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.
8
,
No
.
4
,
Dece
m
b
er
201
9
,
p
p
.
399
~
410
I
SS
N:
2
2
5
2
-
8938
,
DOI
: 1
0
.
1
1
5
9
1
/i
j
ai.
v
8
.i
4
.
p
p
399
-
4
1
0
399
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//
ija
i
.
ia
es
co
r
e
.
co
m
I
m
pro
v
ing
so
ft
w
a
re deve
lo
p
m
en
t
e
f
fort e
st
i
m
a
tion us
ing
supp
o
rt
v
ector
re
g
ress
io
n and
f
ea
t
ure selec
tion
Abdela
li
Z
a
k
ra
n
i
1
,
M
us
t
a
ph
a
H
a
in
2
,
Ali
I
dri
3
1,
2
De
p
a
rtm
e
n
t
o
f
In
d
u
strial
E
n
g
in
e
e
rin
g
,
ENS
A
M
,
Ha
ss
a
n
II
Un
iv
e
rsity
,
M
o
ro
c
c
o
3
S
o
f
tw
a
r
e
P
ro
jec
t
M
a
n
a
g
e
m
e
n
t
Re
se
a
rc
h
Tea
m
,
ENS
I
A
S
,
M
o
h
a
m
m
e
d
V
U
n
iv
e
rsity
,
M
o
ro
c
c
o
Art
icle
I
nfo
AB
ST
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Au
g
1
5
,
2
0
1
9
R
ev
i
s
ed
Oct
3
0
,
2
0
1
9
A
cc
ep
ted
No
v
15
,
2
0
1
9
A
c
c
u
ra
te
a
n
d
re
li
a
b
le
so
f
t
w
a
re
d
e
v
e
lo
p
m
e
n
t
e
ff
o
rt
e
sti
m
a
ti
o
n
(S
DEE)
is
o
n
e
o
f
th
e
m
a
in
c
o
n
c
e
rn
s
f
o
r
p
r
o
j
e
c
t
m
a
n
a
g
e
rs.
P
lan
n
in
g
a
n
d
sc
h
e
d
u
l
in
g
a
so
f
t
w
a
re
p
ro
jec
t
u
sin
g
a
n
i
n
a
c
c
u
ra
te
e
sti
m
a
te
m
a
y
c
a
u
se
se
v
e
re
r
isk
s
to
th
e
so
f
t
w
a
re
p
ro
jec
t
u
n
d
e
r
d
e
v
e
lo
p
m
e
n
t
su
c
h
a
s
d
e
la
y
e
d
d
e
li
v
e
r
y
,
p
o
o
r
q
u
a
li
ty
so
f
t
w
a
re
,
m
issin
g
f
e
a
tu
re
s.
T
h
e
re
f
o
re
,
a
n
a
c
c
u
ra
te
p
re
d
ictio
n
o
f
th
e
so
f
t
w
a
re
e
ff
o
rt
p
la
y
s
a
n
i
m
p
o
rtan
t
ro
le
in
th
e
m
in
im
iza
ti
o
n
o
f
th
e
s
e
risk
s
th
a
t
c
a
n
lea
d
to
th
e
p
ro
jec
t
f
a
il
u
re
.
No
w
a
d
a
y
s,
th
e
a
p
p
l
ica
ti
o
n
o
f
a
rti
f
icia
l
i
n
telli
g
e
n
c
e
tec
h
n
iq
u
e
s
h
a
s
g
ro
w
n
d
ra
m
a
ti
c
a
ll
y
f
o
r
p
re
d
ictin
g
so
f
tw
a
r
e
e
ff
o
rt.
T
h
e
re
se
a
rc
h
e
rs
f
o
u
n
d
th
a
t
th
e
se
tec
h
n
iq
u
e
s
a
re
su
it
a
b
le
to
o
ls
f
o
r
a
c
c
u
ra
te
p
re
d
ictio
n
.
In
th
is
stu
d
y
,
a
n
i
m
p
ro
v
e
d
m
o
d
e
l
is
d
e
sig
n
e
d
f
o
r
e
sti
m
a
ti
n
g
so
f
t
w
a
re
e
ff
o
rt
u
sin
g
su
p
p
o
r
t
v
e
c
to
r
re
g
re
ss
io
n
(S
V
R)
a
n
d
t
w
o
f
e
a
tu
re
se
lec
ti
o
n
(F
S
)
m
e
th
o
d
s.
P
rio
r
t
o
b
u
il
d
i
n
g
m
o
d
e
l
ste
p
,
a
p
re
p
r
o
c
e
ss
in
g
sta
g
e
is
p
e
rf
o
r
m
e
d
b
y
ra
n
d
o
m
f
o
re
st
o
r
Bo
ru
ta
f
e
a
tu
re
se
lec
ti
o
n
m
e
th
o
d
s
to
re
m
o
v
e
u
n
im
p
o
rtan
t
f
e
a
tu
re
s.
Ne
x
t,
th
e
S
V
R
m
o
d
e
l
is
t
u
n
e
d
b
y
a
g
rid
se
a
rc
h
a
p
p
ro
a
c
h
.
T
h
e
p
e
rf
o
r
m
a
n
c
e
o
f
th
e
m
o
d
e
ls
is
th
e
n
e
v
a
lu
a
ted
o
v
e
r
e
ig
h
t
w
e
ll
-
k
n
o
w
n
d
a
tas
e
ts
th
ro
u
g
h
3
0
%
h
o
l
d
o
u
t
v
a
li
d
a
ti
o
n
m
e
th
o
d
.
T
o
sh
o
w
th
e
im
p
a
c
t
o
f
f
e
a
tu
re
se
lec
ti
o
n
o
n
th
e
a
c
c
u
ra
c
y
o
f
S
V
R
m
o
d
e
ls,
t
h
e
p
r
o
p
o
se
d
m
o
d
e
l
wa
s
c
o
m
p
a
re
d
w
it
h
S
V
R
m
o
d
e
l
w
it
h
o
u
t
f
e
a
tu
re
se
lec
ti
o
n
.
T
h
e
re
su
lt
s
in
d
ica
ted
th
a
t
S
V
R
w
it
h
f
e
a
tu
re
se
lec
ti
o
n
o
u
tp
e
rf
o
rm
s
S
V
R
w
it
h
o
u
t
F
S
i
n
te
rm
s
o
f
th
e
th
re
e
a
c
c
u
ra
c
y
m
e
a
su
re
s u
se
d
in
t
h
is em
p
iri
c
a
l
stu
d
y
.
K
ey
w
o
r
d
s
:
A
cc
u
r
ac
y
m
ea
s
u
r
es
R
an
d
o
m
f
o
r
est
So
f
t
w
ar
e
ef
f
o
r
t
esti
m
atio
n
Su
p
p
o
r
t v
ec
to
r
r
eg
r
ess
io
n
f
ea
t
u
r
e
s
elec
tio
n
Co
p
y
rig
h
t
©
2
0
1
9
In
stit
u
te o
f
A
d
v
a
n
c
e
d
E
n
g
i
n
e
e
rin
g
a
n
d
S
c
ien
c
e
.
Al
l
rig
h
ts re
se
rv
e
d
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
A
b
d
elali
Z
ak
r
a
n
i,
Dep
ar
t
m
en
t o
f
I
n
d
u
s
tr
ial
E
n
g
i
n
ee
r
in
g
,
E
co
le
Natio
n
ale
Su
p
ér
ie
u
r
e
d
’
A
r
t
s
et
Mé
tier
s
,
1
5
0
A
v
en
u
e
Nile,
S
id
i O
th
m
a
n
,
2
0
6
7
0
,
C
asab
lan
ca
,
Mo
r
o
cc
o
.
E
m
ail: a
b
d
elali.
za
k
r
a
n
i
@
u
n
i
v
h
2
c.
m
a
1.
I
NT
RO
D
UCT
I
O
N
I
n
an
a
g
e
o
f
r
eg
u
lar
tec
h
n
o
lo
g
ical
d
is
r
u
p
tio
n
,
f
o
r
s
o
f
t
w
ar
e
co
m
p
an
ie
s
,
g
r
o
w
i
n
g
f
ast
h
as
b
ec
o
m
e
ess
e
n
tial
to
s
u
r
v
i
v
al.
Mo
r
eo
v
er
,
s
o
f
t
w
ar
e
co
m
p
a
n
ies
m
u
s
t
al
s
o
tar
g
et
b
ec
o
m
i
n
g
p
r
o
f
itab
le
r
ap
id
l
y
an
d
ef
f
icien
tl
y
.
On
e
o
f
t
h
e
m
ai
n
k
e
y
s
to
ac
h
ie
v
e
t
h
is
g
o
al
i
s
to
a
llo
ca
te
s
o
f
t
w
ar
e
p
r
o
j
ec
t
r
eso
u
r
ce
s
ef
f
icie
n
tl
y
a
n
d
s
ch
ed
u
le
ac
tiv
i
ties
as
o
p
ti
m
a
ll
y
a
s
p
o
s
s
ib
le.
I
n
th
i
s
co
n
te
x
t,
esti
m
ati
n
g
s
o
f
t
w
ar
e
d
ev
elo
p
m
en
t
ef
f
o
r
t
is
cr
itical.
Var
io
u
s
m
et
h
o
d
s
h
a
v
e
b
ee
n
in
v
e
s
ti
g
ated
in
s
o
f
t
w
ar
e
ef
f
o
r
t
esti
m
atio
n
,
in
c
lu
d
i
n
g
tr
ad
itio
n
al
m
e
th
o
d
s
s
u
ch
a
s
th
e
co
n
s
tr
u
ctiv
e
co
s
t
m
o
d
el
(
C
OC
OM
O)
[
1
]
,
an
d
,
r
ec
en
t
l
y
,
m
ac
h
i
n
e
lear
n
in
g
tech
n
iq
u
e
s
s
u
c
h
a
s
M
L
P
n
e
u
r
al
n
et
w
o
r
k
s
[
2
]
,
r
ad
ial
b
asis
f
u
n
c
tio
n
(
R
B
F)
n
e
u
r
al
n
et
w
o
r
k
s
[
3
]
,
r
an
d
o
m
f
o
r
est
(
R
F)
[
4
-
5
]
,
f
u
zz
y
an
a
lo
g
y
(
F
A
)
[
6
]
an
d
s
u
p
p
o
r
t v
ec
to
r
r
eg
r
e
s
s
i
o
n
(
SV
R
)
[
7
]
.
