ab
ilis
tic
o
u
tp
u
ts
,
q
u
a
n
tify
in
g
d
ec
is
io
n
-
er
r
o
r
s
ize,
wh
ile
class
if
icati
o
n
m
etr
ics
i
n
d
icate
o
v
er
all
m
o
d
el
ef
f
icien
c
y
.
T
h
r
ee
m
o
d
el
t
y
p
es
wer
e
test
ed
,
n
a
m
ely
f
u
zz
y
s
y
s
tem
(
FS
)
,
DL
,
a
n
d
E
M,
with
h
y
p
er
p
ar
am
eter
s
ca
r
ef
u
lly
t
u
n
ed
f
o
r
th
e
b
est
p
e
r
f
o
r
m
a
n
ce
a
n
d
least
tr
ai
n
in
g
f
l
u
ctu
atio
n
.
R
ath
er
th
a
n
r
el
y
in
g
o
n
m
ass
iv
ely
d
ee
p
ar
ch
itectu
r
es,
we
f
av
o
r
e
d
s
im
p
ler
m
o
d
els
to
lim
it
o
v
er
f
itti
n
g
f
r
o
m
ex
ce
s
s
p
ar
am
eter
s
.
All
FS
an
d
DL
m
o
d
els
wer
e
o
p
tim
ized
o
v
er
1
0
0
tr
ai
n
in
g
e
p
o
ch
s
with
Ad
am
[
2
6
]
,
ch
o
s
en
f
o
r
its
ef
f
icien
t
co
n
v
er
g
en
ce
an
d
s
tab
le
lear
n
in
g
r
ates.
B
in
ar
y
cr
o
s
s
-
en
tr
o
p
y
was
u
s
ed
f
o
r
s
o
m
e
m
o
d
els
,
p
ar
ticu
lar
l
y
f
o
r
b
i
n
ar
y
class
if
icatio
n
,
as
it
im
p
r
o
v
e
d
class
s
ep
ar
atio
n
(
d
r
o
wsy
v
s
.
n
o
n
-
d
r
o
wsy
)
an
d
r
ed
u
ce
d
m
is
class
if
icatio
n
.
A
b
atch
s
ize
o
f
1
6
was
u
s
ed
in
s
o
m
e
ca
s
es,
alo
n
g
with
d
r
o
p
o
u
t
r
eg
u
lar
izatio
n
to
lim
it
o
v
er
f
itti
n
g
,
wh
ile
'
r
elu
'
an
d
'
s
ig
m
o
id
'
ac
tiv
atio
n
s
in
d
en
s
e
lay
er
s
im
p
r
o
v
ed
th
e
n
etwo
r
k
'
s
ab
ilit
y
to
ca
p
tu
r
e
n
o
n
lin
ea
r
ities
.
T
ab
le
2
s
h
o
ws
th
e
r
ep
r
esen
tativ
e
p
ar
am
eter
s
o
f
ea
c
h
m
o
d
el,
r
ef
lectin
g
t
u
n
in
g
s
p
ec
ialized
to
ea
ch
alg
o
r
ith
m
'
s
r
eq
u
ir
em
e
n
ts
.
T
h
e
h
y
p
er
p
ar
a
m
eter
s
s
u
m
m
ar
ized
in
T
ab
le
2
wer
e
d
ete
r
m
in
ed
em
p
ir
icall
y
,
th
r
o
u
g
h
iter
ativ
e
m
an
u
al
ex
p
er
im
en
tatio
n
g
u
i
d
ed
b
y
p
r
elim
in
ar
y
tr
ials
an
d
v
alu
es
co
m
m
o
n
ly
r
e
p
o
r
ted
in
th
e
liter
atu
r
e
f
o
r
co
m
p
ar
ab
le
task
s
,
r
ath
er
th
a
n
a
f
o
r
m
al
g
r
id
-
s
ea
r
ch
o
r
cr
o
s
s
-
v
alid
atio
n
p
r
o
ce
d
u
r
e.
Fo
r
A
NFI
S,
MA
E
s
er
v
ed
b
o
th
as
th
e
tr
ain
in
g
lo
s
s
an
d
a
s
an
ev
alu
atio
n
m
etr
ic,
ch
o
s
en
f
o
r
its
r
o
b
u
s
tn
ess
to
o
u
tlier
s
an
d
s
tab
le
g
r
ad
ien
t
b
eh
av
io
r
;
it
was
c
o
m
b
in
e
d
wi
th
MSE
an
d
R
MSE
as
e
v
alu
a
tio
n
m
etr
ics
f
o
r
a
f
u
ller
ass
es
s
m
en
t
o
f
p
r
e
d
ictio
n
er
r
o
r
s
an
d
ea
s
ier
co
m
p
a
r
is
o
n
a
cr
o
s
s
m
o
d
els.
T
ab
le
2
.
Hy
p
er
p
ar
a
m
eter
s
o
f
e
ac
h
m
o
d
el
M
o
d
e
l
t
y
p
e
M
o
d
e
l
H
y
p
e
r
p
a
r
a
me
t
e
r
s
FS
A
N
F
I
S
n
_
i
n
p
u
t
:
3
,
n
_
mem
b
:
3
,
mem
b
_
f
u
n
c
:
g
b
e
l
l
mf
,
o
p
t
i
mi
z
e
r
:
a
d
a
m,
l
o
ss
:
M
e
a
n
A
b
so
l
u
t
e
Er
r
o
r
,
n
_
e
p
o
c
h
s:
1
0
0
D
e
e
p
n
e
u
r
a
l
n
e
t
w
o
r
k
s
ANN
l
a
y
e
r
s:
[
D
e
n
se(
3
2
,
r
e
l
u
)
,
D
e
n
s
e
(
1
6
,
r
e
l
u
)
,
D
e
n
s
e
(
1
,
s
i
g
mo
i
d
)
]
,
l
o
s
s:
b
i
n
a
r
y
_
c
r
o
ss
e
n
t
r
o
p
y
,
o
p
t
i
m
i
z
e
r
:
a
d
a
m,
m
e
t
r
i
c
s
:
a
c
c
u
r
a
c
y
,
e
p
o
c
h
s:
1
0
0
,
b
a
t
c
h
_
si
z
e
:
1
6
,
v
a
l
i
d
a
t
i
o
n
_
s
p
l
i
t
:
0
.
2
C
N
N
l
a
y
e
r
s:
[
C
o
n
v
1
D
(
3
2
,
2
,
r
e
l
u
)
,
F
l
a
t
t
e
n
(
)
,
D
e
n
se(
1
,
si
g
m
o
i
d
)
]
,
l
o
s
s:
b
i
n
a
r
y
_
c
r
o
ss
e
n
t
r
o
p
y
,
o
p
t
i
m
i
z
e
r
:
a
d
a
m,
m
e
t
r
i
c
s
:
a
c
c
u
r
a
c
y
,
e
p
o
c
h
s:
1
0
0
,
b
a
t
c
h
_
si
z
e
:
1
6
,
v
a
l
i
d
a
t
i
o
n
_
s
p
l
i
t
:
0
.
2
B
LST
M
l
a
y
e
r
s:
[
B
i
d
i
r
e
c
t
i
o
n
a
l
(
LSTM
(
6
4
,
r
e
t
u
r
n
_
se
q
u
e
n
c
e
s=T
r
u
e
)
)
,
B
i
d
i
r
e
c
t
i
o
n
a
l
(
LST
M
(
6
4
)
)
,
D
e
n
s
e
(
1
,
si
g
m
o
i
d
)
]
,
l
o
ss
:
b
i
n
a
r
y
_
c
r
o
ss
e
n
t
r
o
p
y
,
o
p
t
i
m
i
z
e
r
:
a
d
a
m,
m
e
t
r
i
c
s
:
a
c
c
u
r
a
c
y
,
e
p
o
c
h
s:
1
0
0
,
b
a
t
c
h
_
si
z
e
:
1
6
,
v
a
l
i
d
a
t
i
o
n
_
s
p
l
i
t
:
0
.
2
LSTM
l
a
y
e
r
s:
[
LST
M
(
6
4
,
r
e
t
u
r
n
_
se
q
u
e
n
c
e
s
=
Tr
u
e
)
,
LST
M
(
6
4
)
,
D
e
n
se(1
,
si
g
m
o
i
d
)
]
,
l
o
ss
:
b
i
n
a
r
y
_
c
r
o
ss
e
n
t
r
o
p
y
,
o
p
t
i
m
i
z
e
r
:
a
d
a
m,
m
e
t
r
i
c
s:
a
c
c
u
r
a
c
y
,
e
p
o
c
h
s:
1
0
0
,
b
a
t
c
h
_
s
i
z
e
:
1
6
,
v
a
l
i
d
a
t
i
o
n
_
sp
l
i
t
:
0
.
