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.
13
,
No
.
4
,
Dece
m
b
er
2
0
2
4
,
p
p
.
4
8
8
3
~
4
8
9
4
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
13
.i
4
.
p
p
4
8
8
3
-
4
8
9
4
4883
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
A det
ection mo
de
l of ag
g
ress
iv
e dri
v
ing
behav
io
r bas
ed on
hy
brid deep
learn
ing
No
o
r
Wa
lid
K
ha
lid
,
Wis
a
m
Da
wo
o
d Abdu
lla
h
C
o
l
l
e
g
e
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
M
a
t
h
e
ma
t
i
c
s
,
T
i
k
r
i
t
U
n
i
v
e
r
si
t
y
,
T
i
k
r
i
t
,
I
r
a
q
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
an
1
7
,
2
0
2
4
R
ev
is
ed
Ma
r
4
,
2
0
2
4
Acc
ep
ted
Ma
r
2
1
,
2
0
2
4
A
m
a
jo
r
p
ro
b
lem
in
to
d
a
y
’s
tran
s
p
o
rtati
o
n
sy
ste
m
s
is
d
riv
i
n
g
b
e
h
a
v
io
r,
si
n
c
e
th
e
re
a
re
g
ro
win
g
wo
rries
c
o
n
c
e
rn
in
g
e
n
s
u
rin
g
t
h
e
sa
fe
ty
o
f
m
o
to
rists,
p
a
ss
e
n
g
e
rs,
a
n
d
o
th
e
r
ro
a
d
u
se
r
s.
De
e
p
lea
rn
i
n
g
a
lg
o
rit
h
m
s
c
a
n
c
las
sify
p
e
o
p
le
b
a
se
d
o
n
t
h
e
ir
d
ri
v
i
n
g
b
e
h
a
v
io
rs
a
n
d
i
d
e
n
ti
f
y
d
riv
i
n
g
tr
e
n
d
s
fro
m
se
n
so
r
d
a
ta.
Th
is
p
a
p
e
r
p
re
se
n
ts
a
n
o
v
e
l
m
o
d
e
l
b
a
se
d
o
n
a
d
ri
v
in
g
b
e
h
a
v
i
o
r
d
a
tas
e
t
g
a
th
e
re
d
fro
m
c
e
ll
p
h
o
n
e
s
fo
r
d
e
tec
ti
n
g
a
n
d
c
las
sify
in
g
a
g
g
re
ss
iv
e
d
riv
i
n
g
.
T
h
e
m
o
d
e
l
u
se
s
a
h
y
p
e
r
-
d
e
e
p
lea
rn
i
n
g
m
o
d
e
l
to
c
re
a
te
a
p
re
d
ic
ti
o
n
m
o
d
e
l
t
h
a
t
c
las
sifies
d
riv
e
rs
i
n
t
o
th
re
e
g
ro
u
p
s:
n
o
rm
a
l,
slo
w
,
a
n
d
a
g
g
re
ss
iv
e
.
Th
e
sy
ste
m
sta
rts
with
p
re
-
p
ro
c
e
ss
in
g
m
e
th
o
d
s
n
o
rm
a
li
z
a
ti
o
n
a
n
d
sta
n
d
a
r
d
sc
a
ler
a
p
p
ro
a
c
h
e
s
to
p
re
p
a
re
th
e
d
a
ta.
Two
m
e
th
o
d
o
lo
g
ies
a
re
u
se
d
:
d
irec
tl
y
e
n
terin
g
t
h
e
d
a
ta
in
to
t
h
e
d
e
e
p
m
o
d
e
l
to
c
las
sify
d
ri
v
in
g
b
e
h
a
v
io
r
a
n
d
se
lec
ti
n
g
fe
a
tu
re
s
u
si
n
g
p
ri
n
c
ip
a
l
c
o
m
p
o
n
e
n
t
a
n
a
l
y
sis
(P
CA),
sin
g
u
lar
v
a
l
u
e
d
e
c
o
m
p
o
siti
o
n
(S
VD
),
a
n
d
m
u
tu
a
l
in
f
o
rm
a
ti
o
n
(M
I)
.
T
h
e
h
y
p
e
r
-
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
two
r
k
(
CNN
)
-
d
e
n
se
m
o
d
e
l
is
t
h
e
n
u
se
d
to
trai
n
fe
a
tu
re
s
to
c
las
sify
d
ri
v
e
r
b
e
h
a
v
i
o
r.
T
h
e
e
x
p
e
rime
n
tal
re
su
l
ts
sh
o
w
th
a
t
th
e
CNN
-
d
e
n
se
m
o
d
e
l
wit
h
fe
a
tu
r
e
se
lec
ti
o
n
tec
h
n
iq
u
e
s
S
VD
6
a
n
d
M
I
6
a
c
h
iev
e
s
th
e
b
e
st
re
su
lt
s
with
1
0
0
%
a
c
c
u
ra
c
y
ra
te
fo
r
a
g
g
re
ss
iv
e
d
ri
v
e
r
b
e
h
a
v
i
o
r
d
e
tec
ti
o
n
,
w
h
il
e
t
h
e
ti
m
e
fo
r
S
VD
6
is t
h
e
sh
o
r
tes
t
a
t
4
3
se
c
o
n
d
s.
K
ey
w
o
r
d
s
:
Ag
g
r
ess
iv
e
d
r
iv
in
g
b
eh
a
v
io
r
C
o
n
v
o
lu
tio
n
n
eu
r
al
n
etwo
r
k
Dee
p
lear
n
in
g
Den
s
e
Featu
r
e
s
elec
tio
n
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
:
No
o
r
W
alid
Kh
alid
C
o
lleg
e
o
f
C
o
m
p
u
ter
Scien
ce
an
d
Ma
th
em
atics
,
T
ik
r
it Un
iv
er
s
ity
T
ik
r
it,
I
r
aq
E
m
ail:
n
o
o
r
walid
1
9
9
5
@
g
m
ail
.
co
m
1.
I
NT
RO
D
UCT
I
O
N
Dr
iv
in
g
b
e
h
av
io
r
s
ar
e
th
e
m
o
s
t
co
m
m
o
n
c
au
s
e
o
f
tr
a
f
f
ic
ac
cid
en
ts
an
d
a
la
r
g
e
co
n
tr
ib
u
tio
n
to
in
s
u
r
an
ce
claim
s
[
1
]
.
T
h
e
r
e
h
av
e
b
ee
n
tr
a
f
f
ic
ac
ci
d
en
ts
s
in
c
e
Kar
l
B
en
z
in
v
en
te
d
th
e
v
e
h
icle.
T
h
e
n
u
m
b
er
o
f
ca
r
s
o
n
th
e
r
o
a
d
in
c
r
ea
s
es
in
tan
d
em
with
th
e
ec
o
n
o
m
y
a
n
d
s
o
ciety
,
co
n
tr
i
b
u
tin
g
to
a
n
in
cr
ea
s
e
in
t
r
af
f
ic
ac
cid
en
ts
an
d
c
o
n
g
esti
o
n
[
2
]
.
Acc
o
r
d
in
g
to
r
esear
ch
,
h
u
m
an
f
ac
to
r
s
a
r
e
r
esp
o
n
s
ib
le
f
o
r
ab
o
u
t
9
0
%
o
f
r
o
ad
way
ac
cid
e
n
ts
[
3
]
.
Dr
iv
i
n
g
s
ty
le
ca
n
b
e
d
escr
ib
e
d
as
a
d
r
iv
er
’
s
h
ab
it
u
al
d
r
iv
in
g
b
e
h
av
io
r
th
at
r
e
f
lects
th
eir
ten
d
en
c
y
to
o
p
er
ate
in
p
ar
ticu
lar
way
s
r
e
g
u
lar
ly
[
4
]
.
I
t
also
d
escr
ib
es
h
o
w
a
d
r
iv
er
’
s
s
ty
le
o
f
d
r
iv
in
g
af
f
ec
ts
b
o
th
th
eir
o
wn
an
d
o
th
er
d
r
i
v
er
s
’
s
af
ety
th
r
o
u
g
h
d
r
iv
in
g
[
5
]
.
Ab
n
o
r
m
al
d
r
iv
in
g
is
d
ef
in
ed
as
a
b
n
o
r
m
al
o
r
u
n
s
af
e
b
eh
av
i
o
r
t
h
at
d
e
v
ia
tes
f
r
o
m
th
e
n
o
r
m
s
f
o
r
a
s
p
e
cif
ic
s
et
o
f
d
r
i
v
er
s
[
6
]
.
T
h
er
e
ar
e
o
th
er
ty
p
es
o
f
ir
r
eg
u
lar
d
r
iv
i
n
g
,
b
u
t
th
e
m
o
s
t
r
elev
an
t
b
e
h
av
io
r
s
,
lik
e
s
p
ee
d
in
g
,
a
g
g
r
ess
iv
e
d
r
iv
in
g
,
a
n
d
c
ar
eless
d
r
iv
in
g
,
ar
e
r
elate
d
to
an
in
cr
ea
s
ed
ch
a
n
ce
o
f
an
ac
cid
en
t [
7
]
.
R
o
ad
r
a
g
e
i
s
ch
ar
ac
ter
ized
b
y
v
e
r
b
al
ab
u
s
e,
s
h
o
v
in
g
,
h
itti
n
g
,
th
r
ea
ten
in
g
b
eh
a
v
io
r
,
an
d
m
ay
b
e
m
i
n
o
r
o
r
m
ajo
r
in
j
u
r
ie
s
[
8
]
.
I
t
is
d
escr
ib
ed
as
a
s
h
o
r
t
-
liv
ed
,
in
ten
s
e
em
o
tio
n
al
r
esp
o
n
s
e
to
p
e
r
ce
iv
ed
p
r
o
v
o
ca
tio
n
i
n
a
co
n
f
lict
s
itu
atio
n
in
v
o
lv
in
g
two
o
r
m
o
r
e
in
d
iv
id
u
als
o
n
th
e
r
o
ad
[
9
]
.
Sp
ee
d
in
g
,
tailg
atin
g
,
wea
v
in
g
in
an
d
o
u
t
o
f
tr
af
f
ic,
an
d
r
u
n
n
in
g
r
e
d
s
ig
n
als
ar
e
all
ex
am
p
les
o
f
ag
g
r
ess
iv
e
d
r
iv
in
g
[
1
0
]
.
Ac
co
r
d
in
g
to
a
s
u
r
v
ey
d
o
n
e
b
y
th
e
am
er
ican
au
to
m
o
b
ile
ass
o
ciatio
n
(
AAA)
f
o
u
n
d
atio
n
f
o
r
tr
af
f
ic
s
af
ety
,
ag
g
r
ess
iv
e
d
r
i
v
in
g
b
eh
a
v
io
r
(
ADB)
was
im
p
licated
in
r
o
u
g
h
ly
5
5
.
7
%
o
f
f
atal
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.
13
,
No
.
4
,
Dec
em
b
er
2
0
2
4
:
4
8
8
3
-
4
8
9
4
4884
tr
af
f
ic
ac
cid
en
ts
[
1
1
]
,
an
d
th
e
f
r
eq
u
en
c
y
o
f
r
o
a
d
ac
cid
en
ts
an
d
ADB
ar
e
p
o
s
itiv
ely
co
r
r
elate
d
[
1
2
]
.
ADB,
as
o
n
e
o
f
th
e
lead
in
g
ca
u
s
es
o
f
tr
af
f
ic
is
s
u
es,
is
in
f
lu
en
ce
d
b
y
b
o
th
s
itu
atio
n
al
co
n
d
itio
n
s
lik
e
tr
af
f
ic
c
o
n
g
esti
o
n
[
1
3
]
an
d
h
u
m
an
o
n
es
lik
e
n
eg
ativ
e
em
o
tio
n
s
[
1
4
]
.
B
ec
au
s
e
o
f
th
e
p
r
o
g
r
ess
iv
ely
co
n
g
ested
tr
af
f
ic
s
y
s
tem
an
d
th
e
r
ap
id
p
ac
e
o
f
life
,
it
is
ea
s
ier
f
o
r
d
r
iv
e
r
s
to
d
is
p
lay
ADB,
s
o
p
r
o
p
er
r
ec
o
g
n
itio
n
o
f
ADB
is
cr
itical.
Ho
wev
er
,
n
o
s
in
g
le
d
ef
in
iti
o
n
o
f
ADB
ex
is
ts
[
1
5
]
.
I
n
t
er
v
en
tio
n
s
o
f
tech
n
o
lo
g
y
in
h
ig
h
way
r
ag
e
an
d
ag
g
r
ess
iv
e
d
r
iv
in
g
ar
e
c
r
itical
to
ac
h
iev
in
g
th
is
g
o
al
[
1
6
]
.
Dee
p
lear
n
in
g
h
as
s
ee
n
f
ast
d
ev
elo
p
m
e
n
t
in
th
e
f
ield
o
f
d
r
iv
in
g
b
eh
av
io
r
id
en
t
if
icatio
n
in
r
ec
en
t
y
ea
r
s
[
1
7
]
.
I
t
ca
n
h
elp
wh
en
a
m
o
d
el
is
d
if
f
icu
lt
to
tr
ain
d
u
e
to
a
s
m
all
s
am
p
le
s
ize
o
r
wh
en
d
ata
co
llectio
n
is
p
r
o
b
lem
atic
in
th
e
tar
g
et
d
o
m
ain
[
1
8
]
.
Dee
p
lear
n
i
n
g
h
as
b
ee
n
u
s
ed
in
v
ar
io
u
s
s
tu
d
y
d
o
m
ain
s
d
u
e
to
its
u
s
u
al
ad
v
an
tag
es.
Fig
u
r
e
1
d
ep
icts
th
e
ch
ar
ac
ter
is
tics
th
a
t
in
f
lu
en
ce
d
r
iv
in
g
b
e
h
av
io
r
[
1
9
]
.
T
h
is
p
ap
er
aim
s
to
p
r
esen
t
a
m
eth
o
d
f
o
r
th
e
d
etec
tio
n
o
f
ADB
in
v
eh
icles,
f
o
cu
s
in
g
o
n
d
ev
elo
p
in
g
a
d
ee
p
lear
n
in
g
m
o
d
el
b
y
im
p
lem
e
n
tatio
n
a
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
(
C
NN)
to
th
e
id
en
tific
atio
n
an
d
class
if
icatio
n
o
f
d
r
i
v
in
g
b
e
h
av
io
r
s
,
with
a
f
o
cu
s
o
n
in
v
esti
g
atin
g
h
o
w
f
ea
tu
r
e
s
elec
tio
n
s
tr
ateg
ies
af
f
ec
t
m
o
d
el
p
e
r
f
o
r
m
an
ce
,
th
is
is
s
o
m
et
h
in
g
th
at
p
r
ev
io
u
s
s
tu
d
ies
d
id
n
o
t
g
iv
e
m
u
ch
atten
tio
n
to
it.
W
e
will
c
o
n
d
u
ct
a
co
m
p
ar
at
iv
e
an
aly
s
is
b
etwe
en
th
e
C
N
N
m
o
d
el
with
f
ea
t
u
r
e
s
elec
tio
n
an
d
th
e
m
o
d
el
with
o
u
t
f
ea
tu
r
e
s
elec
tio
n
,
ev
al
u
ate
an
d
q
u
an
tif
y
th
e
im
p
ac
t
o
f
em
p
lo
y
i
n
g
f
ea
tu
r
e
s
elec
tio
n
tech
n
iq
u
es
o
n
k
e
y
p
er
f
o
r
m
an
ce
m
etr
ics
to
d
is
ce
r
n
th
e
ef
f
ec
tiv
e
n
ess
o
f
th
ese
m
eth
o
d
s
.