Ma
ch
in
e
lear
n
i
n
g
tech
n
iq
u
es
u
s
e
d
ata
f
r
o
m
p
ast p
r
o
j
ec
ts
to
b
u
ild
a
r
eg
r
ess
io
n
m
o
d
el
t
h
at
i
s
s
u
b
s
eq
u
en
t
l
y
e
m
p
lo
y
ed
to
p
r
ed
ict
th
e
e
f
f
o
r
t
o
f
n
e
w
s
o
f
t
w
ar
e
p
r
o
j
ec
ts
.
Ho
w
ev
er
,
n
o
s
in
g
le
m
et
h
o
d
h
a
s
b
ee
n
f
o
u
n
d
to
b
e
en
t
ir
el
y
s
tab
le
a
n
d
r
elia
b
le
f
o
r
all
ca
s
es.
F
u
r
th
er
m
o
r
e,
th
e
p
er
f
o
r
m
a
n
ce
o
f
an
y
m
et
h
o
d
d
ep
en
d
s
m
ai
n
l
y
o
n
th
e
ch
ar
ac
ter
is
tic
s
o
f
th
e
d
ataset
u
s
ed
to
c
o
n
s
tr
u
ct
th
e
m
o
d
el.
T
h
ese
ch
ar
ac
ter
is
tic
s
i
n
cl
u
d
e
d
atase
t
s
ize,
o
u
tlier
s
,
n
u
m
b
er
o
f
f
e
atu
r
es,
ca
te
g
o
r
ical
f
ea
t
u
r
es
a
n
d
m
is
s
in
g
v
al
u
es.
T
h
er
ef
o
r
e,
p
er
f
o
r
m
i
n
g
a
p
r
ep
r
o
ce
s
s
in
g
d
ata
p
r
io
r
to
an
y
SD
E
E
m
o
d
el
b
u
ild
in
g
ca
n
co
n
tr
i
b
u
te
to
i
m
p
r
o
v
e
t
h
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8938
IJ
-
A
I
Vo
l.
8
,
No
.
4
,
Dec
em
b
er
201
9
:
3
9
9
–
410
400
ac
cu
r
ac
y
o
f
th
e
g
e
n
er
ated
esti
m
atio
n
.
Dep
e
n
d
in
g
o
n
d
ata
s
et
u
s
ed
,
th
e
p
r
ep
r
o
ce
s
s
i
n
g
d
ata
c
an
b
e
clea
n
i
n
g
d
ata
b
y
i
m
p
u
tin
g
m
is
s
i
n
g
v
al
u
e
o
r
tr
an
s
f
o
r
m
i
n
g
a
n
d
/o
r
r
ed
u
cin
g
th
e
d
ata
b
y
r
e
m
o
v
i
n
g
r
ed
u
n
d
an
t
a
n
d
ir
r
elev
an
t
f
ea
t
u
r
es.
As
o
n
e
o
f
t
h
e
m
aj
o
r
co
n
ce
r
n
s
w
h
e
n
u
s
in
g
d
atase
t
t
o
co
n
s
tr
u
ct
a
S
DE
E
m
o
d
el
is
t
h
e
n
eg
at
iv
e
i
m
p
ac
t
o
f
ir
r
elev
an
t a
n
d
r
ed
u
n
d
a
n
t in
f
o
r
m
at
io
n
o
n
est
i
m
a
tio
n
ac
c
u
r
ac
y
[
8
]
.
Hen
ce
,
w
e
n
ee
d
to
r
e
m
o
v
e
i
r
r
elev
an
t
a
n
d
r
ed
u
n
d
an
t
in
f
o
r
m
atio
n
an
d
k
ee
p
a
s
u
b
s
et
o
f
r
ele
v
an
t
f
ea
t
u
r
es
s
o
o
n
l
y
in
f
o
r
m
a
tio
n
a
b
o
u
t
th
e
e
f
f
o
r
t
(
d
ep
en
d
en
t
v
ar
iab
le)
is
r
eser
v
ed
.
Fo
r
t
h
is
p
u
r
p
o
s
e,
m
an
y
f
ea
t
u
r
e
s
elec
tio
n
(
FS
)
m
et
h
o
d
s
h
a
v
e
b
ee
n
e
m
p
lo
y
ed
in
t
h
e
liter
a
tu
r
e
[
8
-
1
3
]
.
I
n
t
h
is
co
n
te
x
t,
th
is
p
ap
er
ai
m
s
to
in
v
e
s
ti
g
ate
t
h
e
u
s
e
o
f
t
w
o
f
ea
t
u
r
e
s
elec
tio
n
m
et
h
o
d
s
as
p
r
ep
r
o
ce
s
s
in
g
s
tep
b
ef
o
r
e
f
ee
d
i
n
g
d
ata
to
SVR
m
o
d
el
b
u
ild
in
g
s
ta
g
e.
T
h
e
p
ap
er
ai
m
s
also
to
ev
al
u
ate
w
h
et
h
er
o
r
n
o
t th
e
w
r
ap
p
er
f
ea
tu
r
e
s
elec
tio
n
m
eth
o
d
s
i
m
p
r
o
v
e
th
e
ac
cu
r
ac
y
o
f
th
e
SV
R
m
o
d
el.
T
h
er
ef
o
r
e,
w
e
ass
e
s
s
SV
R
m
o
d
el
s
p
r
ep
r
o
ce
s
s
ed
w
i
th
t
wo
w
r
ap
p
er
m
et
h
o
d
s
an
d
w
e
co
m
p
ar
e
th
e
m
w
it
h
SVR
m
o
d
el
b
u
ilt
w
it
h
o
u
t
f
ea
tu
r
e
s
elec
tio
n
m
et
h
o
d
s
.
T
h
e
m
a
in
co
n
tr
ib
u
tio
n
s
o
f
th
is
p
ap
er
ar
e
th
r
ee
f
o
ld
:
(
1
)
ass
ess
i
n
g
t
h
e
i
m
p
ac
t
o
f
f
ea
t
u
r
e
s
elec
tio
n
m
et
h
o
d
s
o
n
t
h
e
p
r
ed
ictiv
e
ca
p
ab
ilit
y
o
f
SVR
m
o
d
els
o
v
er
eig
h
t
d
atase
ts
(
2
)
em
p
lo
y
i
n
g
t
w
o
w
r
ap
p
er
f
ea
t
u
r
e
s
e
lectio
n
m
et
h
o
d
s
to
s
elec
t
t
h
e
attr
ib
u
te
s
u
s
ed
f
o
r
SVR
m
o
d
els
(
3
)
tu
n
i
n
g
th
e
h
y
p
er
p
ar
a
m
eter
v
al
u
es
o
f
SV
R
u
s
i
n
g
a
g
r
id
s
ea
r
ch
ap
p
r
o
ac
h
an
d
1
0
-
f
o
ld
cr
o
s
s
-
v
alid
atio
n
ap
p
r
o
ac
h
.
T
h
is
p
ap
er
is
o
r
g
an
ized
as
f
o
llo
w
s
.
Sectio
n
2
p
r
esen
ts
t
h
e
SV
R
tech
n
iq
u
e
a
n
d
th
e
tw
o
f
ea
tu
r
e
s
e
lectio
n
m
et
h
o
d
s
u
s
ed
in
t
h
is
s
t
u
d
y
an
d
Sectio
n
3
g
iv
e
s
an
o
v
er
v
ie
w
o
f
r
elate
d
w
o
r
k
co
n
d
u
cte
d
o
n
SV
R
in
SDEE
.
I
n
Secti
o
n
4
,
w
e
d
escr
ib
e
t
h
e
ar
ch
itectu
r
e
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
in
clu
d
i
n
g
th
e
m
et
h
o
d
o
lo
g
y
ad
o
p
ted
to
ad
j
u
s
t
it
p
ar
am
eter
s
v
alu
e
s
.
Sect
io
n
5
p
r
esen
ts
a
b
r
ief
d
escr
ip
tio
n
o
f
th
e
d
atasets
,
t
h
e
ac
cu
r
ac
y
m
ea
s
u
r
e
s
,
t
h
e
v
ali
d
atio
n
m
eth
o
d
u
s
ed
i
n
th
i
s
s
tu
d
y
.
T
h
e
e
m
p
ir
ical
r
es
u
lts
ar
e
p
r
esen
ted
a
n
d
d
is
cu
s
s
ed
in
Sect
io
n
6
.
Fin
al
l
y
,
Sectio
n
7
co
n
clu
d
es t
h
e
p
ap
er
.
2.
B
ACK
G
RO
UND
B
ef
o
r
e
en
ter
in
g
i
n
to
d
etail
s
,
w
e
in
tr
o
d
u
ce
th
e
t
h
r
ee
m
a
in
to
o
ls
o
f
t
h
is
p
ap
er
:
s
u
p
p
o
r
t
v
ec
to
r
r
eg
r
ess
io
n
,
f
ea
t
u
r
e
i
m
p
o
r
tan
ce
,
an
d
f
ea
tu
r
e
s
e
lectio
n
.
2
.
1
.