2
G
R
U
l
a
y
e
r
s:
[
G
R
U
(
6
4
,
r
e
t
u
r
n
_
se
q
u
e
n
c
e
s
=
Tr
u
e
)
,
G
R
U
(
6
4
)
,
D
e
n
se(1
,
si
g
m
o
i
d
)
]
,
l
o
ss
:
b
i
n
a
r
y
_
c
r
o
ss
e
n
t
r
o
p
y
,
o
p
t
i
m
i
z
e
r
:
a
d
a
m,
m
e
t
r
i
c
s:
a
c
c
u
r
a
c
y
,
e
p
o
c
h
s:
1
0
0
,
b
a
t
c
h
_
s
i
z
e
:
1
6
,
v
a
l
i
d
a
t
i
o
n
_
sp
l
i
t
:
0
.
2
Tr
a
d
i
t
i
o
n
a
l
s
u
p
e
r
v
i
s
e
d
l
e
a
r
n
i
n
g
me
t
h
o
d
s
S
V
M
k
e
r
n
e
l
:
r
b
f
,
p
r
o
b
a
b
i
l
i
t
y
:
Tr
u
e
K
-
n
e
a
r
e
s
t
n
e
i
g
h
b
o
r
s
(
K
N
N
)
n
_
n
e
i
g
h
b
o
r
s:
7
Lo
g
i
s
t
i
c
r
e
g
r
e
ssi
o
n
so
l
v
e
r
:
l
b
f
g
s
En
se
mb
l
e
met
h
o
d
s
R
a
n
d
o
m
f
o
r
e
s
t
n
_
e
st
i
ma
t
o
r
s
:
1
0
0
,
ma
x
_
d
e
p
t
h
:
1
0
G
r
a
d
i
e
n
t
b
o
o
s
t
i
n
g
n
_
e
st
i
ma
t
o
r
s
:
1
0
0
,
l
e
a
r
n
i
n
g
_
r
a
t
e
:
0
.
0
5
,
ma
x
_
d
e
p
t
h
:
5
X
G
B
o
o
st
n
_
e
st
i
ma
t
o
r
s
:
1
0
0
,
l
e
a
r
n
i
n
g
_
r
a
t
e
:
0
.
0
5
,
ma
x
_
d
e
p
t
h
:
5
A
d
a
B
o
o
st
n
_
e
st
i
ma
t
o
r
s
:
1
0
0
T
h
e
lo
s
s
cu
r
v
es
in
Fig
u
r
e
3
ar
e
s
tu
d
ied
o
n
ly
to
ass
ess
co
n
v
er
g
e
n
ce
an
d
tr
ain
in
g
s
tab
ilit
y
,
n
o
t
f
o
r
d
ir
ec
t p
er
f
o
r
m
an
ce
co
m
p
ar
is
o
n
,
wh
ich
r
elies o
n
th
e
ev
alu
atio
n
m
etr
ics r
ep
o
r
ted
in
T
a
b
les
3
an
d
4
.
Fig
u
r
e
3
(
a)
s
tab
ilizes
ea
r
lie
s
t
an
d
with
th
e
f
ewe
s
t
o
s
cil
latio
n
s
,
wh
er
ea
s
t
h
e
DL
m
o
d
els
in
Fig
u
r
e
3
(b
)
t
o
3
(
f
)
r
e
q
u
ir
e
m
o
r
e
ep
o
ch
s
an
d
d
is
p
lay
co
m
p
ar
ati
v
ely
less
s
tab
le
v
alid
atio
n
lo
s
s
.
As
d
ep
icted
i
n
Fig
u
r
e
3
(
a
)
,
ANFI
S
co
n
v
er
g
es
m
u
ch
f
aster
a
n
d
m
o
r
e
s
tab
le
t
h
an
th
e
o
th
er
m
o
d
els,
r
a
p
id
ly
r
ed
u
cin
g
b
o
th
tr
ain
in
g
an
d
v
al
id
atio
n
lo
s
s
with
in
th
e
f
ir
s
t
3
0
ep
o
ch
s
with
o
n
l
y
m
in
o
r
o
s
cillatio
n
s
.
ANN
as
in
Fig
u
r
e
3
(
c)
an
d
C
NN
as
in
Fig
u
r
e
3
(
e)
,
b
y
co
n
tr
ast,
ar
e
s
lo
wer
lear
n
er
s
,
co
n
v
er
g
in
g
o
n
ly
af
ter
6
0
–
8
0
ep
o
c
h
s
.
L
STM
as
in
Fig
u
r
e
3
(
b
)
,
GR
U
as
in
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
5
,
Octo
b
er
2
0
2
6
:
4
3
0
4
-
4
3
1
4
4310
Fig
u
r
e
3
(
f
)
,
a
n
d
B
L
STM
as
in
Fig
u
r
e
3
(
d
)
co
n
v
er
g
e
f
aster
,
s
tab
ilizin
g
a
r
o
u
n
d
4
0
ep
o
c
h
s
,
b
u
t
th
eir
v
alid
atio
n
-
lo
s
s
cu
r
v
es
ar
e
n
o
ticea
b
ly
m
o
r
e
u
n
s
tab
le
th
an
ANFI
S's
,
in
d
icatin
g
h
ig
h
er
s
en
s
itiv
ity
to
tem
p
o
r
al
d
y
n
a
m
ics
an
d
lo
w
er
g
e
n
er
aliza
tio
n
a
b
ilit
y
;
B
L
STM
co
n
v
e
r
g
es
f
astes
t
a
m
o
n
g
th
em
,
y
et
its
co
n
v
er
g
ed
lo
s
s
r
em
ain
s
h
ig
h
e
r
th
an
ANFI
S's
,
co
n
f
ir
m
in
g
th
e
h
y
b
r
id
n
e
u
r
o
-
FS
g
r
ea
ter
o
p
t
im
izatio
n
s
tab
ilit
y
.
T
h
is
ca
n
b
e
attr
ib
u
ted
to
ANF
I
S's
s
m
all
n
u
m
b
er
o
f
tr
ain
a
b
le
p
ar
am
eter
s
(
th
e
m
em
b
er
s
h
ip
-
f
u
n
ctio
n
p
a
r
am
eter
s
o
f
th
r
ee
f
u
zz
if
ied
in
p
u
ts
an
d
lin
ea
r
co
n
s
eq
u
en
ts
o
f
2
7
T
S
K
r
u
les),
th
e
s
m
o
o
t
h
,
d
if
f
er
e
n
tiab
le
b
ell
-
s
h
ap
ed
m
em
b
er
s
h
ip
f
u
n
ctio
n
s
,
an
d
t
h
e
lo
w
-
d
im
en
s
io
n
al
t
h
r
ee
-
f
ea
tu
r
e
in
p
u
t
s
p
ac
e
,
all
f
av
o
r
in
g
s
tab
le
g
r
ad
ie
n
t
-
b
ased
o
p
tim
izatio
n
.
T
h
e
DL
ar
c
h
itec
tu
r
es,
d
esig
n
ed
to
ex
p
lo
it
lar
g
er
,
m
o
r
e
h
eter
o
g
e
n
eo
u
s
d
atase
ts
,
ar
e
m
o
r
e
p
r
o
n
e
to
o
v
er
f
itti
n
g
g
i
v
en
NT
HU
-
DDD'
s
lim
ited
s
u
b
ject
d
iv
er
s
ity
,
co
n
s
is
ten
t
with
th
e
v
alid
atio
n
-
lo
s
s
f
lu
ctu
atio
n
s
n
o
ted
ab
o
v
e.
(
a)
(
b
)
(
c)
(
d
)
(
e)
(f)
Fig
u
r
e
3
.
T
r
ain
in
g
a
n
d
v
alid
atio
n
lo
s
s
cu
r
v
es o
v
er
1
0
0
e
p
o
ch
s
f
o
r
(
a)
ANFI
S,
(
b
)
L
STM
,
(
c
)
ANN,
(
d
)
B
L
STM
,
(
e)
C
NN,
an
d
(
f
)
GR
U
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Hyb
r
id
mo
d
el
fo
r
d
r
iver d
r
o
w
s
in
ess
p
r
ed
ictio
n
u
s
in
g
n
eu
r
o
-
fu
z
z
y
a
n
d
r
a
n
d
o
m
fo
r
est
(
Mo
u
h
cin
e
La
kh
lo
u
fi
)
4311
3
.