Dem
o
n
s
tr
ate
h
o
w
th
e
p
r
o
p
o
s
ed
d
ee
p
lear
n
in
g
m
o
d
el
co
n
tr
ib
u
tes
to
ad
v
an
ce
m
e
n
ts
in
th
e
f
ield
o
f
d
r
iv
in
g
b
eh
a
v
io
r
class
if
icatio
n
.
Pre
s
en
t
n
ew
in
s
ig
h
ts
,
im
p
r
o
v
ed
m
eth
o
d
o
l
o
g
ies,
an
d
p
o
ten
tial
ap
p
licatio
n
s
th
at
ca
n
s
ig
n
if
ican
tly
en
h
an
c
e
th
e
d
etec
tio
n
an
d
u
n
d
er
s
tan
d
i
n
g
o
f
ADB
.
Hig
h
lig
h
t a
d
v
a
n
ce
m
en
ts
ac
h
iev
e
d
an
d
s
h
o
wca
s
e
th
e
m
o
d
el
’
s
h
ig
h
e
r
p
er
f
o
r
m
an
ce
.
T
h
e
r
em
ain
d
er
o
f
th
is
p
a
p
er
i
s
s
tr
u
ctu
r
ed
as
f
o
llo
ws:
s
ec
tio
n
2
p
r
o
v
i
d
es
an
in
-
d
ep
th
ex
a
m
in
atio
n
o
f
d
r
iv
in
g
b
e
h
av
io
r
d
etec
tio
n
an
d
d
ee
p
lear
n
in
g
ap
p
licatio
n
s
.
S
ec
tio
n
3
d
escr
ib
es
t
h
e
ADB
d
etec
tio
n
m
ec
h
an
is
m
,
wh
ich
is
b
ased
o
n
h
y
p
er
-
d
ee
p
lear
n
in
g
.
Sectio
n
4
p
r
o
v
id
es
th
e
co
m
p
ar
is
o
n
f
in
d
i
n
g
s
an
d
a
d
is
cu
s
s
io
n
o
f
th
e
im
p
lem
en
tat
io
n
o
f
th
e
p
r
o
p
o
s
ed
d
ee
p
AD
B
d
etec
tio
n
m
o
d
el
with
an
d
with
o
u
t
u
s
in
g
f
ea
tu
r
e
s
elec
tio
n
.
Sectio
n
5
s
u
m
m
ar
iz
es th
e
co
n
clu
s
io
n
s
o
f
t
h
is
s
tu
d
y
.
Fig
u
r
e
1
.
Facto
r
s
in
f
l
u
en
cin
g
d
r
iv
in
g
b
eh
av
i
o
r
[
1
9
]
2.
L
I
T
E
R
A
T
U
RE
S
U
RVE
Y
T
h
is
s
ec
tio
n
in
clu
d
es
a
co
m
p
r
eh
en
s
iv
e
r
e
v
iew
o
f
liter
atu
r
e
r
an
g
in
g
f
r
o
m
r
ep
r
esen
tati
v
e
wo
r
k
s
r
an
g
in
g
f
r
o
m
th
e
o
ld
est
to
th
e
latest
ar
o
u
n
d
th
is
s
tu
d
y
.
Se
v
er
al
way
s
t
o
d
etec
t
d
r
i
v
in
g
b
eh
av
io
r
h
a
v
e
b
ee
n
p
r
o
p
o
s
ed
o
v
er
th
e
last
two
d
ec
ad
es
.
Mo
u
k
af
ih
et
a
l
.
[
2
0
]
p
r
o
p
o
s
ed
ag
g
r
ess
iv
e
d
r
iv
er
b
e
h
av
io
r
class
if
icatio
n
m
o
d
el
u
s
in
g
lo
n
g
s
h
o
r
t
-
te
r
m
m
em
o
r
y
(
L
STM
)
-
f
u
lly
c
o
n
v
o
lu
tio
n
al
n
etwo
r
k
(
FC
N)
with
r
ea
l
-
wo
r
ld
d
r
i
v
in
g
d
ata
f
r
o
m
m
o
b
ile
p
h
o
n
es.
T
h
e
UAH
-
d
r
iv
e
s
et
d
ataset
is
u
s
ed
to
v
alid
ate
th
e
tech
n
i
q
u
e
.
T
h
e
m
eth
o
d
o
u
tp
er
f
o
r
m
s
o
t
h
er
d
ee
p
lear
n
in
g
an
d
co
n
v
en
tio
n
al
m
ac
h
i
n
e
lear
n
in
g
m
o
d
els
in
ter
m
s
o
f
ac
cu
r
ac
y
,
with
a
9
5
.
8
8
%
ac
c
u
r
ac
y
s
co
r
e
f
o
r
a
5
-
m
in
u
te
win
d
o
w
d
u
r
atio
n
.
Ma
to
u
s
ek
et
a
l
.
[
2
1
]
f
o
cu
s
e
d
o
n
d
ev
elo
p
i
n
g
a
r
eliab
le
m
eth
o
d
f
o
r
id
en
tif
y
i
n
g
u
n
u
s
u
al
d
r
iv
in
g
b
eh
a
v
io
r
u
s
in
g
n
e
u
r
al
n
etwo
r
k
s
.
T
h
e
y
co
m
p
ar
e
L
STM
n
etwo
r
k
s
an
d
Au
to
E
n
c
o
d
er
r
ep
licato
r
n
eu
r
al
n
etwo
r
k
s
to
an
is
o
latio
n
f
o
r
est.
T
h
ey
s
h
o
w
th
at
a
r
ec
u
r
r
en
t
n
eu
r
al
n
etwo
r
k
(
R
NN)
ca
n
r
eliab
ly
d
etec
t
an
o
m
alies
in
d
r
iv
in
g
b
eh
av
i
o
r
,
with
an
ac
c
u
r
ac
y
r
ate
o
f
9
3
%,
m
ak
in
g
it
s
u
itab
le
f
o
r
lar
g
e
-
s
ca
le
d
etec
tio
n
s
y
s
tem
s
.
Xin
g
et
a
l
.
[
2
2
]
d
e
v
elo
p
ed
a
R
NN
to
ad
d
r
ess
d
r
iv
e
r
b
eh
av
io
r
p
r
o
f
ilin
g
as
an
an
o
m
aly
d
etec
tio
n
p
r
o
b
lem
.
T
h
e
m
o
d
el,
tr
ain
e
d
o
n
d
ata
f
r
o
m
ty
p
ical
d
r
iv
er
s
,
p
r
o
d
u
ce
d
s
ig
n
if
ican
t
r
eg
r
ess
io
n
er
r
o
r
wh
en
p
r
ed
ictin
g
ADB
,
b
u
t
lo
w
er
r
o
r
wh
e
n
r
ec
o
g
n
izi
n
g
r
eg
u
lar
d
r
iv
in
g
b
eh
av
io
r
.
T
h
e
m
o
d
el
ac
h
iev
e
d
an
ac
cu
r
ac
y
r
ate
o
f
8
8
%
wh
en
class
if
y
in
g
ADB
,
s
u
g
g
esti
n
g
it
co
u
ld
b
e
a
u
s
ef
u
l
b
aselin
e
f
o
r
u
n
s
u
p
er
v
is
ed
d
r
iv
er
p
r
o
f
ilin
g
an
d
c
o
n
tr
ib
u
tin
g
to
a
s
m
ar
t
tr
a
n
s
p
o
r
tatio
n
ec
o
lo
g
y
.
T
aleb
lo
o
et
a
l
.
[
2
3
]
p
r
o
p
o
s
ed
a
m
et
h
o
d
to
d
etec
t
ADB
u
s
i
n
g
GPS
s
en
s
o
r
s
o
n
s
m
ar
t
p
h
o
n
es.
T
h
ey
class
if
y
d
r
iv
er
s
’
d
r
i
v
in
g
b
eh
av
io
r
ev
e
r
y
th
r
ee
m
i
n
u
tes
u
s
in
g
R
NN
alg
o
r
ith
m
s
,
ig
n
o
r
in
g
r
o
ad
c
o
n
d
itio
n
s
o
r
d
r
iv
er
’
s
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
A
d
etec
tio
n
mo
d
el
o
f a
g
g
r
ess
i
ve
d
r
ivin
g
b
eh
a
vio
r
b
a
s
ed
o
n
h
yb
r
id
d
ee
p
le
a
r
n
in
g
(
N
o
o
r
W
a
lid
K
h
a
lid
)
4885
b
eh
av
io
r
.
T
h
e
alg
o
r
ith
m
,
wh
ic
h
u
s
es
1
2
0
s
ec
o
n
d
s
o
f
GPS
d
ata,
h
as
a
9
3
%
ac
cu
r
ac
y
r
ate
in
id
en
tify
in
g
v
io
len
t
d
r
iv
in
g
b
eh
av
i
o
r
,
in
d
icatin
g
th
at
th
r
ee
m
in
u
tes
o
r
m
o
r
e
o
f
d
r
iv
i
n
g
is
s
u
f
f
icien
t
.
Al
-
Hu
s
s
ein
et
a
l
.
[
2
4
]
p
r
esen
ted
a
m
eth
o
d
f
o
r
p
r
o
f
ili
n
g
d
r
iv
er
b
eh
av
i
o
r
u
s
in
g
s
eg
m
en
t
lab
elin
g
an
d
r
o
w
la
b
elin
g
.
A
s
af
ety
g
r
ad
e
is
ass
ig
n
ed
b
y
r
o
w
lab
elin
g
t
o
e
v
er
y
s
ec
o
n
d
o
f
d
r
iv
i
n
g
d
ata,
wh
ile
s
eg
m
en
t
lab
elin
g
g
r
ad
e
s
tem
p
o
r
al
s
eg
m
en
ts
b
ased
o
n
n
o
r
m
s
.
T
h
e
r
esear
ch
u
s
es
th
r
ee
d
ee
p
-
lea
r
n
in
g
-
b
ase
d
alg
o
r
ith
m
s
:
d
ee
p
n
eu
r
al
n
et
wo
r
k
(
DNN)
,
R
NN,
an
d
C
NN
to
class
if
y
r
ec
o
r
d
ed
d
r
iv
i
n
g
d
ata.
C
NN
was
s
u
g
g
ested
f
o
r
th
e
s
y
s
tem
o
f
id
en
tific
atio
n
,
o
u
tp
er
f
o
r
m
in
g
th
e
o
th
er
two
t
ec
h
n
iq
u
es with
9
6
.
1
% a
cc
u
r
ac
y
.
T
h
e
s
tu
d
y
s
u
g
g
ests
th
at
th
is
r
ec
o
g
n
itio
n
s
y
s
tem
co
u
ld
in
cr
ea
s
e
r
o
ad
s
af
ety
.
T
h
e
r
esear
ch
aim
s
to
av
o
id
o
v
er
f
itti
n
g
an
d
im
p
r
o
v
e
r
o
ad
s
af
ety
.
Al
-
H
u
s
s
ei
n
et
a
l
.
[
2
4
]
p
r
o
p
o
s
e
d
a
n
A
DB
r
ec
o
g
n
iti
o
n
te
ch
n
i
q
u
e
u
s
i
n
g
c
o
ll
ec
ti
v
e
l
ea
r
n
i
n
g
.
T
h
e
m
aj
o
r
it
y
class
is
g
r
o
u
p
e
d
u
s
i
n
g
a
s
el
f
-
o
r
g
a
n
iz
in
g
m
a
p
a
n
d
li
n
k
ed
wit
h
th
e
m
i
n
o
r
i
ty
cl
as
s
to
c
r
ea
te
m
u
lti
p
l
e
class
-
b
al
an
ci
n
g
d
atas
ets
.
T
h
e
c
lass
i
f
i
er
s
ar
e
b
u
i
lt
u
s
i
n
g
C
NN,
L
S
T
M
,
a
n
d
g
ate
d
r
e
c
u
r
r
en
t
u
n
it
(
GR
U
)
tec
h
n
i
q
u
es.
T
h
e
en
s
em
b
l
e
cl
a
s
s
if
i
er
is
b
ett
er
s
u
i
te
d
f
o
r
i
d
e
n
tif
y
i
n
g
ADBs
i
n
a
ti
n
y
p
er
ce
n
tag
e
o
f
th
e
d
at
aset
,
wh
i
le
th
e
class
if
ie
r
w
it
h
o
u
t
e
n
s
e
m
b
le
le
ar
n
i
n
g
is
b
e
tte
r
f
o
r
d
ete
cti
n
g
m
o
r
e
ab
u
n
d
a
n
t
A
D
B
s
.
T
h
e
L
STM
a
n
d
p
r
o
d
u
ct
r
u
le
-
b
ase
d
e
n
s
em
b
l
e
c
lass
i
f
ie
r
h
as
t
h
e
h
i
g
h
est
a
cc
u
r
a
cy
o
f
9
0
.
5
%
[
2
5
]
.
E
s
co
ttá
et
a
l
.
[
2
6
]
u
s
e
d
i
n
e
r
ti
al
m
ea
s
u
r
e
m
e
n
t
u
n
it
(
I
MU
)
s
e
n
s
o
r
s
o
n
s
m
a
r
t
p
h
o
n
es
t
o
id
e
n
t
if
y
d
r
i
v
i
n
g
e
v
en
ts
u
s
i
n
g
l
in
e
ar
a
cc
e
le
r
at
io
n
a
n
d
an
g
u
la
r
v
e
lo
cit
y
s
i
g
n
a
ls
.
T
h
e
y
e
v
a
lu
at
ed
d
ee
p
-
le
ar
n
i
n
g
m
o
d
els
u
s
i
n
g
1
D
a
n
d
2
D
C
N
Ns,
a
c
h
ie
v
i
n
g
h
i
g
h
ac
c
u
r
a
cy
v
al
u
es
o
f
u
p
t
o
8
2
.
4
0
%
.
C
o
j
o
c
ar
u
et
a
l
.
[
2
7
]
p
r
ese
n
te
d
a
d
ee
p
lea
r
n
i
n
g
-
b
as
e
d
d
r
iv
in
g
b
e
h
av
io
r
esti
m
ati
o
n
s
y
s
te
m
i
n
t
e
g
r
ate
d
in
t
o
a
r
id
e
-
s
h
ar
in
g
ap
p
l
ica
ti
o
n
.
R
esu
lts
t
h
a
t
u
s
e
d
t
h
e
d
r
i
v
i
n
g
b
e
h
a
v
i
o
r
d
atas
et
s
h
o
w
b
et
te
r
a
cc
u
r
ac
y
w
it
h
t
wo
class
es,
wit
h
C
NN
-
L
S
T
M
ac
h
i
ev
in
g
th
e
b
est
r
esu
lt
s
at
9
1
.
9
4
%,
an
d
C
o
n
v
L
STM
o
u
t
p
er
f
o
r
m
i
n
g
c
l
ass
ic
al
L
STM
n
etw
o
r
k
s
[
2
7
]
.