Su
pp
o
rt
v
ec
t
o
r
re
g
re
s
s
io
n
Su
p
p
o
r
t
Vec
to
r
Ma
ch
i
n
es
a
s
d
escr
ib
ed
in
[
1
4
]
h
av
e
s
h
o
w
n
t
o
d
eliv
er
p
r
o
m
is
i
n
g
s
o
l
u
tio
n
s
in
v
ar
io
u
s
class
i
f
icatio
n
an
d
r
eg
r
es
s
io
n
task
s
th
a
n
k
s
to
th
e
ir
ab
ilit
y
t
o
av
o
id
lo
ca
l
m
i
n
i
m
a,
i
m
p
r
o
v
ed
g
en
er
al
izatio
n
ca
p
ab
ilit
y
,
an
d
s
p
ar
s
e
r
ep
r
ese
n
ta
tio
n
o
f
t
h
e
s
o
lu
tio
n
.
SVM
ar
e
b
ased
o
n
Stru
ctu
r
al
R
i
s
k
Min
i
m
izatio
n
(
SR
M)
p
r
in
cip
le
an
d
t
h
u
s
tr
ie
s
to
co
n
tr
o
l
th
e
u
p
p
er
b
o
u
n
d
o
f
g
e
n
er
aliza
tio
n
r
is
k
w
h
ile
r
ed
u
cin
g
t
h
e
m
o
d
el
co
m
p
le
x
it
y
.
I
n
ad
d
itio
n
,
th
e
y
d
o
n
o
t s
u
f
f
er
f
r
o
m
o
v
er
f
i
tti
n
g
p
r
o
b
lem
an
d
lo
ca
l
m
i
n
i
m
izati
o
n
is
s
u
e
s
a
n
d
h
en
c
e
o
f
f
er
e
n
h
a
n
ce
d
g
e
n
er
aliza
tio
n
ca
p
ab
ilit
y
.
Fo
r
r
eg
r
ess
io
n
ta
s
k
s
,
Vap
n
i
k
p
r
o
p
o
s
ed
an
SV
M
ca
lled
ε
-
s
u
p
p
o
r
t
v
ec
to
r
r
eg
r
ess
io
n
(
ε
-
S
VR
)
,
wh
ich
p
er
f
o
r
m
s
p
r
ed
ictio
n
tas
k
s
f
r
o
m
t
h
e
ε
-
i
n
s
e
n
s
iti
v
e
lo
s
s
f
u
n
ctio
n
.
B
ec
au
s
e
th
e
ε
p
ar
a
m
eter
i
s
u
s
e
f
u
l
i
f
th
e
ap
p
r
o
x
i
m
atio
n
ac
cu
r
ac
y
i
s
s
p
ec
i
f
ied
b
ef
o
r
e
h
an
d
,
it
is
b
etter
to
f
i
n
d
a
p
r
o
ce
d
u
r
e
to
o
p
tim
ize
th
is
ac
c
u
r
ac
y
w
i
th
o
u
t
d
ep
en
d
in
g
a
p
r
io
r
i
o
n
a
v
a
lu
e
s
et.
T
h
is
p
r
o
ce
d
u
r
e
w
as
s
tu
d
ied
b
y
Sö
lk
o
p
f
,
S
m
o
la,
W
illi
a
m
s
o
n
an
d
B
ar
tlett
[
1
5
]
.
T
h
ey
p
r
o
p
o
s
ed
a
n
ew
f
o
r
m
u
latio
n
,
ca
lled
ʋ
-
s
u
p
p
o
r
t
v
ec
to
r
r
eg
r
ess
io
n
(ʋ
-
SV
R
)
,
th
at
a
u
to
m
atica
ll
y
m
i
n
i
m
izes
t
h
e
ε
-
i
n
s
e
n
s
it
iv
e
l
o
s
s
f
u
n
ctio
n
a
n
d
ch
a
n
g
es
t
h
e
SVR
f
o
r
m
u
latio
n
b
y
u
s
i
n
g
a
n
e
w
ʋ
p
ar
a
m
eter
w
h
o
s
e
v
al
u
e
is
b
et
w
ee
n
[
0
,
1
]
.
I
n
ad
d
itio
n
to
m
i
n
i
m
izi
n
g
th
e
ε
v
alu
e,
th
e
ʋ
p
ar
am
eter
i
s
u
s
ed
f
o
r
co
n
tr
o
lli
n
g
t
h
e
n
u
m
b
er
o
f
s
u
p
p
o
r
t
v
ec
t
o
r
s
,
s
in
ce
t
h
e
v
al
u
e
o
f
ε
in
f
l
u
e
n
ce
s
t
h
e
ch
o
ice
o
f
s
u
p
p
o
r
t v
ec
to
r
s
.
I
n
th
is
s
t
u
d
y
,
a
s
p
ec
ial
f
o
r
m
o
f
SVM
i.e
.
,
Su
p
p
o
r
t
Vec
to
r
R
eg
r
ess
io
n
(
SVR
)
is
u
ti
lized
f
o
r
m
o
d
elin
g
th
e
in
p
u
t
–
o
u
tp
u
t
f
u
n
ctio
n
al
r
e
latio
n
s
h
ip
o
r
r
eg
r
ess
io
n
p
u
r
p
o
s
e
an
d
is
ex
p
lai
n
ed
n
e
x
t.
Gi
v
en
a
s
e
t
o
f
in
p
u
t
–
o
u
tp
u
t
s
a
m
p
le
p
air
s
{
(
1
,
1
)
,
(
2
,
2
)
,
.
.
.
,
(
,
)
}
w
h
er
e
∈
an
d
∈
,
th
e
o
b
j
ec
tiv
e
o
f
ν
-
SV
R
tech
n
iq
u
e
is
to
ap
p
r
o
x
im
a
te
th
e
n
o
n
l
in
ea
r
r
elatio
n
s
h
ip
g
i
v
en
in
(
1
)
,
s
u
ch
th
at
(
)
s
h
o
u
ld
b
e
as
clo
s
e
as
p
o
s
s
ib
le
to
th
e
tar
g
et
v
alu
e
y
a
n
d
s
h
o
u
ld
b
e
as f
lat
as p
o
s
s
ib
le
in
o
r
d
er
to
av
o
id
o
v
er
-
f
itti
n
g
.
(
)
=
.
(
)
+
(
1
)
w
h
er
e
is
th
e
w
ei
g
h
t
v
ec
to
r
,
is
th
e
b
ias
an
d
(
)
r
ep
r
esen
ts
th
e
tr
an
s
f
o
r
m
a
tio
n
f
u
n
ctio
n
th
a
t
m
ap
s
th
e
lo
w
er
d
i
m
e
n
s
io
n
al
i
n
p
u
t
s
p
a
ce
to
a
h
ig
h
er
d
i
m
e
n
s
io
n
al
s
p
ac
e.
T
h
e
p
r
im
a
l
o
b
j
ec
tiv
e
o
f
t
h
e
p
r
o
b
lem
t
h
u
s
r
ed
u
ce
s
to
(
2
)
,
in
o
r
d
er
to
en
s
u
r
e
t
h
at
t
h
e
ap
p
r
o
x
i
m
ated
f
u
n
ctio
n
m
ee
ts
t
h
e
ab
o
v
e
t
w
o
o
b
jectiv
es
o
f
c
lo
s
en
e
s
s
an
d
f
lat
n
es
s
.
m
in
im
ize
1
2
‖
‖
2
+
{
.
+
1
2
∑
(
+
∗
)
=
1
}
(
2
)
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
-
AI
I
SS
N:
2252
-
8938
I
mp
r
o
vin
g
s
o
ftw
a
r
e
d
ev
elo
p
m
en
t e
ffo
r
t e
s
tima
tio
n
u
s
in
g
s
u
p
p
o
r
t v
ec
to
r
r
eg
r
ess
io
n
a
n
d
... (
A
b
d
ela
li Za
kra
n
i
)
401
s
ubj
e
c
t
to
the
c
on
s
tr
a
in
t
s
{
−
〈
.
(
)
〉
−
≤
+
∗
,
〈
.
(
)
〉
+
−
≤
+
∗
,
,
∗
≥
0
w
h
er
e
ε
is
a
d
ev
iatio
n
o
f
a
f
u
n
ctio
n
f
(
x)
f
r
o
m
its
ac
t
u
a
l
v
alu
e
an
d
,
ξ,
ξ
i
*
ar
e
ad
d
itio
n
al
s
lack
v
ar
iab
les
in
tr
o
d
u
ce
d
b
y
C
o
r
tes
&
Vap
n
ik
,
1
9
9
5
,
w
h
ic
h
d
eter
m
i
n
es
t
h
at,
d
ev
iatio
n
s
o
f
m
a
g
n
itu
d
e
ξ
ab
o
v
e
ε
e
r
r
o
r
ar
e
to
ler
ated
.
T
h
e
co
n
s
tan
t
C
k
n
o
w
n
a
s
r
e
g
u
lar
izat
io
n
p
ar
a
m
ete
r
d
eter
m
in
e
s
t
h
e
tr
ad
eo
f
f
b
et
wee
n
t
h
e
f
lat
n
e
s
s
o
f
f
an
d
to
ler
an
ce
o
f
er
r
o
r
ab
o
v
e
ε.
Fu
r
th
er
ϒ
(
0
≤
ϒ≤
1
)
,
r
ep
r
es
en
ts
t
h
e
u
p
p
er
b
o
u
n
d
o
n
th
e
f
u
n
ctio
n
o
f
m
ar
g
i
n
er
r
o
r
s
in
th
e
tr
ain
i
n
g
s
et
an
d
estab
lis
h
e
s
th
e
lo
w
er
b
o
u
n
d
o
n
t
h
e
f
r
ac
t
io
n
o
f
s
u
p
p
o
r
t
v
ec
to
r
s
.