2
.
Co
m
pa
riso
n o
f
ind
iv
idu
a
l a
nd
hy
brid m
o
del per
f
o
rm
a
nce
T
o
ev
al
u
ate
th
e
ef
f
ec
tiv
e
n
ess
o
f
t
h
e
p
r
o
p
o
s
ed
a
p
p
r
o
ac
h
,
a
c
o
m
p
r
e
h
en
s
iv
e
c
o
m
p
a
r
is
o
n
b
etwe
e
n
in
d
iv
id
u
al
m
o
d
els
an
d
h
y
b
r
id
ANFI
S
–
X
co
n
f
ig
u
r
atio
n
s
was
co
n
d
u
cted
.
T
h
e
o
b
jectiv
e
o
f
th
is
an
aly
s
is
is
n
o
t
o
n
ly
to
ass
ess
th
e
p
r
ed
ictiv
e
c
ap
ab
ilit
y
o
f
s
tan
d
alo
n
e
class
if
ier
s
,
b
u
t
also
to
ex
am
in
e
th
e
im
p
ac
t
o
f
in
teg
r
atin
g
ANFI
S
as
a
f
u
zz
y
in
f
er
en
c
e
co
m
p
o
n
en
t
with
in
h
y
b
r
i
d
ar
ch
itectu
r
es.
R
ath
er
th
an
ac
tin
g
s
o
lely
as
an
in
d
ep
en
d
en
t
class
if
ier
,
ANFI
S
is
in
co
r
p
o
r
ate
d
as
a
c
o
r
e
m
o
d
u
le
d
esig
n
ed
to
h
an
d
le
u
n
c
er
tain
ty
an
d
g
r
a
d
u
al
tr
an
s
itio
n
s
b
etwe
en
aler
t
an
d
d
r
o
wsy
s
tates.
T
h
r
o
u
g
h
its
f
u
zz
y
r
u
le
-
b
ased
r
ea
s
o
n
in
g
,
AN
FIS
p
r
o
v
id
es
s
tab
le
p
r
o
b
a
b
ilis
tic
esti
m
ate
s
an
d
f
ast
co
n
v
er
g
en
ce
b
eh
av
io
r
,
wh
i
ch
ar
e
p
ar
ticu
lar
ly
d
esira
b
le
in
d
r
iv
er
d
r
o
wsi
n
ess
d
etec
tio
n
.
B
y
h
y
b
r
id
izin
g
ANFI
S
with
co
m
p
lem
e
n
tar
y
class
if
ier
s
,
th
e
p
r
o
p
o
s
ed
f
r
am
ewo
r
k
aim
s
to
ex
p
l
o
it
b
o
th
th
e
in
ter
p
r
etab
ilit
y
an
d
s
tab
ilit
y
o
f
f
u
zz
y
in
f
er
e
n
ce
an
d
th
e
s
tr
o
n
g
g
en
er
aliza
ti
o
n
ca
p
ab
ilit
ies
o
f
d
ata
-
d
r
iv
e
n
m
o
d
els.
T
h
e
p
e
r
f
o
r
m
an
ce
o
f
th
e
s
tan
d
alo
n
e
m
o
d
els
is
an
aly
ze
d
f
i
r
s
t.
T
h
e
i
n
d
iv
id
u
al
m
o
d
el
p
er
f
o
r
m
an
ce
s
ar
e
r
ep
o
r
ted
in
T
ab
le
3
.
Am
o
n
g
th
e
e
v
alu
ated
a
p
p
r
o
ac
h
es,
e
n
s
em
b
le
-
b
ased
m
et
h
o
d
s
a
n
d
R
NN
s
d
em
o
n
s
tr
ate
s
tr
o
n
g
p
r
ed
ictiv
e
ca
p
ab
ilit
y
.
RF
ac
h
iev
es
th
e
h
i
g
h
est
p
er
f
o
r
m
a
n
ce
am
o
n
g
s
tan
d
alo
n
e
m
o
d
els,
with
an
ac
cu
r
ac
y
o
f
9
4
.
4
4
%,
an
F
1
-
s
co
r
e
o
f
th
e
ap
p
r
o
x
im
atel
y
9
4
.
6
%,
an
d
an
AUC
o
f
0
.
9
8
4
,
in
d
icatin
g
r
o
b
u
s
t
g
e
n
er
aliza
tio
n
a
n
d
e
f
f
e
ctiv
e
d
is
cr
im
in
atio
n
b
etwe
en
aler
t
an
d
d
r
o
wsy
s
tates.
R
ec
u
r
r
en
t
ar
c
h
itectu
r
es
s
u
ch
as
L
STM
,
B
L
STM
,
an
d
GR
U
also
y
ield
co
m
p
etitiv
e
r
esu
lts
,
with
ac
cu
r
ac
ies
ex
ce
ed
in
g
9
3
%
an
d
AUC
v
alu
es
ab
o
v
e
0
.
9
7
,
co
n
f
ir
m
i
n
g
th
eir
ab
ilit
y
to
ca
p
tu
r
e
te
m
p
o
r
al
d
ep
e
n
d
en
cies
in
b
eh
av
io
r
al
f
ea
tu
r
es.
I
n
co
n
tr
ast,
s
im
p
ler
class
if
ier
s
,
in
clu
d
in
g
lo
g
is
tic
r
eg
r
ess
io
n
,
SVM,
an
d
KNN,
ex
h
i
b
it
co
m
p
ar
ativ
el
y
lo
wer
p
er
f
o
r
m
an
ce
,
r
e
f
lectin
g
th
e
lim
itatio
n
s
o
f
lin
ea
r
o
r
d
is
tan
ce
-
b
ased
d
ec
is
io
n
b
o
u
n
d
ar
ies
f
o
r
th
is
task
.
Alth
o
u
g
h
ANFI
S
alo
n
e
d
o
es
n
o
t
ac
h
iev
e
t
h
e
h
ig
h
est
ac
cu
r
ac
y
(
9
1
.
9
0
%),
it
m
ain
tain
s
a
h
ig
h
r
ec
all
(
9
2
.
8
8
%)
an
d
a
s
tr
o
n
g
AUC
(
0
.
9
7
5
)
,
d
em
o
n
s
tr
atin
g
n
o
ta
b
le
s
en
s
itiv
ity
to
d
r
o
wsy
s
tates
an
d
s
t
ab
le
p
r
o
b
ab
ilis
tic
b
eh
av
io
r
.
T
h
ese
c
h
a
r
a
c
t
e
r
i
s
ti
cs
s
u
g
g
es
t
t
h
at
A
NF
I
S
c
o
n
s
t
it
u
t
e
s
a
r
el
i
a
b
le
a
n
d
c
o
m
p
l
e
m
e
n
ta
r
y
c
o
m
p
o
n
e
n
t
f
o
r
h
y
b
r
i
d
i
n
t
e
g
r
a
t
i
o
n
.
T
ab
le
3
.
C
o
m
p
a
r
is
o
n
o
f
v
alu
at
io
n
m
ea
s
u
r
es f
o
r
in
d
iv
i
d
u
al
m
o
d
els
M
o
d
e
l
M
S
E
R
M
S
E
M
A
E
R
e
c
a
l
l
F1
-
sc
o
r
e
AUC
A
c
c
u
r
a
c
y
LSTM
0
.
0
6
0
3
1
7
0
.
2
4
5
5
9
6
0
.
0
6
0
3
1
7
0
.
9
1
0
9
7
9
0
.
9
4
1
7
1
8
0
.
9
7
8
2
1
6
0
.
9
3
9
6
8
3
G
r
a
d
i
e
n
t
B
o
o
st
i
n
g
0
.
0
6
1
9
0
5
0
.
2
4
8
8
0
7
0
.
0
6
1
9
0
5
0
.
9
1
6
9
1
4
0
.
9
4
0
6
3
9
0
.
9
7
9
6
8
9
0
.
9
3
8
0
9
5
X
G
B
o
o
st
0
.
0
6
3
4
9
2
0
.
2
5
1
9
7
6
0
.
0
6
3
4
9
2
0
.
9
1
0
9
7
9
0
.
9
3
8
8
3
8
0
.