C
o
j
o
c
ar
u
a
n
d
Po
p
es
c
u
[
2
8
]
s
h
o
we
d
a
d
at
ase
t
co
l
le
cte
d
u
t
ili
zi
n
g
a
n
An
d
r
o
i
d
s
m
a
r
t
p
h
o
n
e
t
h
a
t
e
x
cl
u
s
i
v
el
y
u
ti
liz
es
s
en
s
o
r
d
a
ta
f
r
o
m
t
h
e
s
m
ar
t
p
h
o
n
e.
T
h
e
d
at
ase
t
is
cl
ass
if
ie
d
i
n
t
o
t
h
r
ee
ca
t
eg
o
r
ies
:
s
lo
w,
n
o
r
m
a
l,
a
n
d
ag
g
r
ess
i
v
e
,
a
n
d
it
is
a
cc
o
m
p
a
n
ied
b
y
e
x
p
e
r
im
e
n
ts
aim
ed
at
o
f
f
e
r
i
n
g
in
s
ig
h
t
i
n
t
o
th
e
d
a
ta
c
ap
ac
it
y
.
T
h
ey
p
r
o
p
o
s
ed
C
NN
,
L
S
T
M
,
a
n
d
C
o
n
v
L
STM
m
o
d
els
u
s
i
n
g
th
r
e
e
m
a
c
h
i
n
e
l
ea
r
n
in
g
t
ec
h
n
i
q
u
es
.
T
h
e
r
es
u
l
ts
s
h
o
w
t
h
a
t
C
o
n
v
L
S
T
M
a
c
h
ie
v
ed
t
h
e
h
i
g
h
es
t
a
cc
u
r
ac
y
o
f
7
9
.
5
%.
Ab
o
s
a
q
et
a
l
.
[
2
9
]
s
u
g
g
este
d
d
e
ep
le
ar
n
i
n
g
-
b
ase
d
d
e
tec
ti
o
n
m
e
th
o
d
s
f
o
r
a
n
o
m
a
lo
u
s
d
r
i
v
i
n
g
b
e
h
a
v
i
o
r
u
s
i
n
g
a
d
at
ase
t
wi
th
f
iv
e
ca
t
eg
o
r
ies
.
T
h
e
p
r
o
p
o
s
ed
C
NN
-
b
ase
d
m
o
d
el
o
u
t
p
e
r
f
o
r
m
s
p
r
e
-
t
r
a
in
e
d
m
o
d
els
i
n
p
e
r
f
o
r
m
a
n
c
e
m
et
r
i
cs,
a
ch
ie
v
i
n
g
8
9
%,
9
3
%
,
9
3
%
,
9
4
%
,
an
d
9
5
%
a
cc
u
r
ac
y
in
cl
ass
i
f
y
in
g
d
r
i
v
er
’
s
u
n
u
s
u
al
co
n
d
u
ct
.
3.
M
E
T
H
O
DO
L
O
G
Y
T
h
e
m
eth
o
d
o
l
o
g
y
u
s
ed
in
th
i
s
s
tu
d
y
to
co
m
b
in
e
f
ea
tu
r
e
r
e
d
u
ctio
n
with
r
a
p
id
h
y
p
e
r
-
d
ee
p
lear
n
in
g
m
eth
o
d
s
f
o
r
p
r
ec
is
e
class
if
icatio
n
o
f
ADB
is
p
r
esen
te
d
in
t
h
is
s
ec
tio
n
o
f
th
e
p
a
p
er
.
T
h
e
m
eth
o
d
ical
p
r
o
ce
s
s
em
p
lo
y
ed
to
cr
e
ate
a
s
tr
o
n
g
s
y
s
tem
th
at
ca
n
r
eliab
ly
an
d
p
r
ec
is
ely
r
ec
o
g
n
ize
ADB
is
d
e
s
cr
ib
ed
in
th
is
s
ec
tio
n
.
T
h
e
m
o
d
el
ac
cu
r
ately
class
if
ies
d
r
iv
in
g
b
eh
a
v
io
r
s
in
to
th
r
ee
ca
teg
o
r
ies:
s
lo
w,
n
o
r
m
al,
an
d
ag
g
r
ess
iv
e.
I
t
d
o
es
th
is
b
y
u
tili
zin
g
f
ea
tu
r
e
s
elec
tio
n
an
d
r
e
d
u
ctio
n
ap
p
r
o
ac
h
es
i
n
co
n
j
u
n
ctio
n
with
th
e
ca
p
ab
il
ities
o
f
DNN
.
T
h
e
m
o
d
el’
s
ab
ilit
y
to
d
is
cr
im
in
ate
b
etwe
en
d
if
f
er
e
n
t
b
e
h
av
io
r
ca
teg
o
r
ies
with
ac
cu
r
ac
y
ca
n
s
u
p
p
o
r
t
p
r
o
ac
tiv
e
ef
f
o
r
ts
to
im
p
r
o
v
e
tr
af
f
ic
co
n
tr
o
l
a
n
d
r
o
ad
s
af
ety
.
T
h
e
n
ex
t
s
ec
tio
n
s
ex
p
lain
th
e
p
r
o
c
ed
u
r
es,
e
v
alu
atio
n
s
tr
ateg
ies,
an
d
m
et
h
o
d
o
lo
g
ies
em
p
lo
y
ed
i
n
th
is
r
esear
ch
p
r
o
ject,
g
o
i
n
g
i
n
to
d
etail
ab
o
u
t
ea
ch
s
tag
e
o
f
th
e
p
r
o
ce
s
s
.
T
h
e
s
y
s
tem
’
s
co
m
p
o
n
en
ts
,
wh
ich
in
clu
d
e
t
h
e
d
r
i
v
in
g
b
eh
a
v
io
r
co
m
p
o
n
en
t,
ar
e
s
h
o
wn
in
Fig
u
r
e
2
.
3
.
1
.
Driv
ing
beha
v
io
r
da
t
a
s
e
t
des
cr
iptio
n
O
u
r
m
a
i
n
o
b
j
e
c
t
i
v
e
is
t
o
p
r
e
s
e
n
t
a
t
h
o
r
o
u
g
h
c
o
m
p
r
e
h
e
n
s
i
o
n
o
f
t
h
e
d
a
t
a
s
e
t
u
s
e
d
i
n
o
u
r
s
tu
d
y
i
n
t
h
e
s
e
c
ti
o
n
d
e
v
o
t
e
d
t
o
d
a
t
a
s
et
g
a
th
e
r
i
n
g
a
n
d
d
e
s
c
r
i
p
t
i
o
n
.
W
e
u
n
d
e
r
s
t
a
n
d
t
h
a
t
t
h
e
e
f
f
e
ct
i
v
e
n
es
s
o
f
d
e
e
p
l
e
a
r
n
i
n
g
m
o
d
e
l
s
d
e
p
e
n
d
s
c
r
i
ti
c
a
ll
y
o
n
h
i
g
h
-
q
u
a
l
i
t
y
d
a
t
a
.
W
e
s
h
a
ll
p
r
o
v
i
d
e
c
o
m
p
r
e
h
e
n
s
i
v
e
i
n
f
o
r
m
a
ti
o
n
i
n
t
h
e
e
n
s
u
i
n
g
s
u
b
s
e
ct
i
o
n
s
t
o
a
c
h
ie
v
e
t
h
is
g
o
al
.
I
m
p
o
r
t
a
n
t
d
et
a
i
ls
l
i
k
e
t
h
e
d
a
ta
g
a
t
h
e
r
i
n
g
p
r
o
c
es
s
,
t
h
e
s
o
u
r
c
es
it
c
a
m
e
f
r
o
m
,
a
n
d
t
h
e
d
a
t
a
c
le
a
n
i
n
g
a
n
d
p
r
e
p
r
o
c
e
s
s
i
n
g
s
t
e
p
s
wi
l
l
al
l
b
e
c
o
v
e
r
e
d
i
n
o
u
r
i
n
v
e
s
t
i
g
at
i
o
n
.
W
e
h
o
p
e
t
h
a
t
t
h
is
t
h
o
r
o
u
g
h
e
x
p
l
a
n
a
t
i
o
n
wi
l
l
p
r
o
v
i
d
e
r
e
a
d
e
r
s
wi
t
h
a
s
t
r
o
n
g
b
a
s
is
f
o
r
co
m
p
r
e
h
e
n
d
i
n
g
t
h
e
c
o
n
t
e
x
t
o
f
t
h
e
d
a
ta
s
et
a
n
d
i
t
s
s
i
g
n
i
f
i
c
a
n
ce
t
o
o
u
r
r
e
s
e
a
r
c
h
.
T
h
e
d
a
t
a
s
e
t
u
s
e
d
i
n
t
h
i
s
s
t
u
d
y
c
l
o
s
e
l
y
m
a
t
c
h
e
s
t
h
e
g
o
a
ls
o
f
t
h
e
r
e
s
e
a
r
c
h
a
s
w
e
l
l
as
t
h
e
r
e
q
u
i
r
e
m
e
n
t
f
o
r
h
i
g
h
-
q
u
a
l
i
t
y
d
a
t
a
t
o
t
r
a
i
n
h
y
p
e
r
-
d
e
e
p
m
o
d
e
l
.
O
u
r
s
t
u
d
y
f
o
c
u
s
es
o
n
d
e
t
e
c
ti
n
g
a
n
d
c
l
a
s
s
i
f
y
i
n
g
d
r
i
v
i
n
g
b
e
h
a
v
i
o
r
i
n
t
o
t
h
r
e
e
g
r
o
u
p
s
:
n
o
r
m
a
l
,
a
g
g
r
e
s
s
i
v
e
,
a
n
d
s
l
o
w
.
E
i
g
h
t
f
e
a
t
u
r
es
m
a
k
e
u
p
t
h
e
d
a
t
a
s
e
t
t
h
at
t
h
e
a
p
p
l
i
c
a
t
i
o
n
u
s
e
s
[
2
7
]
,
[
2
8
]
:
i
)
t
h
r
e
e
f
o
r
t
h
e
a
c
c
e
l
e
r
a
ti
o
n
i
n
m
e
t
e
r
s
p
e
r
s
e
c
o
n
d
s
q
u
a
r
e
d
o
n
X
,
Y
,
a
n
d
Z
a
x
e
s
;
ii
)
t
h
r
e
e
(
X
,
Y
,
Z
)
a
x
es
r
o
t
at
i
o
n
in
d
e
g
r
e
e
s
p
e
r
s
e
c
o
n
d
(
°
/s
)
;
i
ii
)
l
a
b
e
l
f
o
r
c
l
as
s
i
f
i
c
a
ti
o
n
(
a
g
g
r
e
s
s
i
v
e
,
n
o
r
m
a
l
,
s
l
o
w
)
;
a
n
d
i
v
)
d
a
t
e
a
n
d
t
i
m
e
s
ta
m
p
.
On
l
y
t
h
e
a
c
ce
l
e
r
o
m
e
t
e
r
a
n
d
g
y
r
o
s
c
o
p
e
w
e
r
e
u
t
il
i
z
e
d
a
s
t
h
e
p
r
i
m
a
r
y
s
e
n
s
o
r
s
,
a
n
d
t
h
e
d
a
t
a
w
as
g
a
t
h
e
r
e
d
i
n
s
a
m
p
l
es
(
tw
o
s
a
m
p
l
es
p
e
r
s
e
c
o
n
d
)
a
f
t
e
r
th
e
g
r
a
v
i
t
a
t
i
o
n
a
l
ac
c
e
l
e
r
at
i
o
n
wa
s
e
li
m
i
n
a
t
e
d
.
T
h
e
d
ataset
u
s
ed
in
th
is
s
tu
d
y
was
s
o
u
r
ce
d
f
r
o
m
Ka
g
g
le,
a
p
o
p
u
lar
o
n
lin
e
p
latf
o
r
m
f
o
r
s
h
ar
in
g
d
ataset1
.
T
h
e
d
ata
co
llectio
n
p
r
o
ce
s
s
in
v
o
lv
ed
m
eticu
l
o
u
s
r
ec
o
r
d
in
g
u
s
in
g
a
Sam
s
u
n
g
Gala
x
y
S1
0
s
m
ar
tp
h
o
n
e
an
d
a
Dac
ia
Sen
d
er
o
1
.
4
MPI
v
eh
icle.
I
n
te
r
m
s
o
f
th
e
ch
o
ice
o
f
v
eh
icle
f
o
r
d
ata
co
llectio
n
,
a
s
tan
d
ar
d
ca
r
with
7
5
h
o
r
s
ep
o
wer
was
s
elec
ted
.
T
h
e
g
eo
s
p
atial
co
v
er
ag
e
o
f
th
e
d
ataset
is
f
o
cu
s
ed
o
n
t
h
e
city
o
f
C
r
aio
v
a,
lo
ca
ted
i
n
th
e
D
o
lj
r
eg
io
n
o
f
R
o
m
an
ia.
T
h
is
s
p
ec
if
ic
r
eg
io
n
was
ch
o
s
en
as
th
e
d
ata
co
llectio
n
a
r
ea
to
p
r
o
v
id
e
a
l
o
ca
lized
p
e
r
s
p
ec
tiv
e
an
d
ac
c
o
u
n
t
f
o
r
an
y
u
n
iq
u
e
ch
ar
ac
te
r
is
tics
o
r
d
y
n
a
m
ics
p
r
esen
t
in
th
at
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.
13
,
No
.
4
,
Dec
em
b
er
2
0
2
4
:
4
8
8
3
-
4
8
9
4
4886
lo
ca
tio
n
.
T
h
e
d
ataset
em
p
lo
y
ed
in
th
e
p
r
o
p
o
s
ed
m
o
d
el
is
d
escr
ib
ed
in
ter
m
s
o
f
its
ch
ar
ac
ter
is
tics
.
T
h
e
s
u
m
m
ar
ized
in
f
o
r
m
atio
n
is
p
r
esen
ted
in
T
ab
le
1
,
wh
ich
p
r
o
v
id
es
a
n
o
v
er
v
iew
o
f
th
e
d
ataset’
s
attr
ib
u
tes,
v
alu
es,
an
d
o
th
er
p
er
tin
en
t c
h
a
r
ac
ter
is
tics
.
Fig
u
r
e
2
.
T
h
e
p
r
o
p
o
s
ed
s
y
s
te
m
ar
ch
itectu
r
e
T
ab
le
1
.
C
h
ar
ac
ter
is
tics
an
d
v
alu
es o
f
th
e
d
ataset
em
p
lo
y
ed
in
th
e
p
r
o
p
o
s
ed
m
o
d
el
C
h
a
r
a
c
t
e
r
i
s
t
i
c
S
p
e
c
i
f
i
c
a
t
i
o
n
D
a
t
a
s
e
t
n
a
me
D
r
i
v
i
n
g
b
e
h
a
v
i
o
r
N
u
mb
e
r
o
f
sam
p
l
e
s
3
6
4
4
N
u
mb
e
r
o
f
f
e
a
t
u
r
e
s
8
M
i
ss
i
n
g
d
a
t
a
No
B
a
l
a
n
c
e
d
d
a
t
a
s
e
t
Y
e
s
La
b
e
l
Y
e
s
3
.
2
.
Da
t
a
p
re
pro
ce
s
s
ing
On
e
o
f
th
e
m
o
s
t
cr
u
cial
p
h
ases
o
f
ap
p
licatio
n
s
f
o
r
d
ata
an
aly
s
is
i
s
d
ata
p
r
ep
r
o
ce
s
s
in
g
.