T
o
s
o
lv
e
th
e
p
r
im
a
l p
r
o
b
lem
i
n
(
2
)
,
its
d
u
al
f
o
r
m
u
latio
n
i
s
in
tr
o
d
u
ce
d
b
y
co
n
s
tr
u
ct
in
g
L
ag
r
a
n
g
e
f
u
n
c
tio
n
(
L
)
g
iv
e
n
as:
:
1
2
‖
‖
2
+
{
Υ
.
+
1
∑
(
+
∗
)
=
1
}
−
1
∑
(
.
+
∗
∗
)
=
1
−
1
∑
(
+
1
−
.
(
)
−
)
=
1
+
1
∑
(
+
−
.
(
)
+
)
−
.
=
1
(
3
)
w
h
er
e
,
∗
,
,
∗
ar
e
L
ag
r
an
g
e
m
u
l
tip
lie
r
s
an
d
(
∗
)
=
.
∗
.
T
h
u
s
,
m
a
x
i
m
izi
n
g
th
e
L
ag
r
an
g
e
f
u
n
ctio
n
g
iv
e
s
=
∑
(
−
∗
)
=
1
.
(
)
an
d
y
ield
s
th
e
f
o
llo
w
i
n
g
d
u
al
o
p
ti
m
iza
tio
n
p
r
o
b
le
m
:
m
ax
i
m
izes
−
1
2
∑
(
−
∗
)
.
(
−
∗
)
,
=
1
.
(
,
)
+
∑
.
(
−
∗
)
;
=
1
s
ubj
e
c
t
to
{
∑
(
=
1
−
∗
)
=
0
,
∑
(
=
1
−
∗
)
≤
,
,
∗
∈
[
0
,
]
(
4
)
w
h
er
e
(
,
)
d
en
o
tes
th
e
k
er
n
el
f
u
n
ctio
n
g
i
v
en
b
y
(
,
)
=
(
)
.
(
)
.
T
h
e
s
o
lu
tio
n
to
(
4
)
y
ield
s
th
e
L
ag
r
a
n
g
e
m
u
ltip
lier
s
,
∗
.
Su
b
s
tit
u
ti
n
g
w
ei
g
h
t
w
i
n
(
1
)
,
th
e
ap
p
r
o
x
im
a
ted
f
u
n
ctio
n
is
g
i
v
en
as:
(
)
=
∑
(
−
∗
)
.
=
1
(
,
)
+
(
5
)
T
h
e
ch
o
ice
o
f
k
er
n
el
f
u
n
ctio
n
f
o
r
s
p
ec
if
ic
d
ata
p
atter
n
s
,
w
h
ich
is
a
n
o
th
er
attr
ac
ti
v
e
q
u
est
io
n
in
t
h
e
ap
p
licatio
n
o
f
SV
R
,
ap
p
ea
r
ed
s
o
m
e
w
h
at
ar
b
itra
r
y
till
n
o
w
.
So
m
e
p
r
ev
io
u
s
w
o
r
k
[
6
,
1
6
]
e
m
p
ir
icall
y
in
d
icat
e
th
at
t
h
e
u
s
e
o
f
t
h
e
g
au
s
s
ia
n
R
B
F
k
er
n
el
is
s
u
p
er
io
r
to
o
th
er
k
er
n
el
f
u
n
ctio
n
s
b
ec
au
s
e
o
f
it
s
ac
ce
s
s
ib
ilit
y
to
i
m
p
le
m
en
t
a
n
d
p
o
w
er
f
u
l
m
ap
p
in
g
ca
p
ab
ilit
y
.
T
h
er
ef
o
r
e
,
th
e
g
a
u
s
s
ia
n
R
B
F
k
er
n
el
f
u
n
ctio
n
,
(
6
)
,
w
as
e
m
p
lo
y
ed
in
t
h
i
s
s
t
u
d
y
.
(
,
)
=
e
xp
(
−
‖
−
‖
2
)
ℎ
=
1
2
2
(
6
)
T
h
e
p
ar
am
eter
σ
af
f
ec
ts
t
h
e
m
ap
p
in
g
tr
an
s
f
o
r
m
atio
n
o
f
t
h
e
in
p
u
t
d
ata
to
th
e
f
ea
tu
r
e
s
p
ac
e
an
d
co
n
tr
o
ls
th
e
co
m
p
lex
it
y
o
f
t
h
e
m
o
d
el,
th
u
s
,
an
d
th
e
v
alu
e
o
f
p
ar
am
eter
s
h
o
u
ld
b
e
s
elec
ted
ca
r
ef
u
ll
y
a
n
d
ad
eq
u
atel
y
.
I
n
ad
d
itio
n
,
SV
R
r
eq
u
ir
es
also
s
e
tti
n
g
t
w
o
p
ar
am
eter
s
:
th
e
co
m
p
lex
it
y
p
ar
a
m
eter
u
s
u
all
y
d
en
o
ted
b
y
C
,
t
h
e
e
x
ten
t
to
w
h
ic
h
d
e
v
iatio
n
s
(
i.e
.
,
er
r
o
r
s
)
ar
e
to
ler
ated
d
en
o
ted
b
y
E
p
s
ilo
n
(
ε)
,
an
d
t
h
e
ʋ
p
ar
a
m
eter
w
h
ic
h
is
u
s
ed
f
o
r
co
n
tr
o
lli
n
g
th
e
n
u
m
b
er
o
f
s
u
p
p
o
r
t
v
ec
to
r
s
,
s
in
ce
th
e
v
al
u
e
o
f
ε
i
n
f
lu
e
n
ce
s
t
h
e
c
h
o
ice
o
f
s
u
p
p
o
r
t v
ec
to
r
s
.
2
.
2
.
F
e
a
t
ure
s
elec
t
io
n
m
et
ho
ds
T
h
is
s
u
b
s
ec
tio
n
p
r
o
v
id
es
an
o
v
er
v
ie
w
o
f
t
h
e
f
ea
tu
r
e
s
ele
ctio
n
m
et
h
o
d
s
w
it
h
p
ar
ticu
la
r
f
o
cu
s
o
n
f
ea
t
u
r
e
i
m
p
o
r
tan
ce
co
n
ce
p
t u
s
ed
b
y
th
e
m
et
h
o
d
s
u
s
ed
i
n
t
h
is
p
ap
er
.
2
.
2
.
1
.
F
ea
t
ure
s
elec
t
io
n
m
et
h
o
ds
Featu
r
e
s
elec
tio
n
,
a
ls
o
k
n
o
w
n
as
v
ar
iab
le
s
elec
tio
n
,
is
th
e
p
r
o
ce
s
s
o
f
id
e
n
ti
f
y
in
g
t
h
e
m
o
s
t
p
r
o
m
i
s
i
n
g
f
ea
t
u
r
es
(
v
ar
iab
les,
attr
ib
u
tes)
in
a
g
iv
e
n
d
ataset.
T
h
e
s
elec
t
ed
f
ea
t
u
r
e
w
ill
b
e
u
s
ed
to
co
n
s
tr
u
ct
th
e
m
o
d
el
o
r
as
in
p
u
ts
o
f
a
p
r
ed
ictio
n
s
y
s
te
m
.
T
h
er
e
ar
e
m
a
n
y
p
o
ten
tia
l
b
en
ef
its
o
f
f
ea
t
u
r
e
s
elec
tio
n
s
u
c
h
as
i
m
p
r
o
v
in
g
t
h
e
g
en
er
aliza
tio
n
p
er
f
o
r
m
an
ce
o
f
th
e
p
r
ed
icti
v
e
m
o
d
el,
r
ed
u
c
in
g
t
h
e
co
m
p
u
tatio
n
al
ti
m
e
to
c
o
n
s
tr
u
ct
t
h
e
m
o
d
el,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8938
IJ
-
A
I
Vo
l.
8
,
No
.
4
,
Dec
em
b
er
201
9
:
3
9
9
–
410
402
an
d
b
etter
u
n
d
er
s
tan
d
i
n
g
t
h
e
u
n
d
er
l
y
in
g
p
r
o
ce
s
s
.
Sev
er
al
f
ea
tu
r
e
s
elec
tio
n
m
et
h
o
d
s
h
av
e
b
ee
n
p
r
o
p
o
s
ed
an
d
s
tu
d
ied
i
n
th
e
liter
at
u
r
e
[
1
7
]
.
T
h
ey
ca
n
f
all
i
n
to
t
h
r
ee
ca
te
g
o
r
ies:
th
e
w
r
ap
p
er
,
th
e
f
il
ter
a
n
d
e
m
b
ed
d
ed
.
T
h
e
w
r
ap
p
er
m
e
th
o
d
s
u
s
e
a
p
r
ed
ic
tiv
e
m
o
d
el
to
s
co
r
e
f
ea
tu
r
e
s
u
b
s
ets.
E
ac
h
n
e
w
s
u
b
s
et
is
u
s
e
d
to
t
r
ain
a
m
o
d
el,
w
h
ic
h
i
s
te
s
ted
o
n
a
h
o
ld
-
o
u
t
s
et.
C
o
u
n
tin
g
th
e
n
u
m
b
er
o
f
m
is
tak
e
s
m
ad
e
o
n
t
h
at
h
o
ld
-
o
u
t
s
et
(
t
h
e
er
r
o
r
r
ate
o
f
th
e
m
o
d
el)
g
iv
e
s
t
h
e
s
co
r
e
f
o
r
th
at
s
u
b
s
et
[
1
8
]
.
T
h
e
f
ilter
m
et
h
o
d
s
co
n
s
id
er
s
tati
s
tical
ch
ar
ac
ter
is
tic
s
o
f
a
d
ata
s
et
d
ir
ec
tly
w
it
h
o
u
t
in
v
o
l
v
in
g
an
y
lear
n
in
g
alg
o
r
it
h
m
.
T
h
e
em
b
ed
d
e
d
m
et
h
o
d
s
co
m
b
in
e
f
ea
t
u
r
e
s
elec
tio
n
an
d
t
h
e
lear
n
in
g
p
r
o
ce
s
s
in
o
r
d
er
to
s
elec
t
an
o
p
ti
m
al
s
u
b
s
et
o
f
f
ea
t
u
r
es.