9
8
1
2
6
4
0
.
9
3
6
5
0
8
K
N
N
0
.
0
6
8
2
5
4
0
.
2
6
1
2
5
5
0
.
0
6
8
2
5
4
0
.
9
2
6
4
7
1
0
.
9
3
6
1
0
7
0
.
9
6
5
7
3
0
.
9
3
1
7
4
6
G
R
U
0
.
0
6
8
2
5
4
0
.
2
6
1
2
5
5
0
.
0
6
8
2
5
4
0
.
9
2
2
8
4
9
0
.
9
3
5
3
3
8
0
.
9
7
8
1
1
4
0
.
9
3
1
7
4
6
B
LST
M
0
.
0
7
3
0
1
6
0
.
2
7
0
2
1
4
0
.
0
7
3
0
1
6
0
.
9
0
5
8
8
2
0
.
9
3
0
5
1
4
0
.
9
5
8
6
0
.
9
2
6
9
8
4
ANN
0
.
0
7
6
1
9
0
.
2
7
6
0
2
6
0
.
0
7
6
1
9
0
.
8
8
4
2
7
3
0
.
9
2
5
4
6
6
0
.
9
6
8
3
5
2
0
.
9
2
3
8
1
S
V
M
0
.
0
7
7
7
7
8
0
.
2
7
8
8
8
7
0
.
0
7
7
7
7
8
0
.
8
6
9
4
3
6
0
.
9
2
2
8
3
5
0
.
9
6
3
7
7
4
0
.
9
2
2
2
2
2
A
d
a
B
o
o
st
0
.
0
7
9
3
6
5
0
.
2
8
1
7
1
8
0
.
0
7
9
3
6
5
0
.
8
9
0
2
0
8
0
.
9
2
3
0
7
7
0
.
9
7
3
0
1
0
.
9
2
0
6
3
5
C
N
N
0
.
0
8
0
9
5
2
0
.
2
8
4
5
2
1
0
.
0
8
0
9
5
2
0
.
9
0
5
0
4
5
0
.
9
2
2
8
4
4
0
.
9
7
1
4
1
0
.
9
1
9
0
4
8
R
a
n
d
o
m
f
o
r
e
s
t
0
.
0
5
5
5
5
6
0
.
2
3
5
7
0
2
0
.
0
5
5
5
5
6
0
.
9
2
5
8
1
6
0
.
9
4
6
8
8
9
0
.
9
8
4
1
6
1
0
.
9
4
4
4
4
4
A
N
F
I
S
0
.
0
8
0
9
5
2
0
.
2
8
4
5
2
1
0
.
0
8
0
9
5
2
0
.
9
2
8
7
8
3
0
.
9
2
4
6
6
8
0
.
9
7
5
5
8
3
0
.
9
1
9
0
4
8
Lo
g
i
s
t
i
c
r
e
g
r
e
ss
i
o
n
0
.
1
2
5
3
9
7
0
.
3
5
4
1
1
4
0
.
1
2
5
3
9
7
0
.
8
9
6
1
4
2
0
.
8
8
4
3
3
4
0
.
9
4
1
2
9
1
0
.
8
7
4
6
0
3
T
h
e
im
p
ac
t
o
f
h
y
b
r
id
izatio
n
is
ex
am
in
ed
n
ex
t.
T
h
e
h
y
b
r
i
d
ANFI
S
–
X
r
esu
lts
ar
e
s
u
m
m
ar
ized
in
T
ab
le
4
.
A
co
n
s
is
ten
t
im
p
r
o
v
em
en
t
ca
n
b
e
o
b
s
er
v
e
d
ac
r
o
s
s
all
h
y
b
r
id
co
n
f
ig
u
r
ati
o
n
s
wh
en
c
o
m
p
ar
e
d
t
o
th
eir
co
r
r
esp
o
n
d
in
g
s
tan
d
alo
n
e
co
u
n
ter
p
a
r
ts
,
co
n
f
ir
m
in
g
th
e
b
en
ef
it
o
f
in
teg
r
atin
g
ANFI
S
with
in
th
e
p
r
ed
ictio
n
f
r
am
ew
o
r
k
.
Am
o
n
g
all
h
y
b
r
id
m
o
d
els,
ANFI
S
–
R
F
ac
h
iev
es
th
e
b
est
o
v
er
all
p
e
r
f
o
r
m
a
n
ce
,
with
an
ac
cu
r
ac
y
o
f
9
4
.
6
0
%,
an
F1
-
s
co
r
e
o
f
9
4
.
8
2
%,
a
r
ec
all
o
f
9
2
.
2
8
%,
a
n
d
a
lo
w
MSE
o
f
0
.
0
5
4
,
o
u
tp
e
r
f
o
r
m
in
g
b
o
th
s
tan
d
alo
n
e
ANFI
S
an
d
R
F.
Hy
b
r
id
m
o
d
els
b
ased
o
n
r
ec
u
r
r
en
t
ar
ch
itectu
r
es,
s
u
ch
as
ANFI
S
–
B
L
STM
an
d
ANFI
S
–
L
STM
,
also
d
e
m
o
n
s
tr
ate
s
tr
o
n
g
p
er
f
o
r
m
an
ce
,
p
ar
ticu
lar
ly
in
ter
m
s
o
f
r
ec
all
an
d
AUC
(
b
o
th
e
x
ce
ed
in
g
0
.
9
8
)
,
h
ig
h
li
g
h
tin
g
th
e
ad
v
a
n
tag
e
o
f
co
m
b
in
in
g
f
u
zz
y
in
f
er
en
ce
with
tem
p
o
r
al
m
o
d
elin
g
.
Oth
er
h
y
b
r
id
co
n
f
ig
u
r
atio
n
s
,
in
clu
d
in
g
ANFI
S
–
GB
an
d
A
NFI
S
–
XGBo
o
s
t,
s
h
o
w
co
m
p
etitiv
e
b
u
t
s
lig
h
tly
lo
wer
p
er
f
o
r
m
a
n
ce
,
wh
ile
h
y
b
r
id
s
in
v
o
l
v
in
g
s
im
p
ler
class
if
ier
s
s
u
ch
as
ANFI
S
–
SVM
an
d
ANFI
S
–
KNN
y
ield
m
o
d
er
ate
i
m
p
r
o
v
em
en
ts
with
o
u
t
s
u
r
p
ass
in
g
t
h
e
lead
in
g
h
y
b
r
id
m
o
d
els.
Ov
er
all,
t
h
ese
r
esu
l
ts
in
d
icate
th
at
th
e
p
r
o
p
o
s
ed
ANFI
S
–
X
f
r
am
e
wo
r
k
s
y
s
tem
atica
lly
e
n
h
an
c
es
p
r
ed
ictio
n
p
e
r
f
o
r
m
an
ce
b
y
lev
e
r
ag
in
g
th
e
co
m
p
lem
en
tar
y
s
tr
en
g
th
s
o
f
f
u
zz
y
in
f
er
en
ce
an
d
d
ata
-
d
r
iv
e
n
class
if
ier
s
.
Am
o
n
g
all
h
y
b
r
i
d
co
n
f
ig
u
r
atio
n
s
,
th
e
ANFI
S
–
R
F m
o
d
el
em
er
g
es a
s
th
e
m
o
s
t b
alan
ce
d
a
n
d
r
eliab
le
s
o
lu
tio
n
f
o
r
d
r
iv
er
d
r
o
wsi
n
ess
d
etec
tio
n
.
As
s
u
m
m
ar
ized
in
T
a
b
le
5
,
t
h
e
p
r
o
p
o
s
ed
ANFI
S
–
R
F
h
y
b
r
id
m
o
d
el
d
em
o
n
s
tr
ates
s
u
p
er
io
r
b
alan
ce
in
p
r
ed
ictiv
e
p
er
f
o
r
m
a
n
ce
co
m
p
ar
ed
to
r
ec
en
t
m
eth
o
d
s
ap
p
l
ied
to
th
e
NT
HU
-
DDD
d
ata
s
et.
W
h
ile
ce
r
tain
ap
p
r
o
ac
h
es,
s
u
ch
as
C
NN
–
L
STM
o
f
C
h
en
et
a
l.
[
2
7
]
(
9
3
.
3
0
%
ac
cu
r
ac
y
)
an
d
R
NN
o
f
E
d
-
Do
u
g
h
m
i
et
a
l.
[
1
1
]
(
9
2
.