Ma
n
y
in
co
n
s
is
ten
cies,
o
u
t
-
of
-
r
a
n
g
e
n
u
m
b
er
s
,
m
is
s
in
g
v
alu
es,
n
o
is
es,
an
d
/o
r
ex
ce
s
s
es
ar
e
am
o
n
g
th
e
n
u
m
er
o
u
s
d
ef
ec
ts
th
at
ar
e
f
r
eq
u
en
tly
p
r
esen
t
in
r
aw
d
ata.
L
o
w
-
q
u
a
lity
d
ata
will
im
p
ed
e
th
e
le
ar
n
in
g
a
n
d
m
in
in
g
alg
o
r
ith
m
s
’
ab
ilit
y
to
f
u
n
ctio
n
well
in
th
e
u
p
co
m
in
g
s
tag
e
s
.
B
ec
au
s
e
o
f
th
is
,
n
u
m
er
o
u
s
p
r
ep
r
o
ce
s
s
in
g
s
tep
s
m
u
s
t
b
e
co
m
p
leted
to
im
p
r
o
v
e
th
e
q
u
ality
o
f
r
aw
d
ata.
Un
d
e
r
th
is
to
p
ic,
s
o
m
e
o
f
th
e
m
o
s
t
p
o
p
u
lar
a
n
d
u
s
ef
u
l
d
ata
p
r
ep
ar
atio
n
m
eth
o
d
s
f
o
r
u
s
e
in
d
ata
an
aly
s
is
ap
p
licatio
n
s
ar
e
r
ev
iewe
d
in
ter
m
s
o
f
u
s
ag
e,
p
o
p
u
lar
ity
,
an
d
th
e
alg
o
r
ith
m
s
th
at
s
u
p
p
o
r
t
th
em
[
3
0
]
.
I
n
t
h
is
wo
r
k
,
two
c
o
m
m
o
n
l
y
u
s
ed
tech
n
iq
u
es
in
d
ata
p
r
ep
r
o
ce
s
s
in
g
wer
e
u
s
ed
.
T
h
ese
tech
n
iq
u
es a
r
e
n
o
r
m
aliza
tio
n
an
d
s
tan
d
ar
d
s
ca
ler
.
3.
2
.
1.
No
rma
lize
d
a
t
a
No
r
m
alize
d
ata:
n
o
r
m
aliza
tio
n
,
wh
ich
in
v
o
l
v
es
s
ca
lin
g
f
e
atu
r
e
d
ata
t
o
s
p
ec
if
ic
in
te
r
v
a
ls
s
u
ch
as
[
-
1
.
0
,
1
.
0
]
o
r
[
0
.
0
,
1
.
0
]
,
is
u
s
u
ally
r
eq
u
ir
e
d
wh
e
n
a
d
ataset
c
o
n
tain
s
f
ea
tu
r
es
with
v
er
y
d
if
f
er
en
t
s
ca
les.
I
f
n
o
t,
f
ea
tu
r
es
with
v
alu
es
o
n
a
m
u
ch
lar
g
er
s
ca
le
m
ig
h
t
m
ak
e
a
s
m
aller
s
ca
led
b
u
t
s
till
s
ig
n
i
f
ican
t
f
ea
tu
r
e
less
ef
f
ec
tiv
e
[
3
1
]
.
T
h
is
will
h
av
e
a
d
etr
im
en
tal
ef
f
ec
t
o
n
t
h
e
d
ata
m
in
in
g
m
o
d
el’
s
ac
cu
r
ac
y
p
er
f
o
r
m
an
ce
.
T
o
eq
u
alize
th
e
s
ize
o
f
th
e
f
ea
tu
r
es,
th
e
n
o
r
m
alizin
g
tech
n
iq
u
e
is
th
er
ef
o
r
e
d
o
n
e
to
th
em
.
T
h
e
th
r
ee
m
o
s
t
u
s
ed
tech
n
iq
u
es
ar
e
d
ec
im
al
s
ca
le
n
o
r
m
aliza
tio
n
,
z
-
s
co
r
e
n
o
r
m
al
izatio
n
,
an
d
m
in
-
m
a
x
n
o
r
m
ali
zin
g
[
3
2
]
.
Min
-
m
ax
n
o
r
m
aliza
tio
n
:
T
h
e
d
if
f
er
e
n
c
e
b
etwe
en
th
e
d
ata’
s
lar
g
es
t
an
d
lo
west
v
alu
es
is
u
s
e
d
to
ca
lcu
late
th
e
n
o
r
m
aliza
tio
n
.
In
(
1
)
d
is
p
lay
s
th
e
v
alu
es
o
f
th
e
f
ea
tu
r
e
as
m
in
,
m
ax
,
an
d
v
,
th
e
v
alu
es
to
b
e
n
o
r
m
alize
d
,
a
n
d
th
e
n
ew
r
an
g
e
to
b
e
n
o
r
m
alize
d
is
r
ep
r
esen
ted
b
y
an
d
[
3
3
]
.
=
−
mi
n
(
)
max
(
)
−
mi
n
(
)
(
-
)
+
(
1
)
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
A
d
etec
tio
n
mo
d
el
o
f a
g
g
r
ess
i
ve
d
r
ivin
g
b
eh
a
vio
r
b
a
s
ed
o
n
h
yb
r
id
d
ee
p
le
a
r
n
in
g
(
N
o
o
r
W
a
lid
K
h
a
lid
)
4887
W
h
er
e
x
n
ew
r
ep
r
esen
ts
n
o
r
m
alize
d
x
.
W
e
im
p
lem
en
ted
n
o
r
m
aliza
tio
n
tech
n
i
q
u
es,
n
o
tab
ly
m
in
-
m
a
x
n
o
r
m
aliza
tio
n
,
in
o
u
r
p
r
e
-
p
r
o
c
ess
in
g
d
u
e
t
o
its
s
im
p
licity
an
d
ef
f
ec
tiv
e
n
ess
.
W
h
en
p
r
eser
v
in
g
th
e
r
elatio
n
s
h
ip
b
etwe
en
th
e
o
r
i
g
in
al
d
ataset
is
cr
u
cial,
th
is
m
eth
o
d
is
esp
ec
ially
h
elp
f
u
l.
3.
2
.
2
.
Sta
nd
a
rd
s
ca
ler
Stan
d
ar
d
s
ca
ler
,
wh
ich
im
p
le
m
en
ts
Z
-
s
co
r
e
n
o
r
m
aliza
tio
n
,
s
tan
d
ar
d
izes
ch
ar
ac
ter
is
tics
b
y
r
em
o
v
in
g
th
eir
m
ea
n
f
r
o
m
ea
ch
v
alu
e
a
n
d
d
iv
id
in
g
th
e
o
u
tc
o
m
e
b
y
th
e
attr
ib
u
te’
s
s
tan
d
ar
d
d
ev
iati
o
n
s
,
p
r
o
d
u
cin
g
a
d
is
tr
ib
u
tio
n
with
a
m
ea
n
o
f
ze
r
o
an
d
a
v
ar
ia
n
ce
o
f
o
n
e
u
n
it
[
3
4
]
.
L
et
̅
b
e
th
e
m
ea
n
o
f
th
e
x
v
ar
iab
le,
an
d
(
2
)
tr
an
s
f
o
r
m
s
(
s
ca
les)
a
v
alu
e
in
to
̅
.
̅
=
−
̅
s
(
2
)
T
h
e
tr
an
s
latio
n
al
wo
r
d
in
th
is
ex
am
p
le
is
th
e
attr
ib
u
te’
s
s
am
p
le
m
ea
n
,
an
d
th
e
s
tan
d
ar
d
d
ev
iatio
n
s
er
v
es
as
th
e
s
ca
lin
g
f
ac
to
r
.
T
h
is
tec
h
n
iq
u
e
h
as
t
h
e
ad
v
an
tag
e
o
f
tr
an
s
f
o
r
m
in
g
b
o
th
p
o
s
itiv
e
an
d
n
eg
ativ
e
-
v
alu
e
d
q
u
alities
in
to
a
r
elativ
ely
co
m
p
ar
ab
le
d
is
tr
ib
u
tio
n
.
Ho
wev
er
,
wh
en
co
m
p
ar
ed
to
an
attr
ib
u
te
with
o
u
t o
u
tlier
s
,
th
e
f
in
al
d
is
tr
ib
u
tio
n
o
f
i
n
lier
s
is
ex
ce
s
s
iv
el
y
n
ar
r
o
w
wh
e
n
o
u
tlier
s
ar
e
p
r
e
s
en
t [
3
5
]
.
Stan
d
ar
d
s
ca
ler
is
u
s
ed
in
th
is
s
y
s
tem
to
r
esize
th
e
v
alu
e
d
is
tr
ib
u
tio
n
s
o
th
at
th
e
m
ea
n
o
f
th
e
o
b
s
er
v
e
d
d
ata
is
0
an
d
th
e
s
tan
d
ar
d
d
e
v
iatio
n
is
1
.
3.
3
.
Da
t
a
s
et
s
pli
t
t
ing
Data
s
et
s
p
li
ttin
g
is
a
s
tr
ateg
y
th
at
is
wid
ely
r
eg
ar
d
ed
as
ess
en
tial
f
o
r
r
em
o
v
i
n
g
o
r
r
ed
u
ci
n
g
b
ias
in
tr
ain
in
g
d
ata
in
d
ee
p
lear
n
in
g
m
o
d
els.
Data
s
cien
tis
ts
an
d
a
n
aly
s
ts
alwa
y
s
u
s
e
th
is
m
eth
o
d
to
k
ee
p
m
ac
h
in
e
lear
n
in
g
tech
n
iq
u
es
f
r
o
m
o
v
er
f
itti
n
g
an
d
u
n
d
er
p
er
f
o
r
m
i
n
g
o
n
r
ea
l
test
d
ata
[
3
6
]
.
L
ar
g
e
d
a
tasets
ar
e
ty
p
ically
d
iv
id
ed
in
to
s
ev
er
al
well
-
d
e
f
i
n
ed
s
u
b
g
r
o
u
p
s
b
y
d
ata
s
cien
t
is
ts
an
d
an
aly
s
ts
,
wh
o
th
e
n
u
s
e
th
ese
s
u
b
s
ets
to
tr
ain
d
if
f
e
r
en
t
p
ar
am
eter
s
.
T
h
e
g
o
al
o
f
th
is
s
tu
d
y
is
t
o
d
eter
m
in
e
wh
ich
m
ac
h
in
e
lear
n
in
g
s
y
s
tem
p
ar
am
eter
s
b
est
f
it
th
e
tr
ain
in
g
d
ata
b
y
c
o
n
s
id
er
in
g
t
h
e
s
ig
n
if
ica
n
t
im
p
a
ct
o
f
s
p
litt
in
g
a
d
ataset
in
to
m
u
ltip
le
tr
ain
s
ets
an
d
test
s
ets
[
3
7
]
.
I
n
o
r
d
er
to
ass
ess
th
e
p
r
ed
ictiv
e
ab
ilit
ies
o
f
cl
ass
if
icatio
n
m
o
d
els,
a
clea
n
d
ataset
m
u
s
t
b
e
u
s
ed
f
o
r
test
in
g
.
As
a
r
esu
lt,
th
e
o
r
ig
in
al
d
ataset
is
d
iv
id
ed
in
to
two
s
u
b
s
ets:
th
e
te
s
t
d
ataset
co
m
p
r
is
es
3
0
%
o
f
th
e
to
tal
o
b
s
er
v
atio
n
s
,
an
d
th
e
tr
a
in
in
g
d
ataset
co
m
p
r
is
es
7
0
%
o
f
th
e
to
tal
o
b
s
er
v
atio
n
s
in
th
e
o
r
ig
in
al
d
ataset.
T
h
e
test
d
ataset
is
k
ep
t
clea
n
s
o
th
at
m
o
d
el
d
etec
tio
n
m
ay
b
e
m
ad
e
o
n
it,
wh
ile
th
e
tr
ain
in
g
d
ataset
is
u
tili
ze
d
to
tr
ain
th
e
m
o
d
el
an
d
f
in
e
-
tu
n
e
p
ar
am
eter
s
.
Fin
d
in
g
a
b
ala
n
ce
b
etwe
en
a
s
u
itab
ly
lar
g
e
tr
ain
in
g
s
et
an
d
an
eq
u
ally
s
izab
le
test
in
g
s
et
was
th
e
m
ajo
r
cr
iter
io
n
th
at
g
u
id
e
d
o
u
r
d
ataset
s
p
litt
in
g
wh
ich
o
f
f
er
in
g
a
s
o
lid
ass
es
s
m
en
t
o
f
t
h
e
g
en
er
ality
o
f
th
e
m
o
d
el.
B
y
s
ettin
g
asid
e
7
0
%
o
f
th
e
d
ataset
f
o
r
tr
ain
in
g
,
a
llo
win
g
th
e
m
o
d
el
to
b
ec
o
m
e
f
am
iliar
with
an
d
ad
ju
s
t
to
th
e
u
n
d
er
ly
in
g
p
atte
r
n
s
in
th
e
d
ata.
I
n
ad
d
itio
n
,
s
ettin
g
asid
e
3
0
%
f
o
r
test
in
g
g
u
ar
an
tees
a
s
izab
le
co
llectio
n
o
f
u
n
k
n
o
wn
ca
s
es
f
o
r
ass
ess
in
g
th
e
m
o
d
el'
s
ef
f
ec
tiv
en
ess
,
ac
h
iev
in
g
a
b
ala
n
ce
b
etwe
en
m
o
d
el
lear
n
in
g
an
d
ass
ess
m
en
t.
T
h
e
m
o
d
el
is
les
s
lik
ely
to
o
v
er
f
it
s
in
ce
it
h
as
en
o
u
g
h
d
ata
to
co
m
p
r
eh
en
d
u
n
d
er
ly
i
n
g
p
atter
n
s
with
o
u
t
l
ea
r
n
in
g
n
o
is
e,
th
a
n
k
s
to
th
e
b
ig
g
er
p
ar
t
(
7
0
%)
th
at
is
d
ev
o
ted
to
t
r
ain
in
g
.
W
e
aim
to
im
p
r
o
v
e
th
e
tr
an
s
p
a
r
e
n
cy
an
d
c
r
ed
ib
ilit
y
o
f
o
u
r
r
e
s
u
lts
in
th
e
f
ield
o
f
ag
g
r
ess
iv
e
d
r
iv
er
b
eh
a
v
io
r
id
en
tific
atio
n
b
y
u
s
in
g
th
is
m
e
th
o
d
.
3.
4
.
F
e
a
t
ure
re
lev
a
nce
a
s
s
es
s
m
ent
m
et
ho
ds
A
p
r
ep
r
o
ce
s
s
in
g
tech
n
iq
u
e
th
at
d
eter
m
in
es
ess
en
tial
attr
ib
u
tes
o
f
a
p
r
o
b
lem
is
f
ea
tu
r
e
s
elec
tio
n
.
R
ed
u
cin
g
th
e
n
u
m
b
er
o
f
f
ea
t
u
r
es,
wh
ich
m
ea
n
s
th
e
n
u
m
b
e
r
o
f
co
lu
m
n
s
in
a
d
ataset
is
th
e
p
r
im
a
r
y
m
eth
o
d
u
s
ed
to
d
o
it.