I
n
g
en
er
al,
th
e
r
esu
l
ts
o
f
w
r
ap
p
er
m
et
h
o
d
s
ar
e
b
etter
th
an
th
o
s
e
o
f
f
ilter
m
et
h
o
d
s
.
Ho
w
e
v
er
,
t
h
e
w
r
ap
p
er
m
et
h
o
d
is
s
lo
w
(
t
i
m
e
-
co
n
s
u
m
i
n
g
)
an
d
v
er
y
co
m
p
licated
w
h
e
n
t
h
er
e
ar
e
m
an
y
f
ea
t
u
r
es
i
n
t
h
e
d
ata
s
et.
Fo
r
tu
n
atel
y
,
i
n
o
u
r
ca
s
e,
t
h
e
d
ataset
s
u
s
ed
i
n
th
is
s
t
u
d
y
h
a
v
e
r
elati
v
el
y
a
s
m
all
n
u
m
b
er
o
f
f
ea
tu
r
e
s
.
2
.
2
.
2
.
Ra
nd
o
m
f
o
re
s
t
f
ea
t
ure
i
m
po
rt
a
nce
R
an
d
o
m
f
o
r
est
(
R
F)
is
an
en
s
e
m
b
le
lear
n
i
n
g
tec
h
n
iq
u
e
ba
s
ed
o
n
clas
s
i
f
ic
atio
n
an
d
r
eg
r
ess
io
n
tr
ee
s
[
1
9
]
.
E
ac
h
tr
ee
is
tr
ain
ed
o
n
a
b
o
o
ts
tr
ap
s
a
m
p
le,
a
n
d
o
p
t
i
m
al
v
ar
iab
le
s
at
ea
c
h
s
p
lit
ar
e
id
en
ti
f
ied
f
r
o
m
a
r
a
n
d
o
m
s
u
b
s
et
o
f
all
v
ar
iab
les.
T
h
e
s
ele
ctin
g
cr
iter
ia
ar
e
d
i
f
f
er
e
n
t
f
o
r
class
i
f
icat
io
n
a
n
d
r
eg
r
ess
io
n
p
r
o
b
lem
s
.
Fo
r
th
e
f
o
r
m
er
s
e
tti
n
g
,
t
h
e
Gi
n
i
i
n
d
ex
i
s
ap
p
lied
,
w
h
er
ea
s
v
ar
ia
n
ce
r
e
d
u
ctio
n
i
s
u
s
ed
f
o
r
th
e
la
tter
ap
p
r
o
ac
h
.
T
h
e
g
lo
b
al
p
r
ed
ictio
n
o
f
t
h
e
R
F
is
co
m
p
u
ted
a
s
a
m
aj
o
r
it
y
v
o
te
o
r
av
er
ag
e
f
o
r
class
i
f
icatio
n
o
r
r
eg
r
ess
io
n
,
r
esp
ec
tiv
el
y
[
2
0
]
.
I
n
ad
d
itio
n
to
p
r
e
d
ictio
n
,
R
Fs
ca
n
b
e
u
s
ed
as
m
e
th
o
d
to
esti
m
ate
v
ar
iab
le
i
m
p
o
r
tan
ce
m
ea
s
u
r
es
to
r
a
n
k
v
ar
iab
les
b
y
p
r
ed
ictiv
e
i
m
p
o
r
tan
ce
.
T
o
illu
s
tr
ate
th
is
,
let
’
s
a
p
r
o
j
ec
t
f
ea
tu
r
e.
R
F
f
ea
t
u
r
e
i
m
p
o
r
tan
ce
o
f
is
d
ef
i
n
ed
,
as
d
escr
ib
ed
in
[
2
1
]
,
as
f
o
llo
w
s
.
Fo
r
ea
ch
tr
ee
t
o
f
th
e
f
o
r
est,
co
n
s
id
er
th
e
ass
o
ciate
d
s
am
p
le
(
Ou
t
Of
B
ag
is
t
h
e
d
ata
w
h
ich
w
a
s
n
o
t
in
cl
u
d
ed
in
th
e
b
o
o
s
tr
ap
s
a
m
p
le
u
s
ed
to
co
n
s
tr
u
ct
)
.
Den
o
te
b
y
th
e
m
ea
n
s
q
u
ar
e
er
r
o
r
(
MSE
)
o
f
a
s
in
g
le
tr
ee
t
o
n
t
h
is
s
a
m
p
le.
No
w
,
r
a
n
d
o
m
l
y
p
er
m
u
te
th
e
v
alu
e
s
o
f
in
to
g
et
a
p
er
tu
r
b
ed
s
am
p
le
d
en
o
ted
b
y
̌
an
d
co
m
p
u
te
,
th
e
er
r
o
r
o
f
p
r
e
d
icto
r
t
o
n
th
e
p
er
tu
r
b
ed
s
a
m
p
le.
Fe
atu
r
e
i
m
p
o
r
tan
ce
o
f
is
t
h
en
eq
u
al
to
:
(
)
=
1
∑
(
̌
−
)
,
(
7
)
w
h
er
e
th
e
s
u
m
is
o
v
er
all
tr
ee
s
o
f
th
e
R
F
a
n
d
d
en
o
tes
th
e
n
u
m
b
er
o
f
tr
ee
s
o
f
t
h
e
R
F.
Featu
r
es
th
at
ar
e
r
elev
a
n
t
f
o
r
p
r
ed
ictio
n
w
i
ll
h
a
v
e
lar
g
e
i
m
p
o
r
tan
ce
v
alu
e
s
,
w
h
er
ea
s
f
ea
tu
r
es
t
h
at
ar
e
n
o
t
a
s
s
o
ciate
d
w
it
h
t
h
e
o
u
tco
m
e
h
a
v
e
v
al
u
e
s
clo
s
e
to
ze
r
o
.
2
.
2
.
3
.
B
o
rut
a
f
ea
t
ure
s
elec
t
io
n
m
et
ho
d
B
o
r
u
ta
is
an
a
ll
r
elev
a
n
t
f
ea
t
u
r
e
s
elec
tio
n
al
g
o
r
ith
m
,
i.e
.
,
e
m
b
ed
d
ed
w
it
h
t
h
e
R
F
al
g
o
r
ith
m
a
n
d
u
s
e
s
ca
lcu
lated
Z
-
s
co
r
es
as
a
m
ea
s
u
r
e
o
f
b
an
d
i
m
p
o
r
tan
ce
.
T
h
e
m
ai
n
id
ea
o
f
th
is
ap
p
r
o
ac
h
is
to
co
m
p
ar
e
th
e
i
m
p
o
r
tan
ce
o
f
t
h
e
r
ea
l
p
r
ed
icto
r
v
ar
iab
les
w
it
h
t
h
o
s
e
o
f
r
an
d
o
m
s
o
-
ca
lled
s
h
ad
o
w
v
ar
iab
les
u
s
in
g
s
tat
is
tica
l
test
i
n
g
a
n
d
s
ev
er
al
r
u
n
s
o
f
R
F
s
[
2
2
]
.
I
n
ea
ch
r
u
n
,
th
e
s
et
o
f
p
r
ed
icto
r
v
ar
iab
les
is
d
o
u
b
led
b
y
ad
d
in
g
a
co
p
y
o
f
ea
ch
v
ar
iab
le.
T
h
e
v
alu
es
o
f
th
o
s
e
s
h
ad
o
w
v
ar
iab
les
a
r
e
g
en
er
ated
b
y
p
er
m
u
ti
n
g
t
h
e
o
r
ig
in
a
l
v
al
u
es
ac
r
o
s
s
o
b
s
er
v
atio
n
s
an
d
t
h
er
e
f
o
r
e
d
estro
y
i
n
g
t
h
e
r
elatio
n
s
h
ip
w
it
h
t
h
e
o
u
tco
m
e.
A
R
F is
tr
ain
ed
o
n
t
h
e
e
x
te
n
d
ed
d
ata
s
et
a
n
d
t
h
e
v
ar
iab
le
i
m
p
o
r
tan
ce
v
al
u
es
ar
e
co
llected
.
Fo
r
ea
ch
r
ea
l
v
ar
iab
le
a
s
ta
tis
tica
l
te
s
t
is
p
er
f
o
r
m
e
d
co
m
p
ar
i
n
g
its
i
m
p
o
r
ta
n
ce
w
it
h
t
h
e
m
a
x
i
m
u
m
v
al
u
e
o
f
all
t
h
e
s
h
ad
o
w
v
ar
iab
les.
Var
iab
le
s
w
it
h
s
i
g
n
i
f
ica
n
tl
y
lar
g
er
o
r
s
m
al
ler
i
m
p
o
r
tan
ce
v
alu
e
s
ar
e
d
ec
lar
ed
as
i
m
p
o
r
t
an
t
o
r
u
n
i
m
p
o
r
ta
n
t,
r
esp
ec
ti
v
e
l
y
.
A
ll
u
n
i
m
p
o
r
ta
n
t
v
ar
iab
les
an
d
s
h
ad
o
w
v
ar
ia
b
les
ar
e
r
em
o
v
ed
an
d
th
e
p
r
ev
io
u
s
s
tep
s
ar
e
r
ep
ea
ted
u
n
til
all
v
ar
iab
les
ar
e
class
i
f
ied
o
r
a
ce
r
tain
d
eter
m
i
n
ed
n
u
m
b
er
o
f
r
u
n
s
h
as b
ee
n
d
o
n
e
[
2
0
]
.
3.