0
0
%
ac
cu
r
ac
y
)
,
ac
h
ie
v
ed
s
lig
h
tly
h
ig
h
er
ac
c
u
r
ac
y
,
th
ey
eith
er
lack
F1
-
s
co
r
e
r
ep
o
r
tin
g
o
r
p
r
esen
t
lo
wer
F1
-
s
co
r
es,
in
d
icatin
g
p
o
ten
tia
l
p
r
ec
is
io
n
–
r
ec
all
im
b
alan
ce
.
I
n
co
n
tr
ast,
o
u
r
m
o
d
el
attain
e
d
a
r
o
b
u
s
t
F1
-
s
co
r
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
5
,
Octo
b
er
2
0
2
6
:
4
3
0
4
-
4
3
1
4
4312
o
f
9
0
.
6
%,
o
u
tp
e
r
f
o
r
m
in
g
th
e
d
ee
p
C
NN
b
y
Du
a
et
a
l.
[
2
8
]
(
8
4
.
0
9
%)
an
d
th
e
3
D
C
NN
b
y
Z
h
ao
et
a
l
.
[
1
2
]
.
Fu
r
th
er
m
o
r
e
,
th
e
co
m
b
in
atio
n
o
f
ad
a
p
tiv
e
f
u
zz
y
r
ea
s
o
n
in
g
an
d
en
s
em
b
le
lear
n
in
g
in
ANFI
S
–
R
F
en
ab
les
ef
f
ec
tiv
e
g
en
er
aliza
tio
n
u
n
d
e
r
f
ac
ial
v
ar
ia
b
ilit
y
.
T
h
ese
r
esu
lts
u
n
d
er
lin
e
th
e
r
elev
an
ce
o
f
n
e
u
r
o
-
f
u
zz
y
en
s
em
b
le
m
o
d
els
as
in
ter
p
r
e
tab
le
an
d
co
m
p
etitiv
e
alter
n
a
tiv
es
to
d
ee
p
ar
c
h
itectu
r
es
f
o
r
r
ea
l
-
tim
e
d
r
iv
er
d
r
o
wsi
n
ess
d
etec
tio
n
.
T
h
is
s
h
o
u
ld
b
e
in
ter
p
r
eted
in
lig
h
t
o
f
th
e
d
ataset'
s
s
co
p
e:
o
n
ly
s
ix
s
u
b
jects
ar
e
r
ep
r
esen
ted
,
lim
itin
g
f
ac
ial
an
d
d
em
o
g
r
a
p
h
ic
d
iv
er
s
ity
,
an
d
d
r
o
wsy
-
r
elate
d
b
e
h
av
io
r
s
ar
e
lik
ely
u
n
d
er
r
ep
r
esen
ted
r
elativ
e
to
a
ler
t
s
tates,
in
tr
o
d
u
cin
g
s
o
m
e
c
lass
im
b
alan
ce
.
B
o
th
f
ac
to
r
s
m
o
tiv
ate
o
u
r
u
s
e
o
f
AUC alo
n
g
s
id
e
ac
cu
r
ac
y
an
d
F1
-
s
co
r
e.
T
ab
le
4
.
C
o
m
p
a
r
is
o
n
b
etwe
en
in
d
iv
id
u
al
a
n
d
h
y
b
r
id
m
o
d
els
M
o
d
e
l
M
S
E
R
M
S
E
M
A
E
A
c
c
u
r
a
c
y
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
-
sc
o
r
e
AUC
A
N
F
I
S
-
RF
0
.
0
5
3
9
6
8
0
.
2
3
2
3
1
1
0
.
0
5
3
9
6
8
0
.
9
4
6
0
3
2
0
.
9
7
4
9
2
2
0
.
9
2
2
8
4
9
0
.
9
4
8
1
7
1
0
.
9
7
8
6
5
1
A
N
F
I
S
-
B
LSTM
0
.
0
5
7
1
4
3
0
.
2
3
9
0
4
6
0
.
0
5
7
1
4
3
0
.
9
4
2
8
5
7
0
.
9
7
1
7
8
7
0
.
9
1
9
8
8
1
0
.
9
4
5
1
2
2
0
.
9
8
2
9
9
6
A
N
F
I
S
-
LSTM
0
.
0
5
8
7
3
0
.
2
4
2
3
4
3
0
.
0
5
8
7
3
0
.
9
4
1
2
7
0
.
9
7
7
7
0
7
0
.
9
1
0
9
7
9
0
.
9
4
3
1
6
4
0
.
9
7
8
3
1
2
A
N
F
I
S
-
G
R
U
0
.
0
6
1
9
0
5
0
.
2
4
8
8
0
7
0
.
0
6
1
9
0
5
0
.
9
3
8
0
9
5
0
.
9
6
5
6
2
5
0
.
9
1
6
9
1
4
0
.
9
4
0
6
3
9
0
.
9
7
7
2
1
8
A
N
F
I
S
-
G
B
o
o
st
i
n
g
0
.
0
6
3
4
9
2
0
.
2
5
1
9
7
6
0
.
0
6
3
4
9
2
0
.
9
3
6
5
0
8
0
.
9
7
1
4
2
9
0
.
9
0
8
0
1
2
0
.
9
3
8
6
5
0
.
9
7
9
6
3
9
A
N
F
I
S
-
X
G
B
o
o
st
0
.
0
6
6
6
6
7
0
.
2
5
8
1
9
9
0
.
0
6
6
6
6
7
0
.
9
3
3
3
3
3
0
.
9
7
1
2
4
6
0
.
9
0
2
0
7
7
0
.
9
3
5
3
8
5
0
.
9
8
1
3
9
6
A
N
F
I
S
-
A
d
a
B
o
o
st
0
.
0
6
8
2
5
4
0
.
2
6
1
2
5
5
0
.
0
6
8
2
5
4
0
.
9
3
1
7
4
6
0
.
9
7
7
2
7
3
0
.
8
9
3
1
7
5
0
.
9
3
3
3
3
3
0
.
9
6
2
1
1
8
A
N
F
I
S
-
K
N
N
0
.
0
7
1
4
2
9
0
.
2
6
7
2
6
1
0
.
0
7
1
4
2
9
0
.
9
2
8
5
7
1
0
.
9
4
8
3
2
8
0
.
9
1
7
6
4
7
0
.
9
3
2
7
3
5
0
.
9
6
5
7
A
N
F
I
S
-
ANN
0
.
0
7
4
6
0
3
0
.
2
7
3
1
3
6
0
.
0
7
4
6
0
3
0
.
9
2
5
3
9
7
0
.
9
7
0
7
7
9
0
.
8
8
7
2
4
0
.
9
2
7
1
3
2
0
.
9
6
7
9
7
7
A
N
F
I
S
-
S
V
M
0
.
0
7
4
6
0
3
0
.
2
7
3
1
3
6
0
.
0
7
4
6
0
3
0
.
9
2
5
3
9
7
0
.
9
7
0
7
7
9
0
.
8
8
7
2
4
0
.
9
2
7
1
3
2
0
.
9
7
0
4
8
8
A
N
F
I
S
-
C
N
N
0
.
0
7
7
7
7
8
0
.
2
7
8
8
8
7
0
.
0
7
7
7
7
8
0
.
9
2
2
2
2
2
0
.
9
5
0
.
9
0
2
0
7
7
0
.
9
2
5
4
1
9
0
.
9
7
0
9
2
9
A
N
F
I
S
-
LR
0
.
0
8
2
5
4
0
.
2
8
7
2
9
7
0
.
0
8
2
5
4
0
.
9
1
7
4
6
0
.
9
4
6
7
0
8
0
.
8
9
6
1
4
2
0
.
9
2
0
7
3
2
0
.
9
6
5
8
9
6
T
ab
le
5
.
C
o
m
p
a
r
ativ
e
s
tu
d
ies o
n
d
etec
tin
g
d
r
iv
e
r
d
r
o
wsi
n
ess
u
s
in
g
th
e
NT
HU
-
DDD
d
ataset
S
t
u
d
y
M
e
t
h
o
d
U
sed
v
a
r
i
a
b
l
e
s
D
a
t
a
s
e
t
A
c
c
u
r
a
c
y
(
%)
F1
-
sc
o
r
e
(
%)
[
2
8
]
D
e
e
p
C
N
N
H
a
n
d
g
e
s
t
u
r
e
s
,
f
a
c
i
a
l
e
x
p
r
e
ss
i
o
n
s
,
h
e
a
d
m
o
v
e
me
n
t
s
N
TH
U
-
DDD
8
5
.