T
h
e
m
o
d
el’
s
ac
c
u
r
ac
y
r
ate
an
d
in
f
er
e
n
ce
q
u
alit
y
in
cr
ea
s
e
as
th
e
n
u
m
b
er
o
f
f
e
atu
r
es
is
d
ec
r
ea
s
ed
with
o
u
t
co
m
p
r
o
m
is
in
g
th
e
q
u
ality
o
f
t
h
e
d
ataset,
wh
ile
lear
n
in
g
tim
e
an
d
av
ailab
le
s
p
ac
e
ar
e
d
ec
r
ea
s
ed
.
T
o
g
iv
e
th
ese
a
d
v
an
tag
es,
m
an
y
f
ea
tu
r
e
s
elec
tio
n
alg
o
r
ith
m
s
ar
e
av
ailab
le.
T
h
r
ee
m
eth
o
d
s
w
er
e
em
p
lo
y
ed
in
th
e
s
u
g
g
ested
m
o
d
el:
p
r
in
cip
al
c
o
m
p
o
n
en
t
an
aly
s
is
(
PC
A)
,
s
in
g
u
lar
v
alu
e
d
ec
o
m
p
o
s
itio
n
(
SVD)
,
an
d
m
u
tu
al
in
f
o
r
m
atio
n
(
MI
)
.
I
n
t
h
is
s
ec
ti
o
n
,
m
o
r
e
d
etails ab
o
u
t th
ese
tech
n
iq
u
es will b
e
ex
p
lain
ed
.
3.
4
.
1.
Usi
ng
princip
a
l c
o
m
po
nent
a
na
ly
s
is
t
o
s
elec
t
f
ea
t
ures (
1
s
t
t
ec
hn
iqu
e)
Usi
n
g
PC
A
to
s
elec
t
f
ea
tu
r
es
(
1
s
t
tech
n
iq
u
e)
: T
h
e
f
i
r
s
t
tech
n
iq
u
e
u
s
ed
i
n
th
is
s
y
s
tem
is
PC
A,
PC
A
is
a
tr
an
s
f
o
r
m
atio
n
ap
p
r
o
ac
h
th
a
t
r
ed
u
ce
s
th
e
s
ize
o
f
a
d
ataset
b
y
tr
a
n
s
f
o
r
m
in
g
it
in
to
f
ewe
r
ass
o
ciate
d
v
ar
iab
les
[
3
8
]
.
PC
A
is
a
d
ec
o
m
p
o
s
itio
n
o
f
a
c
o
lu
m
n
-
m
ea
n
-
ce
n
ter
ed
d
a
ta
m
atr
ix
X
o
f
s
ize
N×
K,
wh
e
r
e
N
an
d
K
ar
e
t
h
e
n
u
m
b
er
o
f
s
am
p
les an
d
f
ea
tu
r
es,
r
esp
ec
tiv
ely
.
=
+
(
3
)
T
is
a
s
co
r
i
n
g
m
atr
ix
o
f
s
ize
N×
A
co
n
n
ec
ted
to
th
e
m
atr
ix
X
p
r
o
jectio
n
s
in
to
an
A
-
d
im
e
n
s
io
n
al
s
p
ac
e,
P
is
a
lo
ad
in
g
m
atr
ix
o
f
s
ize
K×
A
r
e
lated
to
th
e
f
ea
tu
r
e
p
r
o
jectio
n
s
in
to
an
A
-
d
im
en
s
io
n
al
s
p
ac
e
(
with
=I
)
,
a
n
d
E
is
a
r
esid
u
al
m
atr
ix
o
f
s
ize
N×
K
[
3
9
]
.
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.
13
,
No
.
4
,
Dec
em
b
er
2
0
2
4
:
4
8
8
3
-
4
8
9
4
4888
3
.
4
.
2
.
Usi
ng
s
ing
ula
r
v
a
lue d
ec
o
m
po
s
it
io
n
t
o
s
elec
t
f
e
a
t
ur
es (
2
nd
t
ec
hn
iq
ue)
T
h
e
s
ec
o
n
d
tech
n
iq
u
e
u
tili
ze
d
in
th
e
p
r
o
p
o
s
ed
s
y
s
tem
t
o
s
elec
t
th
e
b
est
f
ea
tu
r
es
th
at
m
ak
e
th
e
ac
cu
r
ac
y
o
f
d
etec
tio
n
an
d
class
if
icatio
n
alm
o
s
t
id
en
tical
i
s
SVD,
as
P
C
A
b
u
t
m
o
s
t
s
p
ec
if
ically
,
th
e
in
itial
A
p
r
in
cip
al
co
m
p
o
n
e
n
ts
an
d
th
e
SVD
o
f
X
ar
e
u
s
ed
to
id
en
tify
th
e
A
-
d
im
en
s
io
n
al
s
p
ac
e.
W
h
en
we
d
en
o
te
X=
as
th
e
SVD
o
f
X
an
d
̂
,
̂
,
an
d
̂
as
th
e
m
atr
ices
co
n
tain
in
g
th
e
f
ir
s
t
A
co
lu
m
n
s
o
f
U,
S,
an
d
V,
r
esp
ec
tiv
ely
,
we
g
et:
=
̂
×
̂
(
4
)
=
̂
(
5
)
An
d
X=
is
n
am
ed
th
e
r
ec
o
n
s
tr
u
cted
d
ata
m
atr
i
x
[
4
0
]
.
3
.
4
.
3
.
Usi
ng
m
utua
l inf
o
rm
a
t
io
n
t
o
s
elec
t
f
ea
t
ures (
3
rd
t
ec
hn
iq
ue)
T
h
e
th
ir
d
tech
n
iq
u
e
u
s
ed
i
n
t
h
e
p
r
o
p
o
s
ed
m
o
d
el
to
in
cr
ea
s
e
th
e
ac
c
u
r
ac
y
an
d
d
ec
r
ea
s
e
th
e
tim
e
o
f
ex
ec
u
tio
n
is
MI
.
Stu
d
ies
o
n
M
I
d
atin
g
f
r
o
m
ea
r
ly
to
th
e
1
9
9
0
s
s
h
o
w
th
at
it
is
o
n
e
o
f
th
e
m
o
s
t
p
o
p
u
lar
f
ea
t
u
r
e
s
elec
tio
n
tech
n
iq
u
es
[
4
1
]
.
B
y
ca
lcu
latin
g
h
o
w
m
u
ch
d
ata
a
b
o
u
t
o
n
e
r
an
d
o
m
f
ea
tu
r
e
ca
n
b
e
o
b
tain
ed
f
r
o
m
th
e
o
th
er
,
MI
q
u
an
tifie
s
th
e
m
u
tu
ally
d
ep
en
d
en
t
r
elatio
n
s
h
ip
b
etwe
en
two
r
a
n
d
o
m
f
ea
tu
r
es.
I
t
is
th
er
ef
o
r
e
ass
o
ciate
d
with
th
e
en
tr
o
p
y
o
f
a
r
an
d
o
m
f
ea
t
u
r
e,
wh
ic
h
is
estab
lis
h
ed
b
y
th
e
q
u
a
n
tity
o
f
in
f
o
r
m
atio
n
in
clu
d
e
d
in
th
e
f
ea
tu
r
e
.
T
h
e
MI
b
etwe
e
n
two
d
is
cr
ete
r
an
d
o
m
v
ar
iab
l
es X
an
d
Y
is
d
ef
in
ed
t
o
b
e
as
(
6
)
[
4
2
]
.
(
;
)
=
∑
∑
(
,
)
2
(
(
,
)
(
)
(
)
)
∈
∈
(
6
)
T
h
r
e
e
s
e
p
a
r
a
t
e
f
e
at
u
r
e
s
e
l
ec
t
i
o
n
t
e
c
h
n
i
q
u
e
s
w
e
r
e
c
a
r
e
f
u
ll
y
s
e
le
c
t
e
d
i
n
t
h
e
s
t
u
d
y
t
o
h
a
n
d
l
e
t
h
e
p
a
r
t
i
c
u
l
a
r
d
i
f
f
i
c
u
l
t
i
es
i
n
v
o
l
v
e
d
i
n
i
d
e
n
t
i
f
y
i
n
g
A
DB
.
E
ac
h
o
f
t
h
es
e
s
t
r
a
t
e
g
i
es
h
as
u
n
i
q
u
e
b
e
n
e
f
i
t
s
t
h
a
t
c
o
m
p
l
e
m
e
n
t
t
h
e
r
e
s
e
a
r
c
h
o
b
j
e
c
t
i
v
es
a
n
d
i
n
c
r
e
a
s
e
t
h
e
s
t
a
b
i
l
it
y
a
n
d
e
f
f
i
c
a
c
y
o
f
t
h
e
s
u
g
g
e
s
t
e
d
m
u
l
t
i
-
s
t
a
g
e
s
y
s
t
em
.
I
t
w
a
s
d
e
c
i
d
e
d
t
o
c
o
m
b
i
n
e
P
C
A
,
S
V
D
,
a
n
d
M
I
in
o
r
d
e
r
t
o
t
h
e
h
a
r
n
e
s
s
a
d
v
a
n
t
a
g
e
s
o
f
e
a
c
h
t
e
c
h
n
i
q
u
e
.
T
h
e
d
r
i
v
i
n
g
b
e
h
a
v
i
o
r
d
a
t
a
s
e
t
h
a
s
h
i
g
h
d
i
m
e
n
s
i
o
n
a
l
it
y
,
a
n
d
b
e
c
a
u
s
e
P
C
A
e
f
f
e
c
t
i
v
el
y
l
o
w
e
r
s
d
i
m
e
n
s
i
o
n
a
l
it
y
a
n
d
p
r
e
s
e
r
v
e
s
i
m
p
o
r
t
a
n
t
i
n
f
o
r
m
a
t
i
o
n
,
i
t
is
a
g
o
o
d
f
it
f
o
r
o
u
r
s
t
u
d
y
s
i
n
c
e
o
u
r
o
b
j
e
ct
i
v
e
i
s
t
o
d
is
c
o
v
e
r
d
r
i
v
i
n
g
b
e
h
a
v
i
o
r
'
s
i
n
f
l
u
e
n
t
i
a
l
f
e
a
t
u
r
es
.
A
d
i
f
f
e
r
e
n
t
v
i
e
w
p
o
i
n
t
o
n
t
h
e
l
at
e
n
t
s
t
r
u
c
t
u
r
e
s
i
n
t
h
e
d
a
t
as
e
t
i
s
o
f
f
e
r
e
d
b
y
S
V
D
,
w
h
i
c
h
e
n
h
a
n
ce
s
PC
A
.
C
a
p
t
u
r
i
n
g
s
u
b
t
l
e
c
o
r
r
e
l
at
i
o
n
s
i
n
d
r
i
v
i
n
g
b
e
h
a
v
i
o
r
f
e
a
t
u
r
e
s
w
as
t
h
e
m
o
ti
v
a
t
i
n
g
f
o
r
c
e
b
e
h
i
n
d
i
t
s
u
s
a
g
e
M
I
w
a
s
s
e
l
e
ct
e
d
i
n
o
r
d
e
r
t
o
e
v
a
l
u
a
t
e
t
h
e
i
n
f
o
r
m
a
t
io
n
g
a
i
n
e
d
r
e
l
a
t
e
d
t
o
s
e
v
e
r
al
ch
a
r
a
c
t
e
r
is
t
i
cs
i
n
r
e
l
at
i
o
n
t
o
AD
B
.
T
h
e
i
n
t
r
i
c
a
c
y
o
f
d
r
i
v
i
n
g
b
e
h
a
v
i
o
r
d
a
t
a
s
e
ts
is
i
n
a
c
c
o
r
d
a
n
c
e
w
i
t
h
i
t
s
c
a
p
a
c
it
y
t
o
m
a
n
a
g
e
n
o
n
-
l
i
n
e
a
r
i
n
t
e
r
a
c
ti
o
n
s
.
3.
5
.
Co
nv
o
lutio
n neura
l net
wo
rk
t
o
cla
s
s
if
y
da
t
a
C
NN,
also
k
n
o
wn
as
C
o
n
v
Net,
is
a
k
in
d
o
f
ar
tific
ial
n
eu
r
al
n
etwo
r
k
(
ANN)
with
r
em
ar
k
ab
le
g
en
er
aliza
tio
n
ca
p
ab
ilit
ies
an
d
a
d
ee
p
f
ee
d
-
f
o
r
war
d
d
esig
n
[
4
3
]
.
I
t
ca
n
lear
n
h
ig
h
l
y
ab
s
tr
ac
ted
f
ea
tu
r
es
o
f
th
in
g
s
,
esp
ec
ially
s
p
atial
d
ata,
an
d
r
ec
o
g
n
ize
th
em
m
o
r
e
ef
f
ec
tiv
ely
th
an
o
th
e
r
n
etwo
r
k
s
with
FC
lay
er
s
[
4
4
]
‒
[
4
6
]
.
A
d
ee
p
C
NN
m
o
d
el
co
n
s
is
ts
o
f
a
lim
ited
n
u
m
b
er
o
f
p
r
o
ce
s
s
in
g
lay
er
s
th
at
ca
n
b
e
tr
ain
ed
at
d
if
f
er
en
t
lev
els
o
f
ab
s
tr
ac
tio
n
to
lear
n
d
if
f
er
en
t
f
ea
t
u
r
es
o
f
in
p
u
t
d
ata
(
lik
e
im
ag
es)
[
4
7
]
.
Hig
h
er
ab
s
tr
ac
tio
n
is
ac
h
iev
ed
b
y
th
e
d
ee
p
er
lay
er
s
in
lear
n
in
g
an
d
ex
tr
ac
tin
g
lo
w
-
lev
el
d
ata,
wh
ile
lo
wer
a
b
s
tr
ac
tio
n
is
ac
h
iev
ed
b
y
th
e
in
itiato
r
y
lev
els
[
4
8
]
.
F
ig
u
r
e
3
d
ep
icts
th
e
co
n
ce
p
t
u
al
f
o
r
m
o
f
th
e
p
r
o
p
o
s
ed
C
NN
-
d
en
s
e,
with
d
if
f
er
en
t
s
o
r
ts
o
f
lay
er
s
d
is
cu
s
s
ed
in
th
e
f
o
llo
win
g
s
ec
tio
n
.
−
C
o
n
v
o
lu
tio
n
lay
e
r
:
th
e
co
n
v
o
l
u
tio
n
al
lay
er
is
th
e
m
o
s
t
cr
u
ci
al
p
ar
t
o
f
an
y
C
NN
ar
ch
itectu
r
e.
T
o
cr
ea
te
an
o
u
tp
u
t
f
ea
tu
r
e
m
ap
,
it
c
o
n
s
is
ts
o
f
a
s
et
o
f
c
o
n
v
o
lu
tio
n
a
l
k
er
n
els,
s
o
m
etim
es
r
ef
er
r
e
d
to
as
f
ilter
s
,
co
n
v
o
l
v
ed
with
th
e
i
n
p
u
t im
a
g
e
(
N
-
d
im
en
s
io
n
al
m
et
r
ics)
[
4
9
]
,
[
5
0
]
.