RE
L
AT
E
D
WO
RK
T
h
e
SVR
tec
h
n
iq
u
e
h
a
s
b
ee
n
u
s
ed
in
m
a
n
y
e
m
p
ir
ical
s
o
f
t
w
ar
e
e
n
g
i
n
ee
r
i
n
g
s
tu
d
ie
s
es
p
ec
iall
y
i
n
p
r
ed
ictin
g
s
e
v
er
al
s
o
f
t
w
ar
e
c
h
ar
ac
ter
is
tic
s
s
u
c
h
as
b
u
g
a
n
d
d
ef
ec
t
[
2
3
-
2
4
]
,
r
eliab
ilit
y
[
2
5
]
,
q
u
alit
y
[
2
6
]
an
d
en
h
a
n
ce
m
en
t
ef
f
o
r
t
[
2
7
]
.
R
eg
ar
d
in
g
ap
p
licatio
n
o
f
an
SV
R
f
o
r
esti
m
ati
n
g
s
o
f
t
w
ar
e
d
ev
e
lo
p
m
e
n
t
e
f
f
o
r
t,
w
e
id
en
ti
f
ied
1
3
r
elev
a
n
t
s
tu
d
ie
s
in
th
e
liter
at
u
r
e
[
7
,
2
7
-
3
8
]
.
T
h
e
f
ir
s
t
i
n
v
e
s
ti
g
atio
n
o
f
S
VR
i
n
SDEE
w
a
s
o
r
ig
in
all
y
ca
r
r
ied
o
u
t
b
y
Oli
v
eir
a
[
7
]
.
He
h
as
co
n
s
id
er
ed
SVR
w
ith
li
n
ea
r
as
w
e
ll
as
R
B
F
k
er
n
els
a
n
d
o
p
tim
ized
its
p
ar
a
m
eter
s
e
m
p
l
o
y
i
n
g
g
r
id
s
e
lectio
n
.
T
h
e
ex
p
er
i
m
en
ts
w
er
e
p
er
f
o
r
m
ed
u
s
i
n
g
s
o
f
t
w
ar
e
p
r
o
j
ec
ts
f
r
o
m
N
A
S
A
d
ata
s
et
an
d
th
e
r
esu
lts
h
av
e
s
h
o
w
n
t
h
at
S
V
R
s
i
g
n
i
f
ica
n
tl
y
o
u
tp
er
f
o
r
m
s
R
B
FNs
an
d
li
n
ea
r
r
eg
r
ess
io
n
.
Hi
s
w
o
r
k
d
id
n
o
t
i
n
v
e
s
ti
g
ate
f
ea
t
u
r
e
s
e
lectio
n
m
eth
o
d
s
;
all
i
n
p
u
t
f
ea
tu
r
es
w
er
e
u
s
ed
f
o
r
b
u
ild
in
g
th
e
r
e
g
r
ess
io
n
m
o
d
els.
In
[
2
8
,
3
9
]
u
s
ed
a
g
en
et
ic
al
g
o
r
ith
m
(
GA
)
ap
p
r
o
ac
h
to
s
elec
t
an
o
p
ti
m
al
s
u
b
s
et
f
ea
t
u
r
e
an
d
o
p
ti
m
ize
SV
R
p
ar
a
m
eter
f
o
r
SDEE
.
T
h
e
y
u
s
ed
b
in
ar
y
co
d
ed
ch
r
o
m
o
s
o
m
e
a
s
s
o
l
u
tio
n
r
ep
r
esen
ta
tio
n
f
o
r
s
u
b
s
et
f
ea
tu
r
e
a
n
d
S
VR
p
ar
am
eter
.
T
h
eir
s
i
m
u
latio
n
s
h
a
v
e
s
h
o
w
n
t
h
at
th
e
p
r
o
p
o
s
ed
GA
-
b
ased
ap
p
r
o
ac
h
w
as
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
-
AI
I
SS
N:
2252
-
8938
I
mp
r
o
vin
g
s
o
ftw
a
r
e
d
ev
elo
p
m
en
t e
ffo
r
t e
s
tima
tio
n
u
s
in
g
s
u
p
p
o
r
t v
ec
to
r
r
eg
r
ess
io
n
a
n
d
... (
A
b
d
ela
li Za
kra
n
i
)
403
ab
le
to
i
m
p
r
o
v
e
s
u
b
s
tan
tiall
y
t
h
e
p
er
f
o
r
m
a
n
ce
o
f
S
VR
an
d
o
u
tp
er
f
o
r
m
b
a
g
g
i
n
g
ML
P
n
et
w
o
r
k
an
d
b
ag
g
i
n
g
M5
P
.
T
h
e
au
th
o
r
s
in
[
3
6
]
in
v
esti
g
ated
p
ar
ticle
s
w
ar
m
o
p
ti
m
iza
tio
n
(
P
SO)
ap
p
licatio
n
to
s
elec
t
s
u
b
s
et
f
ea
t
u
r
e
an
d
SV
R
p
ar
a
m
eter
ap
p
lied
to
s
o
f
t
w
ar
e
ef
f
o
r
t
es
ti
m
atio
n
.
T
h
e
y
u
s
ed
co
n
ti
n
u
o
u
s
v
al
u
e
t
y
p
e
to
o
p
tim
ize
SV
R
p
ar
a
m
eter
a
n
d
d
is
cr
ete
v
al
u
e
t
y
p
e
to
s
elec
t
s
u
b
s
et
f
ea
t
u
r
e.
Ho
w
e
v
er
,
th
e
s
t
u
d
y
w
a
s
li
m
ited
to
Desh
ar
n
ais
d
ataset
a
n
d
d
o
es
n
o
t
s
h
o
w
th
e
p
er
f
o
r
m
a
n
ce
o
f
th
e
r
esu
lt
in
g
SV
R
m
o
d
el
u
s
i
n
g
co
m
m
o
n
l
y
e
m
p
lo
y
ed
ac
c
u
r
ac
y
m
ea
s
u
r
es
in
SDEE
.
S
u
p
p
o
r
t
v
ec
to
r
r
eg
r
ess
io
n
h
as
b
ee
n
also
u
s
e
d
to
esti
m
a
te
t
h
e
d
ev
elo
p
m
en
t
e
f
f
o
r
t
o
f
w
eb
p
r
o
j
ec
ts
u
s
in
g
T
u
k
u
t
u
k
u
d
atas
et
in
[
3
0
,
4
0
-
4
1
]
.
T
h
e
r
esu
lt
s
o
f
th
ese
s
tu
d
ies
s
h
o
w
ed
t
h
at
SVR
h
a
s
p
o
ten
ti
al
s
i
n
ce
it
o
u
tp
er
f
o
r
m
ed
th
e
m
o
s
t
co
m
m
o
n
l
y
ad
o
p
ted
p
r
ed
ictio
n
tec
h
n
iq
u
es.
I
t
w
a
s
ar
g
u
ed
th
at
S
VR
i
s
a
f
le
x
ib
le
m
et
h
o
d
th
at
u
s
e
k
er
n
els
a
n
d
p
ar
am
e
ter
s
etti
n
g
s
w
h
ic
h
en
ab
le
th
e
lear
n
i
n
g
m
ec
h
a
n
i
s
m
to
b
etter
s
u
it
t
h
e
ch
ar
ac
ter
is
tic
s
o
f
d
if
f
er
e
n
t
c
h
u
n
k
s
o
f
d
ata,
w
h
ich
is
a
t
y
p
i
ca
l
ch
ar
ac
ter
is
ti
c
o
f
cr
o
s
s
-
co
m
p
a
n
y
d
ataset
s
.
I
n
o
r
d
er
to
au
to
m
atica
ll
y
s
elec
t
s
u
itab
le
SV
R
p
ar
a
m
e
ter
s
i
n
clu
d
in
g
t
h
e
k
er
n
e
l
f
u
n
ctio
n
,
th
e
a
u
th
o
r
s
i
n
[
3
3
]
p
r
o
p
o
s
ed
th
e
u
s
e
o
f
an
ap
p
r
o
ac
h
b
ased
o
n
T
ab
u
Sear
ch
(
T
S).
T
h
ey
ev
al
u
ated
e
m
p
ir
icall
y
t
h
e
p
r
o
p
o
s
ed
m
o
d
el
u
s
in
g
d
if
f
er
en
t
t
y
p
es
o
f
d
atasets
f
r
o
m
P
R
O
MI
SE
r
ep
o
s
ito
r
y
a
n
d
T
u
k
u
t
u
k
u
d
ataset.
T
h
eir
r
esu
lts
s
h
o
w
ed
th
at
SV
R
co
m
b
i
n
ed
w
it
h
T
S
s
i
g
n
i
f
ica
n
tl
y
o
u
tp
er
f
o
r
m
ed
C
B
R
an
d
m
an
u
al
s
tep
w
is
e
r
eg
r
es
s
io
n
m
eth
o
d
s
.
T
h
is
s
ec
tio
n
h
a
s
atte
m
p
ted
to
p
r
o
v
id
e
a
b
r
ief
s
u
m
m
ar
y
o
f
t
h
e
m
aj
o
r
liter
atu
r
e
r
elatin
g
to
s
o
f
t
w
ar
e
e
f
f
o
r
t e
s
ti
m
atio
n
u
s
in
g
s
u
p
p
o
r
t v
ec
to
r
r
e
g
r
ess
io
n
.
4.
SVR
M
O
DE
L
S WI
T
H
F
E
A
T
UR
E
SE
L
E
CT
I
O
N
M
E
T
H
O
DS
T
h
is
s
ec
tio
n
p
r
ese
n
ts
a
n
o
v
er
v
ie
w
o
f
t
h
e
t
w
o
S
VR
m
o
d
els
d
esig
n
ed
in
t
h
is
p
ap
er
n
a
m
el
y
SVR
w
i
th
b
ac
k
w
ar
d
f
ea
t
u
r
e
eli
m
in
a
tio
n
an
d
SV
R
w
i
t
h
B
o
r
u
ta
f
ea
t
u
r
e
s
elec
t
io
n
(
h
e
n
ce
f
o
r
th
S
V
R
-
B
FE
a
n
d
SV
R
-
B
OR
UT
A
r
esp
ec
tiv
el
y
)
an
d
il
lu
s
tr
ate
s
h
o
w
t
h
ese
m
o
d
el
s
w
e
r
e
tr
ain
ed
an
d
o
p
tim
ized
b
y
g
r
id
s
ea
r
ch
m
et
h
o
d
.