0
0
8
4
.
0
9
[
1
1
]
R
N
N
D
r
i
v
e
r
i
ma
g
e
N
TH
U
-
DDD
9
2
.
0
0
8
5
.
0
0
[
1
2
]
3
D
C
N
N
D
r
i
v
e
r
f
a
c
e
a
n
d
f
a
c
i
a
l
l
a
n
d
m
a
r
k
s
N
TH
U
-
DDD
8
8
.
6
0
[
2
9
]
D
e
e
p
C
N
N
e
n
s
e
mb
l
e
H
a
n
d
g
e
s
t
u
r
e
s
,
f
a
c
i
a
l
e
x
p
r
e
ss
i
o
n
s
,
h
e
a
d
m
o
v
e
me
n
t
s
N
TH
U
-
DDD
8
5
.
0
0
8
4
.
0
9
[
1
1
]
R
N
N
D
r
i
v
e
r
f
a
c
e
a
n
d
f
a
c
i
a
l
l
a
n
d
m
a
r
k
s
N
TH
U
-
DDD
9
2
.
0
0
8
5
.
0
0
[
1
2
]
3
D
C
N
N
D
r
i
v
e
r
f
a
c
e
a
n
d
f
a
c
i
a
l
l
a
n
d
m
a
r
k
s
N
TH
U
-
DDD
8
8
.
6
0
[
2
7
]
C
N
N
+
LSTM
Ey
e
a
n
d
f
a
c
e
r
e
g
i
o
n
N
TH
U
-
DDD
9
3
.
3
0
[
2
9
]
M
LP
+
K
N
N
Ey
e
c
l
o
s
u
r
e
,
y
a
w
n
i
n
g
N
TH
U
-
DDD
9
4
.
3
1
7
9
.
0
0
Th
i
s
st
u
d
y
A
N
F
I
S
-
RF
F
a
c
i
a
l
l
a
n
d
mar
k
(
e
y
e
a
n
d
mo
u
t
h
)
N
TH
U
-
DDD
9
4
.
6
0
9
4
.
8
1
4.
CO
NCLU
SI
O
N
T
h
is
r
esear
ch
p
r
o
p
o
s
ed
a
h
y
b
r
id
m
o
d
el
o
f
ANFI
S
an
d
R
F
to
im
p
r
o
v
e
th
e
p
r
e
cisi
o
n
an
d
tr
u
s
two
r
th
in
ess
o
f
d
r
iv
er
d
r
o
wsi
n
ess
d
etec
tio
n
.
E
f
f
icien
t
ANFI
S,
h
an
d
lin
g
u
n
ce
r
tai
n
ty
,
c
o
u
p
led
with
R
F
'
s
en
s
em
b
le
lear
n
in
g
ca
p
a
b
ilit
y
,
en
ab
led
th
e
h
y
b
r
id
m
o
d
el
to
o
u
tp
e
r
f
o
r
m
th
e
s
tan
d
alo
n
e
an
d
s
ev
er
al
s
tate
-
of
-
th
e
-
ar
t
m
o
d
els
o
n
th
e
NT
HU
-
DDD
d
ataset.
Sp
ec
if
ically
,
th
e
m
o
d
el
r
ea
c
h
ed
a
n
a
cc
u
r
ac
y
o
f
9
4
.
6
0
%
an
d
an
F1
-
s
co
r
e
o
f
9
4
.
8
1
%,
wh
ich
in
d
icate
s
its
ca
p
ab
ilit
y
in
h
an
d
lin
g
a
v
ar
iety
o
f
f
ac
ial
ex
p
r
ess
io
n
s
.
Fu
r
th
er
m
o
r
e
,
th
e
m
o
d
el,
b
ei
n
g
b
o
th
ef
f
icien
t
an
d
in
ter
p
r
etab
le,
ca
n
b
e
a
g
o
o
d
ca
n
d
id
ate
f
o
r
r
ea
l
-
tim
e
ap
p
licatio
n
o
n
em
b
e
d
d
ed
s
y
s
tem
s
.
E
x
p
lo
r
atio
n
s
f
o
r
f
u
tu
r
e
r
esear
ch
will
in
clu
d
e
co
m
b
in
in
g
m
o
r
e
b
eh
a
v
io
r
al
an
d
p
h
y
s
io
lo
g
ical
s
ig
n
als
(
e.
g
.
,
h
ea
r
t
r
ate
,
s
teer
in
g
p
atter
n
s
)
an
d
d
iv
er
s
if
y
in
g
th
e
m
o
d
el'
s
d
r
iv
in
g
e
n
v
ir
o
n
m
en
t
to
im
p
r
o
v
e
g
en
e
r
aliza
tio
n
,
r
o
b
u
s
tn
ess
,
an
d
r
ea
l
-
wo
r
ld
a
p
p
lic
ab
ilit
y
.
ACK
NO
WL
E
DG
M
E
N
T
S
T
h
e
au
t
h
o
r
s
t
h
an
k
th
e
f
ac
u
lty
an
d
ad
m
i
n
is
tr
ativ
e
s
taf
f
at
th
e
Natio
n
al
Sch
o
o
l
o
f
Ap
p
lied
Scien
ce
s
,
Sid
i M
o
h
am
ed
B
en
Ab
d
ella
h
Un
iv
er
s
ity
,
f
o
r
t
h
eir
s
u
p
p
o
r
t.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
T
h
is
r
esear
ch
was n
o
t f
u
n
d
ed
b
y
an
y
o
r
g
an
izatio
n
.
AUT
H
O
R
CO
NT
RI
B
UT
I
O
NS ST
A
T
E
M
E
N
T
T
h
is
jo
u
r
n
al
u
s
es
th
e
C
o
n
tr
ib
u
to
r
R
o
les
T
ax
o
n
o
m
y
(
C
R
ed
iT)
to
r
ec
o
g
n
ize
in
d
iv
id
u
al
au
th
o
r
co
n
tr
ib
u
tio
n
s
,
r
ed
u
ce
au
th
o
r
s
h
ip
d
is
p
u
tes,
an
d
f
ac
ilit
ate
co
llab
o
r
atio
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Hyb
r
id
mo
d
el
fo
r
d
r
iver d
r
o
w
s
in
ess
p
r
ed
ictio
n
u
s
in
g
n
eu
r
o
-
fu
z
z
y
a
n
d
r
a
n
d
o
m
fo
r
est
(
Mo
u
h
cin
e
La
kh
lo
u
fi
)
4313
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
Mo
u
h
cin
e
L
a
k
h
lo
u
f
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
E
l M
eh
d
i M
ello
u
li
✓
✓
✓
✓
✓
✓
✓
C
:
C
o
n
c
e
p
t
u
a
l
i
z
a
t
i
o
n
M
:
M
e
t
h
o
d
o
l
o
g
y
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
d
a
t
i
o
n
Fo
:
Fo
r
mal
a
n
a
l
y
s
i
s
I
:
I
n
v
e
s
t
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
su
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
s
i
o
n
P
:
P
r
o
j
e
c
t
a
d
mi
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
si
t
i
o
n
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
Au
th
o
r
s
s
tate
n
o
co
n
f
lict o
f
in
t
er
est.
I
NF
O
RM
E
D
CO
NS
E
N
T
W
e
h
av
e
o
b
tain
ed
in
f
o
r
m
ed
c
o
n
s
en
t f
r
o
m
all
in
d
iv
id
u
als in
c
lu
d
ed
in
t
h
is
s
tu
d
y
.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
d
ataset
u
s
ed
in
th
is
s
tu
d
y
was
p
r
o
v
id
ed
b
y
th
e
C
o
m
p
u
t
er
Vis
io
n
L
ab
o
r
ato
r
y
at
Natio
n
al
T
s
in
g
Hu
a
Un
iv
er
s
ity
an
d
c
an
b
e
ac
c
ess
ed
o
n
lin
e
at
h
ttp
://cv
.
cs.n
th
u
.
ed
u
.
tw/p
h
p
/callf
o
r
p
ap
er
/d
ata
s
ets/
D
DD/.
RE
F
E
R
E
NC
E
S
[
1
]
S
.
O
t
ma
n
i
,
T.
P
e
b
a
y
l
e
,
J.
R
o
g
e
,
a
n
d
A
.