−
P
o
o
l
i
n
g
l
a
y
e
r
:
la
y
e
r
s
s
u
b
-
s
a
m
p
l
e
f
e
a
t
u
r
e
m
a
p
s
p
r
o
d
u
c
e
d
a
f
te
r
c
o
n
v
o
l
u
t
i
o
n
o
p
e
r
a
t
i
o
n
s
,
p
r
es
e
r
v
i
n
g
d
o
m
i
n
a
n
t
f
e
a
t
u
r
e
s
i
n
e
a
c
h
p
o
o
l
s
t
e
p
.
P
o
o
l
i
n
g
o
p
e
r
a
t
i
o
n
s
s
p
e
ci
f
y
t
h
e
p
o
o
l
e
d
r
e
g
i
o
n
s
i
z
e
a
n
d
s
t
r
i
d
e
,
li
k
e
c
o
n
v
o
l
u
t
i
o
n
.
D
i
f
f
e
r
e
n
t
t
e
c
h
n
i
q
u
es
l
i
k
e
m
a
x
p
o
o
l
i
n
g
,
m
i
n
p
o
o
l
i
n
g
,
a
v
e
r
a
g
e
p
o
o
l
i
n
g
,
g
a
t
e
d
p
o
o
l
i
n
g
,
a
n
d
t
r
e
e
p
o
o
l
i
n
g
a
r
e
u
s
e
d
i
n
d
i
f
f
e
r
e
n
t
l
a
y
e
r
s
,
wi
t
h
m
a
x
p
o
o
l
i
n
g
b
e
i
n
g
t
h
e
m
o
s
t
p
o
p
u
l
a
r
a
n
d
c
o
m
m
o
n
l
y
u
s
e
d
t
e
c
h
n
iq
u
e
[
5
1
]
‒
[
5
3
]
.
−
L
ea
k
y
R
eL
U:
th
is
ac
tiv
atio
n
f
u
n
ctio
n
,
i
n
co
n
tr
ast
to
R
eL
U
,
d
o
wn
s
ca
les
th
e
n
eg
ativ
e
in
p
u
ts
r
ath
er
th
an
to
tally
ig
n
o
r
i
n
g
th
em
.
T
h
e
Dy
in
g
R
eL
U
p
r
o
b
lem
is
r
es
o
lv
ed
b
y
u
s
in
g
leak
y
R
eL
U.
leak
y
R
eL
U
is
r
ep
r
esen
ted
m
ath
e
m
atica
lly
as
(
7
)
[
5
4
]
:
(
)
=
{
>
0
≤
0
(
7
)
W
h
er
e
m
i
s
a
cons
t
an
t
,
al
s
o
know
n
as
t
he
l
e
ak
f
a
ct
o
r
,
and
i
s
of
t
en
se
t
t
o
a
l
ow
num
be
r
(
e.
g.,
0.001)
.
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
A
d
etec
tio
n
mo
d
el
o
f a
g
g
r
ess
i
ve
d
r
ivin
g
b
eh
a
vio
r
b
a
s
ed
o
n
h
yb
r
id
d
ee
p
le
a
r
n
in
g
(
N
o
o
r
W
a
lid
K
h
a
lid
)
4889
−
Den
s
e
:
th
is
lay
er
o
f
a
s
tan
d
a
r
d
DNN
is
wh
at
it
is
ca
lled
.
I
t
i
s
th
e
m
o
s
t
o
f
ten
u
s
ed
a
n
d
co
m
m
o
n
lay
er
.
T
h
e
f
o
llo
win
g
p
r
o
ce
s
s
is
ca
r
r
ied
o
u
t
o
n
th
e
in
p
u
t
b
y
th
e
d
en
s
e
lay
er
,
wh
ich
th
en
r
etu
r
n
s
t
h
e
o
u
tco
m
e.
T
h
e
f
o
r
m
u
latio
n
o
f
t
h
is
lay
er
is
(
8
)
[
5
5
]
:
=
(
(
,
)
+
)
(
8
)
−
Flatten
:
th
e
o
u
tp
u
t
o
f
th
e
p
o
o
lin
g
lay
er
will
b
e
a
m
atr
ix
,
wh
ich
th
e
n
eu
r
al
n
etwo
r
k
ca
n
n
o
t
r
ec
eiv
e.
T
h
e
n
×n
m
atr
ix
f
r
o
m
th
e
p
o
o
lin
g
l
ay
er
is
co
n
v
er
ted
in
to
n
2
×1
m
atr
ix
b
y
th
e
f
latten
in
g
lay
e
r
s
o
th
at
it
m
ay
b
e
f
ed
in
to
th
e
n
eu
r
al
n
etwo
r
k
[
5
6
]
.
−
F
u
l
l
y
c
o
n
n
e
c
t
e
d
l
a
y
e
r
s
:
i
n
a
C
N
N
m
o
d
e
l
,
o
n
e
o
r
m
o
r
e
f
u
l
l
y
c
o
n
n
e
c
t
e
d
l
a
y
e
r
s
a
r
e
o
f
t
e
n
i
n
c
l
u
d
e
d
j
u
s
t
b
e
f
o
r
e
t
h
e
c
l
a
s
s
i
f
ic
a
t
i
o
n
o
u
t
p
u
t
.
S
i
m
i
l
a
r
t
o
n
e
u
r
a
l
n
e
t
w
o
r
k
l
a
y
e
r
t
o
p
o
l
o
g
i
e
s
,
n
e
u
r
o
n
s
b
e
t
w
e
e
n
n
e
i
g
h
b
o
r
i
n
g
l
a
y
e
r
s
a
r
e
f
u
l
l
y
c
o
n
n
e
c
t
e
d
,
a
n
d
a
c
o
m
p
l
e
t
e
l
y
c
o
n
n
e
c
t
e
d
l
a
y
e
r
c
o
n
s
i
s
t
s
o
f
a
f
i
x
e
d
n
u
m
b
e
r
o
f
d
i
s
c
o
n
n
e
c
t
e
d
n
e
u
r
o
n
s
[
5
7
]
,
[
5
8
]
.
Fig
u
r
e
3
.
Ar
c
h
itectu
r
e
o
f
th
e
p
r
o
p
o
s
ed
C
NN
-
Den
s
e
m
o
d
el
3.
5.
1
.
T
h
e
p
r
o
p
o
s
e
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
-
d
e
n
s
e
m
o
d
e
l
f
o
r
d
r
i
v
e
r
b
e
h
a
v
i
o
r
d
e
t
e
c
t
i
o
n
a
n
d
c
l
a
s
s
i
f
i
c
a
t
i
o
n
T
h
e
p
r
o
p
o
s
e
d
C
N
N
-
D
e
n
s
e
M
o
d
e
l
f
o
r
d
r
i
v
e
r
b
e
h
a
v
i
o
r
d
e
t
e
c
t
i
o
n
a
n
d
c
l
a
s
s
i
f
i
c
a
t
i
o
n
:
T
h
e
p
r
o
p
o
s
e
d
C
N
N
-
D
e
n
s
e
m
o
d
e
l
f
o
r
ADB
i
s
e
x
p
l
a
i
n
e
d
i
n
t
h
i
s
s
e
c
t
i
o
n
.
T
h
e
p
r
o
p
o
s
e
d
C
N
N
m
o
d
e
l
i
s
u
t
i
l
i
z
e
d
t
o
c
l
a
s
s
i
f
y
d
a
t
a
i
m
m
e
d
i
a
t
e
l
y
a
f
t
e
r
t
h
e
d
a
t
a
s
e
t
i
s
l
o
a
d
e
d
,
p
r
o
c
e
s
s
e
d
,
a
n
d
s
p
l
i
t
i
n
t
h
i
s
t
e
c
h
n
i
q
u
e
.
T
h
e
s
u
g
g
e
s
t
e
d
C
N
N
-
d
e
n
s
e
m
o
d
e
l
h
a
s
2
6
l
a
y
e
r
s
,
w
h
i
c
h
a
r
e
a
s
f
o
l
l
o
w
s
:
i
)
C
N
N
w
i
t
h
8
l
a
y
e
r
s
,
i
i
)
l
e
a
k
y
R
e
L
U
w
i
t
h
7
l
a
y
e
r
s
,
i
i
i
)
M
a
x
P
o
o
l
i
n
g
w
i
t
h
7
l
a
y
e
r
s
,
i
v
)
1
l
a
y
e
r
s
h
o
u
l
d
b
e
f
l
a
t
t
e
n
e
d
,
a
n
d
v
)
d
e
n
s
e
i
s
3
l
a
y
e
r
s
.
T
a
b
l
e
2
g
o
e
s
i
n
t
o
m
u
c
h
d
e
t
a
i
l
a
b
o
u
t
t
h
e
s
e
l
a
y
e
r
s
.
T
ab
le
2
.
T
h
e
p
r
o
p
o
s
ed
h
y
p
e
r
C
NN
-
d
en
s
e
lay
er
s
NO.
La
y
e
r
t
y
p
e
F
i
l
t
e
r
s
S
i
z
e
/
S
t
r
i
d
e
A
c
t
i
v
a
t
i
o
n
f
u
n
c
t
i
o
n
#
P
a
r
a
m
1
C
o
n
v
o
l
u
t
i
o
n
a
l
16
3
/
1
ــ
64
3
M
a
x
P
o
o
l
i
n
g
ــ
2
/
2
ــ
0
3
Le
a
k
y
R
e
LU
ــ
ــ
ــ
0
4
C
o
n
v
o
l
u
t
i
o
n
a
l
32
3
/
1
ــ
1
5
6
8
5
M
a
x
P
o
o
l
i
n
g
ــ
2
/
1
ــ
0
6
Le
a
k
y
R
e
LU
ــ
ــ
ــ
0
7
C
o
n
v
o
l
u
t
i
o
n
a
l
64
3
/
1
ــ
6
2
0
8
8
M
a
x
P
o
o
l
i
n
g
ــ
2
/
1
ــ
0
9
Le
a
k
y
R
e
LU
ــ
ــ
ــ
0
10
C
o
n
v
o
l
u
t
i
o
n
a
l
64
3
/
1
ــ
1
2
3
5
2
11
M
a
x
P
o
o
l
i
n
g
ــ
2
/
1
ــ
0
12
Le
a
k
y
R
e
LU
ــ
ــ
ــ
0
13
D
e
n
se
64
ــ
Li
n
e
a
r
4
1
6
0
14
C
o
n
v
o
l
u
t
i
o
n
a
l
32
3
/
1
ــ
6
1
7
6
15
M
a
x
P
o
o
l
i
n
g
ــ
2
/
1
ــ
0
16
Le
a
k
y
R
e
LU
ــ
ــ
ــ
0
17
C
o
n
v
o
l
u
t
i
o
n
a
l
32
3
/
1
ــ
3
1
0
4
18
M
a
x
P
o
o
l
i
n
g
ــ
2
/
2
ــ
0
19
Le
a
k
y
R
e
LU
ــ
ــ
ــ
0
20
D
e
n
se
32
ــ
Li
n
e
a
r
1
0
5
6
21
C
o
n
v
o
l
u
t
i
o
n
a
l
16
3
/
1
ــ
1
5
5
2
22
M
a
x
P
o
o
l
i
n
g
ــ
2
/
2
ــ
0
23
Le
a
k
y
R
e
LU
ــ
ــ
ــ
0
24
C
o
n
v
o
l
u
t
i
o
n
a
l
45
3
/
1
ــ
2
2
0
5
25
F
l
a
t
t
e
n
ــ
ــ
ــ
0
26
D
e
n
se
32
ــ
S
o
f
t
ma
x
1
3
8
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.
13
,
No
.
4
,
Dec
em
b
er
2
0
2
4
:
4
8
8
3
-
4
8
9
4
4890
4.
RE
SU
L
T
AND
D
I
SCUS
I
O
N
I
n
th
is
s
ec
tio
n
,
we
p
r
esen
t
o
u
r
r
esear
ch
f
in
d
in
g
s
an
d
p
r
o
v
id
e
a
th
o
r
o
u
g
h
an
aly
s
is
an
d
in
te
r
p
r
etatio
n
o
f
th
em
in
th
e
co
n
tex
t
o
f
o
u
r
s
tu
d
y
o
b
jectiv
es.
W
e
h
av
e
d
iv
id
ed
th
is
s
ec
tio
n
in
to
s
ev
er
al
s
u
b
s
ec
tio
n
s
to
m
ak
e
s
u
r
e
th
e
p
r
esen
tatio
n
is
well
-
o
r
g
an
ized
.
T
wo
m
eth
o
d
o
lo
g
ies we
r
e
u
s
ed
in
th
e
p
r
o
p
o
s
ed
m
o
d
el
as f
o
llo
ws
.
4
.
1
.
Cla
s
s
if
y
da
t
a
us
ing
hy
per
CNN
-
d
ens
e
wit
ho
ut
f
ea
t
ure
s
elec
t
io
n “
1
st
m
et
ho
do
l
o
g
y
”
I
n
th
is
m
et
h
o
d
o
lo
g
y
,
th
e
d
ata
s
et
is
f
ir
s
t
p
r
o
ce
s
s
ed
u
s
in
g
t
wo
p
r
e
-
p
r
o
ce
s
s
in
g
tech
n
iq
u
es,
th
en
th
e
d
ata
is
s
ep
ar
ated
in
to
two
g
r
o
u
p
s
,
th
e
f
ir
s
t
is
u
s
ed
to
tr
ai
n
th
e
p
r
o
p
o
s
ed
m
o
d
el
an
d
th
e
o
th
er
is
u
s
ed
f
o
r
test
in
g
.
T
h
e
d
ata
is
en
ter
ed
a
s
is
to
th
e
class
if
icatio
n
s
tag
e
an
d
th
e
r
esu
lts
o
f
th
is
s
tag
e
u
s
in
g
ev
alu
atio
n
m
etr
ics [
5
9
]
,
[
6
0
]
ar
e
s
h
o
wn
in
T
ab
l
es
3
an
d
Fig
u
r
e
4
.
T
ab
le
3
.
T
h
e
r
esu
lts
o
f
p
r
o
p
o
s
ed
C
NN
-
d
en
s
e
with
o
u
t f
ea
tu
r
e
s
elec
tio
n
Te
c
h
n
i
q
u
e
A
c
c
u
r
a
c
y
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
f
-
mea
su
r
e
Ti
me
i
n
s
e
c
.
C
N
N
-
D
e
n
se
9
5
.
2
%
9
5
%
9
4
.
7
%
9
4
.
8
%
41
Fig
u
r
e
4
.
C
h
ar
t
o
f
r
esu
lts
o
f
p
r
o
p
o
s
ed
C
NN
-
d
en
s
e
with
o
u
t
f
e
atu
r
e
s
elec
tio
n
4
.
2
.
Cla
s
s
if
y
da
t
a
us
ing
hy
per
CNN
-
d
ens
e
u
s
ing
f
ea
t
ure
s
elec
t
io
n “
2
nd
m
et
ho
do
lo
g
y
”
A
f
ea
tu
r
e
s
elec
tio
n
is
m
er
ely
ch
o
o
s
in
g
o
r
elim
in
atin
g
s
p
ec
if
ic
f
ea
tu
r
es
with
o
u
t
alter
in
g
th
em
in
an
y
m
an
n
er
.
Dim
en
s
io
n
ality
r
ed
u
ctio
n
is
th
e
p
r
o
ce
s
s
o
f
r
ed
u
cin
g
th
e
d
im
e
n
s
io
n
ality
o
f
f
ea
tu
r
es.