4
.
1
.
SVR
m
o
dels
w
it
h
ba
c
kw
a
rd
f
e
a
t
ure
eli
m
i
na
t
io
n
(
SVR
-
B
F
E
)
I
n
th
e
p
r
ep
r
o
ce
s
s
in
g
s
ta
g
e
o
f
t
h
i
s
m
o
d
el,
w
e
u
s
ed
a
s
i
m
p
ler
f
o
r
m
o
f
b
ac
k
w
ar
d
f
ea
tu
r
e
eli
m
i
n
atio
n
s
o
th
at
i
n
s
tead
o
f
iter
atin
g
t
h
e
b
ac
k
w
ar
d
eli
m
in
atio
n
p
r
o
ce
d
u
r
e
u
n
t
il
t
h
e
en
d
,
w
e
s
to
p
p
ed
th
is
p
r
o
ce
d
u
r
e
at
th
e
f
o
u
r
t
h
eli
m
in
a
tio
n
.
T
h
is
m
e
t
h
o
d
is
p
ar
ticu
lar
l
y
u
s
e
f
u
l
in
s
tu
d
y
in
g
th
e
ac
c
u
r
ac
y
o
f
th
e
m
o
d
el
af
ter
ea
ch
iter
atio
n
a
n
d
co
m
p
ar
in
g
t
h
e
r
esu
lt
s
o
b
tain
ed
w
i
th
th
e
B
o
r
u
ta
b
ased
SVR
.
Fo
llo
w
in
g
t
h
i
s
m
et
h
o
d
an
d
u
s
i
n
g
v
ar
iab
le
i
m
p
o
r
tan
ce
co
m
p
u
te
d
b
y
r
an
d
o
m
f
o
r
est,
f
o
u
r
s
u
b
s
ets
o
f
f
ea
t
u
r
es
w
er
e
g
e
n
er
ated
b
y
r
e
m
o
v
in
g
ea
ch
ti
m
e
t
h
e
least
i
m
p
o
r
tan
t
v
ar
ia
b
le.
So
,
in
th
e
f
ir
s
t
s
u
b
s
et
d
en
o
ted
B
FE_
1
,
w
e
eli
m
i
n
ate
t
h
e
least
s
ig
n
i
f
ica
n
t
f
ea
t
u
r
e
an
d
in
th
e
s
ec
o
n
d
s
u
b
s
et
B
FE_
2
,
w
e
r
e
m
o
v
ed
th
e
n
ex
t
least
i
m
p
o
r
tan
t
f
ea
tu
r
e
ac
co
r
d
in
g
to
v
ar
iab
le
i
m
p
o
r
tan
ce
r
a
n
k
i
n
g
a
n
d
s
o
o
n
.
Star
ti
n
g
f
r
o
m
t
h
ese
s
u
b
s
et
s
,
f
o
u
r
S
V
R
m
o
d
el
s
,
d
en
o
ted
SVR
-
B
FE_
i
w
er
e
o
p
tim
ized
u
s
in
g
g
r
id
s
ea
r
ch
o
p
ti
m
izatio
n
m
e
th
o
d
an
d
1
0
-
f
o
ld
cr
o
s
s
v
alid
atio
n
ap
p
r
o
ac
h
.
Fig
u
r
e
1
d
ep
icts
th
e
m
o
d
el
g
r
ap
h
icall
y
a
n
d
s
h
o
w
s
th
e
d
if
f
er
en
t
s
ta
g
es
o
f
SVR
m
o
d
el
b
u
ild
in
g
i
n
cl
u
d
in
g
f
ea
tu
r
e
s
elec
tio
n
s
tep
an
d
h
y
p
er
-
p
ar
a
m
eter
o
p
ti
m
izatio
n
s
tep
.
4
.
2
.
SVR
m
o
del
w
it
h
bo
rut
a
f
ea
t
ure
s
elec
t
io
n
m
et
ho
d
(
S
VR
-
B
O
RUTA)
T
h
is
SVR
m
o
d
el
is
co
m
p
o
s
ed
,
lik
e
th
e
f
ir
s
t
o
n
e,
f
r
o
m
o
n
e
p
r
ep
r
o
ce
s
s
in
g
s
ta
g
e
w
h
e
r
e
B
o
r
u
ta
alg
o
r
ith
m
is
p
er
f
o
r
m
ed
to
r
e
m
o
v
e
all
u
n
i
m
p
o
r
tan
t
f
ea
t
u
r
es
an
d
k
ee
p
o
n
l
y
th
e
r
ele
v
a
n
t
o
n
es.
Ne
x
t,
th
e
h
y
p
er
p
ar
a
m
eter
o
f
SV
R
m
o
d
el
(
C
,
µ)
w
er
e
ad
j
u
s
ted
b
y
th
e
s
a
m
e
p
r
o
ce
d
u
r
e
u
s
ed
f
o
r
SVR
w
ith
B
FE
i
n
o
r
d
er
to
ev
alu
ate
t
h
e
m
u
n
d
er
th
e
s
a
m
e
co
n
d
itio
n
s
.
Fi
g
u
r
e
1
ill
u
s
tr
ates th
e
m
o
d
el
b
u
ild
in
g
ar
ch
i
t
ec
tu
r
e.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8938
IJ
-
A
I
Vo
l.
8
,
No
.
4
,
Dec
em
b
er
201
9
:
3
9
9
–
410
404
Fig
u
r
e
1
.
A
r
ch
itectu
r
e
o
f
SVR
m
o
d
els
w
it
h
FS
S
4
.
3
.
P
a
ra
m
et
er
s
s
et
t
ing
I
t
is
w
ell
k
n
o
w
n
t
h
at
t
h
e
p
ar
a
m
eter
s
etti
n
g
s
co
u
ld
h
a
v
e
a
s
ig
n
i
f
ica
n
t
i
m
p
ac
t
o
n
t
h
e
esti
m
atio
n
ac
cu
r
ac
y
o
f
tr
ai
n
ed
SDEE
tech
n
iq
u
es.
T
h
er
ef
o
r
e,
b
u
ild
in
g
an
ac
cu
r
ate
m
o
d
el
r
eq
u
ir
es
s
elec
tio
n
o
f
o
p
ti
m
al
v
alu
e
s
o
f
its
lear
n
i
n
g
p
ar
a
m
eter
s
[
1
6
]
.
Ho
w
e
v
er
,
f
i
n
d
i
n
g
o
p
ti
m
al
v
a
lu
e
s
is
co
m
p
licat
ed
task
a
n
d
v
ar
io
u
s
ap
p
r
o
ac
h
es
h
av
e
b
ee
n
p
r
o
p
o
s
ed
in
th
e
liter
atu
r
e
to
ad
d
r
ess
th
is
i
s
s
u
e,
s
u
c
h
as
g
r
id
s
ea
r
ch
(
GS)
[
4
2
]
,
p
ar
ticle
s
w
ar
m
o
p
ti
m
izatio
n
(
P
SO)
[
4
3
]
an
d
g
en
etic
al
g
o
r
ith
m
(
G
A)
[
3
9
]
.
I
n
o
r
d
er
to
en
ab
le
SVR
m
o
d
els,
d
ev
elo
p
ed
in
t
h
i
s
s
tu
d
y
,
to
ac
h
ie
v
e
a
h
i
g
h
er
p
r
ed
ictio
n
ac
cu
r
ac
y
o
v
er
t
h
e
eig
h
t d
ata
s
ets
u
s
ed
i
n
S
D
E
E
,
w
e
e
m
p
lo
y
ed
g
r
id
s
ea
r
ch
(
GS)
a
s
o
p
ti
m
izatio
n
m
et
h
o
d
co
m
b
in
ed
w
i
th
cr
o
s
s
-
v
alid
atio
n
p
r
o
ce
d
u
r
e.
T
h
e
m
ain
id
ea
b
e
h
in
d
t
h
e
g
r
id
s
ea
r
ch
m
et
h
o
d
is
t
h
at
d
if
f
er
e
n
t
p
air
s
o
f
p
ar
a
m
eter
s
ar
e
test
ed
an
d
t
h
e
o
n
e
w
it
h
t
h
e
h
ig
h
es
t
cr
o
s
s
v
alid
atio
n
ac
c
u
r
ac
y
i
s
s
elec
te
d
.
T
h
e
m
aj
o
r
ad
v
an
tag
e
o
f
G
S
m
et
h
o
d
is
its
h
i
g
h
lear
n
i
n
g
ac
cu
r
ac
y
a
n
d
th
e
ab
ilit
y
o
f
p
ar
allel
p
r
o
ce
s
s
in
g
o
n
th
e
tr
ain
i
n
g
o
f
e
v
er
y
SV
R
,
b
ec
au
s
e
t
h
e
y
ar
e
in
d
ep
en
d
en
t
o
f
ea
c
h
o
th
er
.
A
lt
h
o
u
g
h
G
S
m
eth
o
d
ca
n
f
in
d
th
e
o
p
ti
m
u
m
p
ar
a
m
e
ter
s
,
t
h
e
co
m
p
u
tatio
n
al
co
m
p
le
x
it
y
i
s
v
er
y
b
ig
o
b
v
io
u
s
l
y
,
an
d
t
h
e
ti
m
e
s
p
en
t
i
s
v
er
y
lar
g
e,
esp
ec
iall
y
f
o
r
lar
g
e
s
a
m
p
l
e
d
ata.