M
u
z
e
t
,
“
Ef
f
e
c
t
o
f
d
r
i
v
i
n
g
d
u
r
a
t
i
o
n
a
n
d
p
a
r
t
i
a
l
sl
e
e
p
d
e
p
r
i
v
a
t
i
o
n
o
n
s
u
b
s
e
q
u
e
n
t
a
l
e
r
t
n
e
ss
a
n
d
p
e
r
f
o
r
ma
n
c
e
o
f
c
a
r
d
r
i
v
e
r
s
,
”
P
h
y
si
o
l
o
g
y
&
Be
h
a
v
i
o
r
,
v
o
l
.
8
4
,
n
o
.
5
,
p
p
.
7
1
5
–
7
2
4
,
A
p
r
.
2
0
0
5
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
p
h
y
s
b
e
h
.
2
0
0
5
.
0
2
.
0
2
1
.
[
2
]
M
.
A
w
a
i
s,
N
.
B
a
d
r
u
d
d
i
n
,
a
n
d
M
.
D
r
i
e
b
e
r
g
,
“
A
h
y
b
r
i
d
a
p
p
r
o
a
c
h
t
o
d
e
t
e
c
t
d
r
i
v
e
r
d
r
o
w
si
n
e
ss
u
t
i
l
i
z
i
n
g
p
h
y
si
o
l
o
g
i
c
a
l
s
i
g
n
a
l
s
t
o
i
mp
r
o
v
e
s
y
st
e
m
p
e
r
f
o
r
ma
n
c
e
a
n
d
w
e
a
r
a
b
i
l
i
t
y
,
”
S
e
n
s
o
rs
,
v
o
l
.
1
7
,
n
o
.
9
,
A
u
g
.
2
0
1
7
,
d
o
i
:
1
0
.
3
3
9
0
/
s
1
7
0
9
1
9
9
1
.
[
3
]
A
.
G
.
C
o
r
r
e
a
,
L.
O
r
o
s
c
o
,
a
n
d
E
.
La
c
i
a
r
,
“
A
u
t
o
ma
t
i
c
d
e
t
e
c
t
i
o
n
o
f
d
r
o
w
s
i
n
e
ss
i
n
EEG
r
e
c
o
r
d
s
b
a
s
e
d
o
n
m
u
l
t
i
mo
d
a
l
a
n
a
l
y
s
i
s,
”
Me
d
i
c
a
l
E
n
g
i
n
e
e
ri
n
g
& P
h
y
si
c
s
,
v
o
l
.
3
6
,
n
o
.
2
,
p
p
.
2
4
4
–
2
4
9
,
F
e
b
.
2
0
1
4
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
me
d
e
n
g
p
h
y
.
2
0
1
3
.
0
7
.
0
1
1
.
[
4
]
H
.
Z
h
a
n
g
,
C
.
W
u
,
X
.
Y
a
n
,
a
n
d
T.
Z.
Q
i
u
,
“
Th
e
e
f
f
e
c
t
o
f
f
a
t
i
g
u
e
d
r
i
v
i
n
g
o
n
c
a
r
f
o
l
l
o
w
i
n
g
b
e
h
a
v
i
o
r
,
”
T
r
a
n
s
p
o
rt
a
t
i
o
n
R
e
se
a
rc
h
P
a
rt
F:
T
ra
f
f
i
c
P
syc
h
o
l
o
g
y
a
n
d
Be
h
a
v
i
o
u
r
,
v
o
l
.
4
3
,
p
p
.
8
0
–
8
9
,
N
o
v
.
2
0
1
6
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
t
r
f
.
2
0
1
6
.
0
6
.
0
1
7
.
[
5
]
X
.
Z
h
a
n
g
,
X
.
W
a
n
g
,
X
.
Y
a
n
g
,
C
.
X
u
,
X
.
Zh
u
,
a
n
d
J
.
W
e
i
,
“
D
r
i
v
e
r
d
r
o
w
si
n
e
ss
d
e
t
e
c
t
i
o
n
u
si
n
g
m
i
x
e
d
-
e
f
f
e
c
t
o
r
d
e
r
e
d
l
o
g
i
t
mo
d
e
l
c
o
n
s
i
d
e
r
i
n
g
t
i
me
c
u
m
u
l
a
t
i
v
e
e
f
f
e
c
t
,
”
A
n
a
l
y
t
i
c
Me
t
h
o
d
s
i
n
Ac
c
i
d
e
n
t
R
e
se
a
rc
h
,
v
o
l
.
2
6
,
Ju
n
.
2
0
2
0
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
a
mar.
2
0
2
0
.
1
0
0
1
1
4
.
[
6
]
A
.
Q
u
d
d
u
s
,
A
.
S
.
Za
n
d
i
,
L
.
P
r
e
s
t
,
a
n
d
F
.
J.
E.
C
o
m
e
a
u
,
“
U
si
n
g
l
o
n
g
s
h
o
r
t
-
t
e
r
m
mem
o
r
y
a
n
d
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
s
f
o
r
d
r
i
v
e
r
d
r
o
w
s
i
n
e
ss
d
e
t
e
c
t
i
o
n
,
”
A
c
c
i
d
e
n
t
A
n
a
l
y
s
i
s &
Pre
v
e
n
t
i
o
n
,
v
o
l
.
1
5
6
,
J
u
n
.
2
0
2
1
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
a
a
p
.
2
0
2
1
.
1
0
6
1
0
7
.
[
7
]
S
.
B
a
k
h
e
e
t
a
n
d
A
.
A
l
-
H
a
m
a
d
i
,
“
A
f
r
a
mew
o
r
k
f
o
r
i
n
st
a
n
t
a
n
e
o
u
s d
r
i
v
e
r
d
r
o
w
si
n
e
ss d
e
t
e
c
t
i
o
n
b
a
s
e
d
o
n
i
m
p
r
o
v
e
d
H
O
G
f
e
a
t
u
r
e
s a
n
d
N
a
ï
v
e
B
a
y
e
s
i
a
n
c
l
a
ss
i
f
i
c
a
t
i
o
n
,
”
Bra
i
n
S
c
i
e
n
c
e
s
,
v
o
l
.
1
1
,
n
o
.
2
,
F
e
b
.
2
0
2
1
,
d
o
i
:
1
0
.
3
3
9
0
/
b
r
a
i
n
s
c
i
1
1
0
2
0
2
4
0
.
[
8
]
B
.
C
h
e
n
g
,
W
.
Z
h
a
n
g
,
Y
.
L
i
n
,
R
.
F
e
n
g
,
a
n
d
X
.
Z
h
a
n
g
,
“
D
r
i
v
e
r
d
r
o
w
si
n
e
s
s
d
e
t
e
c
t
i
o
n
b
a
se
d
o
n
m
u
l
t
i
s
o
u
r
c
e
i
n
f
o
r
mat
i
o
n
,
”
H
u
m
a
n
Fa
c
t
o
rs
a
n
d
Erg
o
n
o
m
i
c
s
i
n
M
a
n
u
f
a
c
t
u
ri
n
g
&
S
e
r
v
i
c
e
I
n
d
u
s
t
ri
e
s
,
v
o
l
.
2
2
,
n
o
.
5
,
p
p
.
4
5
0
–
4
6
7
,
S
e
p
.
2
0
1
2
,
d
o
i
:
1
0
.
1
0
0
2
/
h
f
m.
2
0
3
9
5
.
[
9
]
S
.
E.
B
e
k
h
o
u
c
h
e
,
Y
.
R
u
i
c
h
e
k
,
a
n
d
F
.
D
o
r
n
a
i
k
a
,
“
D
r
i
v
e
r
d
r
o
w
s
i
n
e
ss
d
e
t
e
c
t
i
o
n
i
n
v
i
d
e
o
s
e
q
u
e
n
c
e
s
u
s
i
n
g
h
y
b
r
i
d
sel
e
c
t
i
o
n
o
f
d
e
e
p
f
e
a
t
u
r
e
s,
”
K
n
o
w
l
e
d
g
e
-
B
a
se
d
S
y
s
t
e
m
s
,
v
o
l
.
2
5
2
,
S
e
p
.
2
0
2
2
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
k
n
o
sy
s
.
2
0
2
2
.
1
0
9
4
3
6
.
[
1
0
]
A
.
A
.
B
a
mi
d
e
l
e
e
t
a
l
.