T
h
e
s
et
o
f
f
ea
tu
r
es
p
r
o
d
u
c
ed
b
y
f
ea
tu
r
e
s
elec
tio
n
,
o
n
t
h
e
o
th
er
h
an
d
,
m
u
s
t
b
e
a
s
u
b
s
et
o
f
th
e
o
r
ig
in
al
s
et
o
f
f
ea
tu
r
es.
T
h
e
s
et
p
r
o
d
u
ce
d
b
y
d
im
en
s
io
n
alit
y
r
ed
u
ctio
n
d
o
es n
o
t h
av
e
to
b
e
(
f
o
r
ex
am
p
le,
PC
A
d
ec
r
ea
s
e
s
d
im
en
s
io
n
ality
b
y
g
en
er
atin
g
n
ew
s
y
n
th
etic
f
ea
t
u
r
es
b
y
lin
ea
r
ly
m
ix
in
g
th
e
e
x
is
tin
g
f
ea
tu
r
es
a
n
d
r
em
o
v
in
g
th
e
less
s
ig
n
if
ican
t
o
n
es).
I
n
th
is
s
en
s
e,
f
ea
tu
r
e
s
e
lectio
n
is
a
s
u
b
s
et
o
f
d
im
en
s
io
n
ality
r
ed
u
ctio
n
.
Featu
r
e
s
elec
tio
n
an
d
r
e
d
u
ctio
n
ap
p
r
o
ac
h
es we
r
e
em
p
lo
y
ed
in
th
is
s
tu
d
y
to
im
p
r
o
v
e
th
e
ef
f
ic
ien
cy
o
f
o
u
r
s
u
g
g
ested
h
y
p
er
C
NN
-
Den
s
e
m
o
d
el.
T
h
is
s
ec
tio
n
d
ig
s
in
to
h
o
w
v
a
r
io
u
s
s
tr
ateg
ies
af
f
ec
t
m
o
d
el
p
er
f
o
r
m
an
ce
an
d
co
m
p
u
tatio
n
al
co
m
p
lex
ity
.
T
h
e
em
p
h
asis
is
o
n
id
e
n
tify
in
g
th
e
m
o
s
t
u
s
ef
u
l
tr
aits
an
d
h
o
w
th
ey
co
n
tr
ib
u
te
t
o
im
p
r
o
v
e
d
p
r
ed
ictio
n
ac
c
u
r
ac
y
.
T
ab
le
4
a
n
d
Fig
u
r
e
5
d
is
p
lay
th
e
r
esu
lts
o
f
th
r
ee
f
ea
t
u
r
e
s
el
ec
tio
n
s
tr
ateg
ies
(
PC
A,
SVD,
an
d
MI
)
co
m
b
in
ed
with
th
e
p
r
o
p
o
s
ed
C
NN
-
d
en
s
e
m
o
d
el.
T
ab
le
4
.
T
h
e
r
esu
lts
o
f
p
r
o
p
o
s
ed
C
NN
-
Den
s
e
with
PC
A,
S
V
D,
an
d
MI
f
ea
tu
r
e
s
elec
tio
n
Te
c
h
n
i
q
u
e
A
c
c
u
r
a
c
y
(
%
)
P
r
e
c
i
s
i
o
n
(
%
)
R
e
c
a
l
l
(
%
)
F
-
mea
su
r
e
(
%
)
Ti
me
i
n
s
e
c
o
n
d
P
C
A
3
7
5
.
4
7
8
.
5
7
8
.
3
7
8
.
3
30
P
C
A
4
9
6
.
8
9
8
.
4
9
8
.
4
9
8
.
4
54
P
C
A
5
8
5
.
7
8
2
.
4
8
2
.
4
8
2
.
3
14
P
C
A
6
9
8
.
7
9
8
.
9
9
8
.
9
9
8
.
9
24
S
V
D
3
73
7
5
.
1
75
75
36
S
V
D
4
9
7
.
6
9
7
.
6
9
7
.
6
9
7
.
6
48
S
V
D
5
9
9
.
9
9
9
.
9
9
9
.
9
9
9
.
9
40
S
V
D
6
1
0
0
1
0
0
1
0
0
1
0
0
43
M
I
3
7
0
.
5
7
3
.
7
7
2
.
9
7
2
.
8
44
M
I
4
9
1
.
8
9
4
.
5
9
4
.
5
9
4
.
5
50
M
I
5
99
9
9
.
3
9
9
.
3
9
9
.
3
59
M
I
6
1
0
0
1
0
0
1
0
0
1
0
0
51
P
a
r
a
me
t
e
r
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
A
d
etec
tio
n
mo
d
el
o
f a
g
g
r
ess
i
ve
d
r
ivin
g
b
eh
a
vio
r
b
a
s
ed
o
n
h
yb
r
id
d
ee
p
le
a
r
n
in
g
(
N
o
o
r
W
a
lid
K
h
a
lid
)
4891
Fig
u
r
e
5
.
C
h
ar
t
o
f
s
u
g
g
ested
C
NN
-
d
en
s
e
r
esu
lts
with
f
ea
tu
r
e
s
elec
tio
n
T
h
e
s
u
g
g
ested
m
o
d
el
with
SVD6
an
d
MI
6
p
r
o
d
u
ce
d
th
e
b
est
r
esu
lts
,
ev
en
wh
en
u
tili
z
in
g
f
ea
tu
r
e
s
elec
tio
n
tech
n
iq
u
es
with
a
1
0
0
%
ac
cu
r
ac
y
r
ate
f
o
r
ag
g
r
ess
iv
e
d
r
iv
er
b
e
h
av
io
r
d
etec
tio
n
,
wh
ile
th
e
tim
e
f
o
r
SVD6
was
th
e
s
h
o
r
test
,
at
4
3
s
ec
o
n
d
s
.
Featu
r
e
s
elec
tio
n
is
u
s
ed
in
th
e
d
ee
p
lear
n
in
g
p
r
o
ce
s
s
to
im
p
r
o
v
e
ac
cu
r
ac
y
.
I
t
also
im
p
r
o
v
es
th
e
d
etec
tio
n
ca
p
ac
ity
o
f
t
h
e
alg
o
r
ith
m
s
b
y
id
en
tify
i
n
g
t
h
e
m
o
s
t
im
p
o
r
ta
n
t
v
ar
iab
les
an
d
r
em
o
v
in
g
th
e
r
ed
u
n
d
a
n
t
an
d
ir
r
elev
a
n
t
o
n
es.
T
h
is
is
wh
y
f
ea
tu
r
e
s
elec
tio
n
is
s
o
cr
u
cial.
T
h
e
f
o
llo
win
g
ar
e
th
r
ee
m
ajo
r
ad
v
an
tag
es o
f
f
ea
tu
r
e
s
elec
tio
n
:
−
R
e
d
u
c
e
s
o
v
e
r
-
f
i
t
t
i
n
g
:
l
e
s
s
d
u
p
l
i
c
a
t
e
d
d
a
t
a
i
m
p
l
i
e
s
f
e
w
e
r
o
p
p
o
r
t
u
n
i
t
i
e
s
t
o
m
a
k
e
c
o
n
c
l
u
s
i
o
n
s
b
a
s
e
d
o
n
n
o
i
s
e
.
−
I
m
p
r
o
v
es a
cc
u
r
ac
y
: le
s
s
m
is
lead
in
g
d
ata
im
p
lies
m
o
r
e
ac
cu
r
ate
m
o
d
elin
g
.
−
Sh
o
r
ten
s
tr
ain
in
g
tim
e:
less
d
a
ta
im
p
lies
f
aster
alg
o
r
ith
m
s
.
T
h
is
s
tu
d
y
in
clu
d
ed
th
r
ee
d
is
t
in
ct
f
ea
tu
r
e
s
elec
tio
n
s
tr
ateg
ie
s
th
at
wer
e
ca
r
ef
u
lly
ch
o
s
en
to
ad
d
r
ess
th
e
u
n
iq
u
e
c
h
allen
g
es
ass
o
ciate
d
with
class
if
y
in
g
ADB
.
T
h
e
d
is
tin
ct
ad
v
an
tag
es
o
f
ea
ch
o
f
th
ese
ap
p
r
o
ac
h
es
en
h
an
ce
th
e
g
o
als
o
f
t
h
e
r
e
s
ea
r
ch
wh
ile
s
tr
en
g
th
en
in
g
t
h
e
s
tab
ilit
y
an
d
e
f
f
ec
tiv
en
es
s
o
f
t
h
e
p
r
o
p
o
s
ed
m
u
lti
-
s
tag
e
s
y
s
tem
.
T
h
e
d
ec
is
io
n
was
m
ad
e
to
u
s
e
PC
A,
SVD,
an
d
MI
to
f
u
ll
y
u
tili
ze
t
h
e
b
e
n
ef
its
o
f
ea
c
h
m
eth
o
d
a
n
d
th
e
n
co
m
p
ar
e
t
h
e
r
esu
lts
an
d
d
eter
m
in
e
th
e
b
est.
T
h
e
Dr
iv
in
g
B
eh
av
i
o
r
d
ataset
is
h
ig
h
d
im
en
s
io
n
al,
an
d
s
in
ce
o
u
r
g
o
al
is
to
id
en
tify
th
e
in
f
lu
en
ti
al
asp
ec
ts
o
f
d
r
iv
i
n
g
b
e
h
av
io
r
,
PC
A
s
u
cc
ess
f
u
ll
y
lo
wer
s
d
im
en
s
io
n
ality
wh
ile
p
r
eser
v
in
g
r
ele
v
an
t
in
f
o
r
m
ati
o
n
,
m
ak
in
g
it
a
s
tr
o
n
g
f
it
f
o
r
o
u
r
s
tu
d
y
.
SVD
im
p
r
o
v
es
PC
A
b
y
p
r
o
v
id
in
g
an
alter
n
ativ
e
p
er
s
p
ec
tiv
e
o
n
th
e
laten
t
s
tr
u
ctu
r
es
in
th
e
d
ataset.
T
h
e
d
r
iv
i
n
g
f
o
r
ce
b
eh
in
d
its
u
s
e
was
th
e
a
b
ilit
y
to
id
en
tif
y
tin
y
co
r
r
elatio
n
s
in
f
ea
t
u
r
es
ass
o
ciate
d
with
d
r
iv
i
n
g
b
eh
av
io
r
.
MI
was
ch
o
s
en
to
ass
ess
th
e
k
n
o
wled
g
e
ac
q
u
ir
ed
o
n
m
a
n
y
tr
aits
ass
o
ciate
d
with
ADB
.
Dr
iv
in
g
b
e
h
av
io
r
d
atasets
ar
e
co
m
p
lex
b
ec
au
s
e
o
f
th
eir
ab
ilit
y
to
h
an
d
le
n
o
n
-
lin
ea
r
in
ter
ac
tio
n
s
.
T
h
e
d
r
awb
ac
k
s
o
f
t
h
ese
ap
p
r
o
ac
h
es
in
clu
d
e
lim
ited
in
ter
p
r
etab
ilit
y
o
f
th
e
m
ajo
r
co
m
p
o
n
e
n
t
in
ter
m
s
o
f
o
r
ig
in
al
f
ea
tu
r
es.
Fo
r
SVD,
th
e
d
ataset
was sen
s
itiv
e
to
n
o
is
e,
an
d
MI
r
eq
u
ir
ed
a
lo
t o
f
c
o
m
p
u
tatio
n
,
p
ar
ticu
lar
l
y
f
o
r
b
i
g
f
ea
tu
r
e
s
ets.
4.
3
.
Resul
t
s
c
o
m
pa
riso
n
W
h
en
co
m
p
ar
in
g
th
e
r
esu
lts
o
b
tain
ed
f
r
o
m
t
h
e
p
r
o
p
o
s
ed
h
y
p
er
C
NN
-
d
en
s
e
s
y
s
tem
with
th
e
r
esu
lts
o
f
p
r
e
v
io
u
s
s
tu
d
ies
th
at
wo
r
k
e
d
o
n
th
e
s
am
e
d
ataset
in
T
ab
l
e
5
an
d
Fig
u
r
e
6
,
we
n
o
tice
th
e
s
u
p
er
io
r
ity
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
in
all
ca
s
es,
ev
en
u
s
in
g
t
h
e
f
ir
s
t
m
eth
o
d
o
l
o
g
y
with
o
u
t
f
ea
tu
r
e
ex
tr
ac
tio
n
th
e
ac
cu
r
ac
y
r
esu
lt
was
9
5
.
2
%.
I
n
o
th
er
ca
s
es,
w
h
en
f
ea
t
u
r
e
e
x
tr
ac
tio
n
tech
n
iq
u
es
wer
e
u
s
ed
th
e
r
esu
lts
o
b
t
ain
ed
f
o
r
ac
c
u
r
ac
y
wer
e
1
0
0
%
with
SVD6
an
d
MI
6
as
th
e
b
est
ac
cu
r
ac
y
,
a
n
d
with
o
th
er
tech
n
iq
u
es
th
e
a
cc
u
r
ac
y
also
r
ea
ch
e
d
9
9
.
9
%
an
d
9
9
%
as
well,
an
d
t
h
e
r
est
o
f
th
e
r
esu
lts
wer
e
al
s
o
g
o
o
d
co
m
p
ar
ed
to
th
e
r
esu
lts
o
f
p
r
ev
io
u
s
s
tu
d
ies
[
2
7
]
,
[
2
8
]
th
at
g
av
e
d
etec
tio
n
ac
cu
r
ac
y
o
f
9
1
.
9
4
%
an
d
7
9
.
5
%
r
esp
ec
tiv
ely
wh
e
n
u
s
in
g
th
e
s
am
e
Dr
iv
in
g
b
eh
av
io
r
d
ataset.
I
n
th
ese
two
s
tu
d
ies
th
ey
d
id
n
’
t
u
s
e
f
ea
tu
r
e
s
elec
tio
n
tech
n
i
q
u
es
ca
u
s
e
o
f
th
is
o
u
r
d
etec
tio
n
ac
cu
r
ac
y
was
b
etter
b
y
u
s
in
g
th
r
ee
o
f
f
ea
r
u
r
e
s
elec
tio
n
tech
n
iq
u
es
(
PC
A,
SVD,
an
d
MI
)
.
I
n
ad
d
itio
n
,
tim
e
o
f
ex
ec
u
tio
n
f
o
r
o
u
r
p
r
o
p
o
s
ed
s
y
s
tem
was f
ew
ca
u
s
es o
f
th
ese
u
s
ed
ap
p
r
a
o
ch
.
T
ab
le
5
.
C
o
m
p
a
r
is
o
n
r
esu
lts
o
n
d
r
iv
in
g
b
e
h
av
io
r
d
ataset
R
e
f
e
r
e
n
c
e
A
c
c
u
r
a
c
y
(
%)
[
2
7
]
9
1
.
9
4
[
2
8
]
7
9
.
5
O
u
r
p
r
o
p
o
s
e
d
C
N
N
-
d
e
n
s
e
1
0
0
P
a
r
a
me
t
e
r
Te
c
h
n
i
q
u
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.
13
,
No
.
4
,
Dec
em
b
er
2
0
2
4
:
4
8
8
3
-
4
8
9
4
4892
Fig
u
r
e
6
.
C
o
m
p
a
r
is
o
n
r
esu
lts
with
r
elate
d
wo
r
k
s
wh
ich
u
s
ed
th
e
s
am
e
d
ataset
5.