,
“
N
o
n
-
i
n
t
r
u
s
i
v
e
d
r
i
v
e
r
d
r
o
w
s
i
n
e
ss
d
e
t
e
c
t
i
o
n
b
a
s
e
d
o
n
f
a
c
e
a
n
d
e
y
e
t
r
a
c
k
i
n
g
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
Ad
v
a
n
c
e
d
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
A
p
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
1
0
,
n
o
.
7
,
p
p
.
5
4
9
–
5
6
9
,
2
0
1
9
,
d
o
i
:
1
0
.
1
4
5
6
9
/
I
JA
C
S
A
.
2
0
1
9
.
0
1
0
0
7
7
5
.
[
1
1
]
Y
.
E
d
-
D
o
u
g
h
m
i
,
N
.
I
d
r
i
ss
i
,
a
n
d
Y
.
H
b
a
l
i
,
“
R
e
a
l
-
t
i
me
s
y
st
e
m
f
o
r
d
r
i
v
e
r
f
a
t
i
g
u
e
d
e
t
e
c
t
i
o
n
b
a
se
d
o
n
a
r
e
c
u
r
r
e
n
t
n
e
u
r
o
n
a
l
n
e
t
w
o
r
k
,
”
J
o
u
rn
a
l
o
f
I
m
a
g
i
n
g
,
v
o
l
.
6
,
n
o
.
3
,
M
a
r
.
2
0
2
0
,
d
o
i
:
1
0
.
3
3
9
0
/
j
i
m
a
g
i
n
g
6
0
3
0
0
0
8
.
[
1
2
]
L.
Zh
a
o
,
Z.
W
a
n
g
,
G
.
Zh
a
n
g
,
a
n
d
H
.
G
a
o
,
“
D
r
i
v
e
r
d
r
o
w
s
i
n
e
ss
r
e
c
o
g
n
i
t
i
o
n
v
i
a
t
r
a
n
sf
e
r
r
e
d
d
e
e
p
3
D
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
t
w
o
r
k
a
n
d
st
a
t
e
p
r
o
b
a
b
i
l
i
t
y
v
e
c
t
o
r
,
”
Mu
l
t
i
m
e
d
i
a
T
o
o
l
s
a
n
d
Ap
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
7
9
,
n
o
.
3
5
–
3
6
,
p
p
.
2
6
6
8
3
–
2
6
7
0
1
,
S
e
p
.
2
0
2
0
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
1
0
4
2
-
020
-
0
9
2
5
9
-
w.
[
1
3
]
A
.
Y
i
l
d
i
z
,
M
.
A
k
i
n
,
M
.
P
o
y
r
a
z
,
a
n
d
G
.
K
i
r
b
a
s,
“
A
p
p
l
i
c
a
t
i
o
n
o
f
a
d
a
p
t
i
v
e
n
e
u
r
o
-
f
u
z
z
y
i
n
f
e
r
e
n
c
e
s
y
st
e
m
f
o
r
v
i
g
i
l
a
n
c
e
l
e
v
e
l
e
st
i
mat
i
o
n
b
y
u
s
i
n
g
w
a
v
e
l
e
t
-
e
n
t
r
o
p
y
f
e
a
t
u
r
e
e
x
t
r
a
c
t
i
o
n
,
”
Ex
p
e
r
t
S
y
s
t
e
m
s
w
i
t
h
A
p
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
3
6
,
n
o
.
4
,
p
p
.
7
3
9
0
–
7
3
9
9
,
M
a
y
2
0
0
9
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
e
sw
a
.
2
0
0
8
.
0
9
.
0
0
3
.
[
1
4
]
S
.
A
r
e
f
n
e
z
h
a
d
,
S
.
S
a
mi
e
e
,
A
.
Ei
c
h
b
e
r
g
e
r
,
a
n
d
A
.
N
a
h
v
i
,
“
D
r
i
v
e
r
d
r
o
w
si
n
e
ss
d
e
t
e
c
t
i
o
n
b
a
se
d
o
n
st
e
e
r
i
n
g
w
h
e
e
l
d
a
t
a
a
p
p
l
y
i
n
g
a
d
a
p
t
i
v
e
n
e
u
r
o
-
f
u
z
z
y
f
e
a
t
u
r
e
s
e
l
e
c
t
i
o
n
,
”
S
e
n
s
o
rs
,
v
o
l
.
1
9
,
n
o
.
4
,
F
e
b
.
2
0
1
9
,
d
o
i
:
1
0
.
3
3
9
0
/
s1
9
0
4
0
9
4
3
.
[
1
5
]
I
.
A
.
F
o
u
a
d
,
“
A
r
o
b
u
s
t
a
n
d
e
f
f
i
c
i
e
n
t
EEG
-
b
a
s
e
d
d
r
o
w
si
n
e
ss
d
e
t
e
c
t
i
o
n
sy
s
t
e
m
u
si
n
g
d
i
f
f
e
r
e
n
t
mac
h
i
n
e
l
e
a
r
n
i
n
g
a
l
g
o
r
i
t
h
ms,
”
A
i
n
S
h
a
m
s
En
g
i
n
e
e
ri
n
g
J
o
u
rn
a
l
,
v
o
l
.
1
4
,
n
o
.
3
,
A
p
r
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
a
s
e
j
.
2
0
2
2
.
1
0
1
8
9
5
.
[
1
6
]
A
.
D
.
M
c
D
o
n
a
l
d
,
J.
D
.
L
e
e
,
C
.
S
c
h
w
a
r
z
,
a
n
d
T
.
L.
B
r
o
w
n
,
“
S
t
e
e
r
i
n
g
i
n
a
r
a
n
d
o
m
f
o
r
e
s
t
,
”
H
u
m
a
n
Fa
c
t
o
rs:
T
h
e
J
o
u
rn
a
l
o
f
t
h
e
H
u
m
a
n
F
a
c
t
o
rs
a
n
d
Erg
o
n
o
m
i
c
s
S
o
c
i
e
t
y
,
v
o
l
.
5
6
,
n
o
.
5
,
p
p
.
9
8
6
–
9
9
8
,
A
u
g
.
2
0
1
4
,
d
o
i
:
1
0
.
1
1
7
7
/
0
0
1
8
7
2
0
8
1
3
5
1
5
2
7
2
.
[
1
7
]
C.
-
H
.
W
e
n
g
,
Y
.
-
H
.
L
a
i
,
a
n
d
S
.
-
H
.
La
i
,
“
D
r
i
v
e
r
d
r
o
w
si
n
e
ss
d
e
t
e
c
t
i
o
n
v
i
a
a
h
i
e
r
a
r
c
h
i
c
a
l
t
e
m
p
o
r
a
l
d
e
e
p
b
e
l
i
e
f
n
e
t
w
o
r
k
,
”
i
n
C
o
m
p
u
t
e
r
Vi
si
o
n
–
A
C
C
V
2
0
1
6
W
o
rks
h
o
p
s
(
AC
C
V
2
0
1
6
)
,
C
h
a
m
,
S
w
i
t
z
e
r
l
a
n
d
:
S
p
r
i
n
g
e
r
,
2
0
1
7
,
p
p
.
1
1
7
–
133
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
319
-
5
4
5
2
6
-
4
_
9
.
[
1
8
]
T.
S
o
u
k
u
p
o
v
á
a
n
d
J.
C
e
c
h
,
“
R
e
a
l
-
t
i
me
e
y
e
-
b
l
i
n
k
d
e
t
e
c
t
i
o
n
u
s
i
n
g
f
a
c
i
a
l
l
a
n
d
mar
k
s,
”
i
n
2
1
st
C
o
m
p
u
t
e
r
Vi
s
i
o
n
Wi
n
t
e
r
W
o
r
k
sh
o
p
,
2
0
1
6
,
p
p
.
1
–
8.
[
1
9
]
Q
.
Z
h
u
a
n
g
,
Z.
K
e
h
u
a
,
J.
W
a
n
g
,
a
n
d
Q
.
C
h
e
n
,
“
D
r
i
v
e
r
f
a
t
i
g
u
e
d
e
t
e
c
t
i
o
n
m
e
t
h
o
d
b
a
se
d
o
n
e
y
e
s
t
a
t
e
s
w
i
t
h
p
u
p
i
l
a
n
d
i
r
i
s
seg
m
e
n
t
a
t
i
o
n
,
”
I
EEE
Ac
c
e
ss
,
v
o
l
.
8
,
p
p
.
1
7
3
4
4
0
–
1
7
3
4
4
9
,
2
0
2
0
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
0
.
3
0
2
5
8
1
8
.
Evaluation Warning : The document was created with Spire.PDF for Python.