CO
NCLU
SI
O
N
T
h
e
ac
cu
r
ate
d
etec
tio
n
o
f
AD
B
is
th
e
f
o
u
n
d
atio
n
f
o
r
ea
r
ly
a
n
d
ef
f
ec
tiv
e
war
n
in
g
o
r
ass
is
t
an
ce
to
th
e
d
r
iv
er
,
wh
ich
is
cr
itical
f
o
r
in
cr
ea
s
in
g
d
r
i
v
in
g
s
af
ety
.
I
n
th
is
s
tu
d
y
,
a
n
ADB
d
etec
tio
n
m
o
d
el
b
ased
o
n
h
y
p
er
-
d
ee
p
lear
n
in
g
C
NN
-
d
e
n
s
e
is
b
u
ilt
u
s
in
g
th
e
d
r
iv
in
g
b
eh
a
v
io
r
d
ataset;
a
p
r
o
p
o
s
ed
class
if
y
m
o
d
el
is
b
u
ilt;
f
ea
tu
r
e
s
elec
tio
n
tech
n
iq
u
es
ar
e
u
s
ed
;
an
d
th
e
m
o
d
el
is
tr
ain
ed
an
d
test
ed
u
s
in
g
th
e
d
r
iv
in
g
b
e
h
av
io
r
d
ataset
o
b
tain
ed
in
a
d
r
i
v
in
g
en
v
ir
o
n
m
en
t
th
at
is
r
ea
lis
tic.
R
esu
lts
in
d
icate
th
at
th
e
p
r
o
p
o
s
ed
d
ee
p
lear
n
in
g
m
o
d
el
ac
h
iev
es
g
r
ea
ter
ac
cu
r
a
cy
,
p
r
e
d
ictio
n
,
r
ec
all,
a
n
d
F1
-
m
ea
s
u
r
e
o
f
1
0
0
%
with
SVD6
in
4
3
s
ec
o
n
d
s
an
d
MI
6
in
5
1
s
ec
o
n
d
s
.
I
n
c
o
n
tr
ast,
th
e
p
r
o
p
o
s
ed
m
o
d
el
d
esig
n
ed
with
o
u
t
f
ea
t
u
r
e
s
elec
tio
n
ac
h
iev
e
d
9
5
.
2
%
ac
cu
r
ac
y
i
n
4
1
s
ec
o
n
d
s
,
wh
e
r
e
th
ese
r
esu
lts
wer
e
t
h
e
wo
r
est r
esu
lts
f
o
r
t
h
e
p
r
o
p
o
s
ed
s
y
s
tem
.
T
h
is
c
o
m
p
ar
is
o
n
r
esu
lt
in
d
icate
s
th
at
th
e
s
u
g
g
e
s
ted
m
o
d
el
with
f
ea
tu
r
e
s
elec
t
io
n
is
b
etter
s
u
ited
f
o
r
ac
cu
r
at
ely
d
etec
tin
g
ADB,
ev
en
with
a
lim
ited
p
ar
t
o
f
th
e
d
ataset.
I
n
ter
m
s
o
f
f
u
tu
r
e
wo
r
k
in
th
is
f
ield
,
we
s
h
o
u
ld
n
o
te
th
at
th
e
d
ataset
ca
n
b
e
en
h
an
ce
d
with
d
ata
th
at
ca
n
b
e
m
ea
s
u
r
ed
to
id
e
n
tify
em
o
tio
n
al,
en
v
ir
o
n
m
en
tal,
an
d
p
s
y
ch
o
lo
g
ical
co
m
p
o
n
en
ts
r
ath
er
th
an
ju
s
t
b
eh
av
io
r
al
f
ac
to
r
s
.
T
h
e
p
r
o
p
o
s
ed
ar
ch
itectu
r
e
en
ab
les
its
ad
ap
tatio
n
to
d
iv
er
s
e
d
atasets
an
d
s
ce
n
ar
io
s
,
m
ak
in
g
it
a
v
alu
ab
le
ass
et
f
o
r
ad
d
r
e
s
s
in
g
v
ar
io
u
s
ch
allen
g
es
in
tr
a
n
s
p
o
r
tatio
n
,
s
af
ety
,
an
d
u
r
b
an
p
lan
n
in
g
.
Fu
tu
r
e
a
p
p
licatio
n
s
ca
n
b
u
ild
o
n
th
is
r
esear
ch
’
s
f
o
u
n
d
atio
n
to
f
u
r
th
er
m
an
y
asp
ec
ts
o
f
in
tellig
en
t
s
y
s
tem
s
an
d
d
ee
p
en
o
u
r
u
n
d
er
s
tan
d
i
n
g
o
f
h
o
w
p
eo
p
le
b
eh
av
e
i
n
d
y
n
am
ic
co
n
t
ex
ts
,
s
u
ch
as
u
s
e
in
ex
p
an
d
th
e
m
o
d
el
’
s
u
s
e
b
e
y
o
n
d
a
g
g
r
ess
io
n
a
n
aly
s
is
o
f
d
r
i
v
in
g
b
eh
av
i
o
r
.
Ma
k
e
u
s
e
o
f
th
e
ar
ch
itectu
r
e
to
ca
teg
o
r
ize
an
d
c
o
m
p
r
e
h
en
d
d
if
f
er
en
t
d
r
iv
in
g
b
e
h
av
io
r
s
,
s
u
ch
as
f
o
llo
win
g
tr
a
f
f
ic
laws,
b
ein
g
d
ef
en
s
iv
e,
o
r
d
r
iv
in
g
wh
ile
d
is
tr
ac
ted
.
T
h
e
ca
p
ac
ity
o
f
th
e
m
o
d
el
to
id
e
n
tify
s
u
b
tle
d
r
iv
in
g
p
atter
n
s
ca
n
h
elp
im
p
r
o
v
e
th
e
way
s
elf
-
d
r
iv
in
g
ca
r
s
m
ak
e
d
e
cisi
o
n
s
in
in
tr
icate
tr
af
f
ic
s
itu
atio
n
s
.
RE
F
E
R
E
NC
E
S
[
1
]
S
.
A
r
u
m
u
g
a
m
a
n
d
R
.
B
h
a
r
g
a
v
i
,
“
A
s
u
r
v
e
y
o
n
d
r
i
v
i
n
g
b
e
h
a
v
i
o
r
a
n
a
l
y
s
i
s
i
n
u
sa
g
e
b
a
se
d
i
n
s
u
r
a
n
c
e
u
s
i
n
g
b
i
g
d
a
t
a
,
”
J
o
u
rn
a
l
o
f
B
i
g
D
a
t
a
,
v
o
l
.
6
,
n
o
.
1
,
D
e
c
.
2
0
1
9
,
d
o
i
:
1
0
.
1
1
8
6
/
s4
0
5
3
7
-
019
-
0
2
4
9
-
5.
[
2
]
J.
H
u
,
X
.
Zh
a
n
g
,
a
n
d
S
.
M
a
y
b
a
n
k
,
“
A
b
n
o
r
mal
d
r
i
v
i
n
g
d
e
t
e
c
t
i
o
n
w
i
t
h
n
o
r
ma
l
i
z
e
d
d
r
i
v
i
n
g
b
e
h
a
v
i
o
r
d
a
t
a
:
a
d
e
e
p
l
e
a
r
n
i
n
g
a
p
p
r
o
a
c
h
,
”
I
EEE
T
r
a
n
sa
c
t
i
o
n
s
o
n
Ve
h
i
c
u
l
a
r
T
e
c
h
n
o
l
o
g
y
,
v
o
l
.
6
9
,
n
o
.
7
,
p
p
.
6
9
4
3
–
6
9
5
1
,
Ju
l
.
2
0
2
0
,
d
o
i
:
1
0
.
1
1
0
9
/
TV
T
.
2
0
2
0
.
2
9
9
3
2
4
7
.
[
3
]
C
.
Z
h
a
n
g
,
R
.
L
i
,
W
.
K
i
m,
D
.
Y
o
o
n
,
a
n
d
P
.
P
a
t
r
a
s,
“
D
r
i
v
e
r
b
e
h
a
v
i
o
r
r
e
c
o
g
n
i
t
i
o
n
v
i
a
i
n
t
e
r
w
o
v
e
n
d
e
e
p
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
s
w
i
t
h
m
u
l
t
i
-
st
r
e
a
m
i
n
p
u
t
s
,
”
I
EEE
Ac
c
e
ss
,
v
o
l
.
8
,
p
p
.
1
9
1
1
3
8
–
1
9
1
1
5
1
,
2
0
2
0
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
0
.
3
0
3
2
3
4
4
.
[
4
]
E.
K
h
o
sr
a
v
i
,
A
.
M
.
A
.
H
e
mm
a
t
y
a
r
,
M
.
J.
S
i
a
v
o
s
h
a
n
i
,
a
n
d
B
.
M
o
s
h
i
r
i
,
“
S
a
f
e
d
e
e
p
d
r
i
v
i
n
g
b
e
h
a
v
i
o
r
d
e
t
e
c
t
i
o
n
(
S
3
D
)
,
”
I
EEE
Ac
c
e
ss
,
v
o
l
.
1
0
,
p
p
.
1
1
3
8
2
7
–
1
1
3
8
3
8
,
2
0
2
2
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
2
.
3
2
1
7
6
4
4
.
[
5
]
M
.
M
a
l
i
k
a
n
d
R
.
N
a
n
d
a
l
,
“
A
f
r
a
m
e
w
o
r
k
o
n
d
r
i
v
i
n
g
b
e
h
a
v
i
o
r
a
n
d
p
a
t
t
e
r
n
u
si
n
g
on
-
b
o
a
r
d
d
i
a
g
n
o
s
t
i
c
s
(
O
B
D
-
I
I
)
t
o
o
l
,
”
M
a
t
e
ri
a
l
s
T
o
d
a
y
:
Pro
c
e
e
d
i
n
g
s
,
v
o
l
.
8
0
,
p
p
.
3
7
6
2
–
3
7
6
8
,
2
0
2
3
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
m
a
t
p
r
.
2
0
2
1
.
0
7
.
3
7
6
.
[
6
]
C
.
K
a
t
r
a
k
a
z
a
s
,
E.
M
i
c
h
e
l
a
r
a
k
i
,
M
.
S
e
k
a
d
a
k
i
s
,
a
n
d
G
.
Y
a
n
n
i
s,
“
A
d
e
scri
p
t
i
v
e
a
n
a
l
y
si
s
o
f
t
h
e
e
f
f
e
c
t
o
f
t
h
e
C
O
V
I
D
-
1
9
p
a
n
d
e
mi
c
o
n
d
r
i
v
i
n
g
b
e
h
a
v
i
o
r
a
n
d
r
o
a
d
safe
t
y
,
”
T
ra
n
sp
o
r
t
a
t
i
o
n
Re
s
e
a
rc
h
I
n
t
e
r
d
i
sci
p
l
i
n
a
ry
Pe
rs
p
e
c
t
i
v
e
s
,
v
o
l
.
7
,
S
e
p
.
2
0
2
0
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
t
r
i
p
.
2
0
2
0
.
1
0
0
1
8
6
.
[
7
]
K
.
W
a
n
g
,
Q
.
X
u
e
,
Y
.
X
i
n
g
,
a
n
d
C
.
Li
,
“
I
mp
r
o
v
e
a
g
g
r
e
ss
i
v
e
d
r
i
v
e
r
r
e
c
o
g
n
i
t
i
o
n
u
s
i
n
g
c
o
l
l
i
si
o
n
s
u
r
r
o
g
a
t
e
m
e
a
s
u
r
e
me
n
t
a
n
d
i
mb
a
l
a
n
c
e
d
c
l
a
ss
b
o
o
st
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
En
v
i
r
o
n
m
e
n
t
a
l
R
e
se
a
rch
a
n
d
P
u
b
l
i
c
H
e
a
l
t
h
,
v
o
l
.
1
7
,
n
o
.
7
,
M
a
r
.
2
0
2
0
,
d
o
i
:
1
0
.
3
3
9
0
/
i
j
e
r
p
h
1
7
0
7
2
3
7
5
.
[
8
]
Y
.
M
a
,
Z
.
X
i
e
,
S
.
C
h
e
n
,
F
.
Q
i
a
o
,
a
n
d
Z.
Li
,
“
R
e
a
l
-
t
i
me
d
e
t
e
c
t
i
o
n
o
f
a
b
n
o
r
mal
d
r
i
v
i
n
g
b
e
h
a
v
i
o
r
b
a
se
d
o
n
l
o
n
g
s
h
o
r
t
-
t
e
r
m
mem
o
r
y
n
e
t
w
o
r
k
a
n
d
r
e
g
r
e
ss
i
o
n
r
e
si
d
u
a
l
s,”
T
ra
n
s
p
o
rt
a
t
i
o
n
R
e
se
a
rc
h
P
a
rt
C
:
Em
e
r
g
i
n
g
T
e
c
h
n
o
l
o
g
i
e
s
,
v
o
l
.
1
4
6
,
Ja
n
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
t
r
c
.
2
0
2
2
.
1
0
3
9
8
3
.
[
9
]
Y
.
Zh
a
n
g
,
Y
.
H
e
,
a
n
d
L.
Zh
a
n
g
,
“
R
e
c
o
g
n
i
t
i
o
n
m
e
t
h
o
d
o
f
a
b
n
o
r
mal
d
r
i
v
i
n
g
b
e
h
a
v
i
o
r
u
si
n
g
t
h
e
b
i
d
i
r
e
c
t
i
o
n
a
l
g
a
t
e
d
r
e
c
u
r
r
e
n
t
u
n
i
t
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
,
”
P
h
y
si
c
a
A:
S
t
a
t
i
s
t
i
c
a
l
Me
c
h
a
n
i
c
s
a
n
d
i
t
s
Ap
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
6
0
9
,
2
0
2
3
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
p
h
y
sa.
2
0
2
2
.
1
2
8
3
1
7
.
[
10
]
H
.
Z
h
u
,
R
.
X
i
a
o
,
J.
Z
h
a
n
g
,
J.
L
i
u
,
C
.
Li
,
a
n
d
L.
Y
a
n
g
,
“
A
d
r
i
v
i
n
g
b
e
h
a
v
i
o
r
r
i
s
k
c
l
a
ssi
f
i
c
a
t
i
o
n
f
r
a
mew
o
r
k
v
i
a
t
h
e
u
n
b
a
l
a
n
c
e
d
t
i
m
e
seri
e
s
s
a
m
p
l
e
s
,”
I
E
EE
T
r
a
n
s
a
c
t
i
o
n
s
o
n
I
n
st
r
u
m
e
n
t
a
t
i
o
n
a
n
d
Me
a
su
r
e
m
e
n
t
,
v
o
l
.
7
1
,
p
p
.
1
–
1
2
,
2
0
2
2
,
d
o
i
:
1
0
.
1
1
0
9
/
TI
M
.
2
0
2
2
.
3
1
4
5
3
5
9
.
0
20
40
60
80
1
0
0
1
2
0
[
27]
[
28]
our
pr
opos
ed
C
NN-
Dens
e
A
c
c
u
r
a
c
y
A
c
c
u
ra
c
y Co
mp
a
riso
n
R
e
su
l
t
s
Evaluation Warning : The document was created with Spire.PDF for Python.