I
n
o
u
r
ca
s
e,
w
e
li
m
ited
th
e
s
ea
r
c
h
s
p
ac
e
to
m
o
s
t p
r
o
m
is
in
g
v
a
lu
e
s
g
u
id
ed
b
y
p
r
ev
io
u
s
s
tu
d
ie
s
[
4
4
]
.
T
a
b
le
1
s
h
o
w
s
GS
p
ar
a
m
eter
f
o
r
S
VR
m
o
d
els
.
T
ab
le
1
.
Gr
id
s
ea
r
ch
p
ar
am
e
te
r
f
o
r
SVR
m
o
d
els
T
e
c
h
n
i
q
u
e
s
P
a
r
a
me
t
e
r
s
S
V
R
Ty
p
e
=
{
µ
-
r
e
g
r
e
ssi
o
n
}
K
e
r
n
e
l
f
u
n
c
t
i
o
n
=
{
R
B
F
}
C
o
mp
l
e
x
i
t
y
=
{f
r
o
m 0
,
0
0
5
t
o
0
,
1
,
s
t
e
p
=
0
,
0
0
5
}
K
e
r
n
e
l
p
a
r
a
me
t
e
r
=
{1
/
n
u
mb
e
r
o
f
f
e
a
t
u
r
e
s}
µ
=
{
0
,
1
t
o
1
,
0
,
s
t
e
p
=
0
,
1
}
T
h
e
GS
m
eth
o
d
f
i
n
d
s
t
h
e
b
est
co
n
f
i
g
u
r
atio
n
o
f
SV
R
m
o
d
el
s
b
y
ev
a
lu
at
in
g
e
v
er
y
p
o
s
s
ib
l
e
co
m
b
i
n
atio
n
o
f
T
ab
le
2
w
i
th
r
esp
ec
t
to
m
ea
n
s
q
u
ar
e
er
r
o
r
(
MSE
)
b
ased
er
r
o
r
f
u
n
ctio
n
u
s
in
g
1
0
-
f
o
ld
cr
o
s
s
-
v
alid
atio
n
ap
p
r
o
ac
h
.
T
h
e
b
est co
n
f
i
g
u
r
atio
n
o
f
ea
ch
tec
h
n
iq
u
e
th
at
m
i
n
i
m
izes
M
SE
is
th
e
n
s
elec
ted
.
No
te
th
at
th
e
s
a
m
e
r
an
g
e
o
f
p
ar
a
m
eter
v
alu
e
s
w
er
e
u
s
ed
f
o
r
SVR
w
it
h
B
o
r
u
ta
f
ea
tu
r
e
s
elec
tio
n
m
et
h
o
d
.
R
eg
ar
d
in
g
th
e
p
ar
am
eter
s
o
f
r
an
d
o
m
f
o
r
est
f
ea
t
u
r
e
s
elec
tio
n
an
d
B
o
r
u
ta
alg
o
r
ith
m
w
er
e
ad
j
u
s
ted
as
s
h
o
w
n
i
n
T
ab
le
3
.
I
n
f
ac
t,
t
h
e
s
e
p
ar
a
m
e
ter
s
d
o
n
o
t
h
av
e
a
s
i
g
n
i
f
ica
n
t
i
m
p
ac
t
o
n
v
ar
iab
le
i
m
p
o
r
tan
ce
r
a
n
k
i
n
g
e
x
ce
p
t
m
a
x
R
u
n
s
p
ar
am
eter
o
f
B
o
r
u
ta
m
et
h
o
d
w
h
ic
h
s
h
o
u
ld
b
e
in
cr
ea
s
ed
i
n
ce
r
tain
ca
s
e
to
r
eso
lv
e
attr
ib
u
t
es
le
f
t
T
en
tati
v
e
b
y
th
e
alg
o
r
it
h
m
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
-
AI
I
SS
N:
2252
-
8938
I
mp
r
o
vin
g
s
o
ftw
a
r
e
d
ev
elo
p
m
en
t e
ffo
r
t e
s
tima
tio
n
u
s
in
g
s
u
p
p
o
r
t v
ec
to
r
r
eg
r
ess
io
n
a
n
d
... (
A
b
d
ela
li Za
kra
n
i
)
405
T
ab
le
2
.
P
ar
am
eter
s
u
s
ed
f
o
r
R
F a
n
d
B
o
r
u
ta
Feat
u
r
e
Selecti
o
n
Me
th
o
d
s
M
e
t
h
o
d
P
a
r
a
me
t
e
r
D
e
scri
p
t
i
o
n
R
a
n
d
o
m
F
o
r
e
st
n
t
r
e
e
=
5
0
0
0
mt
r
y
=
5
N
u
mb
e
r
o
f
t
r
e
e
s
N
u
mb
e
r
o
f
v
a
r
i
a
b
l
e
s se
l
e
c
t
e
d
a
t
e
a
c
h
sp
l
i
t
B
o
r
u
t
a
p
V
a
l
u
e
=
0
.
0
1
max
R
u
n
s
=
5
0
0
C
o
n
f
i
d
e
n
c
e
L
e
v
e
l
M
a
x
i
m
a
l
n
u
m
b
e
r
o
f
i
mp
o
r
t
a
n
c
e
so
u
r
c
e
r
u
n
s.
5.
E
XP
E
R
I
M
E
NT
A
L
DE
SI
G
N
T
h
is
s
ec
tio
n
p
r
ese
n
ts
t
h
e
e
x
p
er
i
m
en
tal
d
esi
g
n
o
f
th
is
s
t
u
d
y
in
c
lu
d
i
n
g
:
(
1
)
th
e
ac
cu
r
ac
y
m
ea
s
u
r
e
s
u
s
ed
to
ev
al
u
ate
th
e
p
r
o
p
o
s
ed
SVR
m
o
d
el
s
,
(
2
)
th
e
d
escr
ip
tio
n
o
f
th
e
d
ataset
s
u
s
ed
,
an
d
(
3
)
th
e
ex
p
er
i
m
en
tal
p
r
o
ce
s
s
f
o
llo
w
ed
to
co
n
s
tr
u
ct
an
d
co
m
p
ar
e
th
e
d
if
f
er
en
t
SV
R
m
o
d
els.
5
.
1
.
Acc
ura
cy
m
ea
s
ure
s
W
e
e
m
p
lo
y
t
h
e
f
o
llo
w
in
g
cr
iter
ia
to
ass
e
s
s
a
n
d
co
m
p
ar
e
th
e
ac
c
u
r
ac
y
o
f
t
h
e
e
f
f
o
r
t
esti
m
atio
n
m
o
d
el
s
.
A
co
m
m
o
n
cr
iter
io
n
f
o
r
th
e
ev
alu
at
io
n
o
f
ef
f
o
r
t
esti
m
atio
n
m
o
d
el
s
is
m
a
g
n
itu
d
e
o
f
r
elativ
e
er
r
o
r
(
MRE)
,
w
h
ich
i
s
d
ef
i
n
ed
as
=
|
(
−
)
|
(
8
)
T
h
e
MRE
v
al
u
es
ar
e
ca
lcu
lat
ed
f
o
r
ea
ch
p
r
o
j
ec
t
in
th
e
d
at
aset,
w
h
ile
m
ea
n
m
a
g
n
it
u
d
e
o
f
r
elati
v
e
er
r
o
r
(
MM
R
E
)
c
o
m
p
u
tes t
h
e
a
v
er
ag
e
o
v
er
N
p
r
o
j
ec
ts
as f
o
llo
w
s
:
=
1
∑
=
1
(
9
)
Gen
er
all
y
,
t
h
e
ac
ce
p
tab
le
tar
g
et
v
al
u
e
f
o
r
MM
R
E
is
2
5
%.
T
h
is
in
d
icate
s
th
a
t
o
n
th
e
a
v
er
ag
e,
th
e
ac
cu
r
ac
y
o
f
t
h
e
e
s
tab
lis
h
ed
es
ti
m
atio
n
m
o
d
els
w
o
u
ld
b
e
les
s
th
a
n
2
5
%.
A
n
o
t
h
er
w
id
el
y
u
s
ed
cr
iter
io
n
is
t
h
e
P
r
ed
(
l
)
w
h
ic
h
r
ep
r
esen
t
s
t
h
e
p
er
ce
n
tag
e
o
f
M
R
E
t
h
at
is
le
s
s
th
a
n
o
r
eq
u
al
to
th
e
v
al
u
e
l
a
m
o
n
g
all
p
r
o
j
ec
ts
.
T
h
is
m
ea
s
u
r
e
is
o
f
ten
u
s
ed
in
th
e
liter
at
u
r
e
an
d
is
t
h
e
p
r
o
p
o
r
tio
n
o
f
t
h
e
p
r
o
j
ec
ts
f
o
r
a
g
iv
en
lev
el
o
f
ac
c
u
r
ac
y
.
T
h
e
d
ef
in
itio
n
o
f
P
r
ed
(
l)
is
g
iv
en
as
f
o
llo
w
s
:
(
)
=
(
1
0
)
W
h
er
e
N
is
th
e
to
tal
n
u
m
b
er
o
f
o
b
s
er
v
atio
n
s
a
n
d
k
is
t
h
e
n
u
m
b
er
o
f
o
b
s
er
v
atio
n
s
w
h
o
s
e
MRE
is
le
s
s
o
r
eq
u
al
to
l.
A
co
m
m
o
n
v
al
u
e
f
o
r
l
is
0
.
2
5
,
w
h
ic
h
is
also
u
s
e
d
in
th
e
p
r
esen
t
s
tu
d
y
.
T
h
e
P
r
e
d
(
0
.
2
5
)
r
ep
r
esen
ts
th
e
p
er
ce
n
ta
g
e
o
f
p
r
o
j
ec
ts
w
h
o
s
e
MRE
i
s
les
s
o
r
eq
u
al
to
2
5
%.
T
h
e
P
r
ed
(
0
.
2
5
)
v
alu
e
id
en
tifie
s
t
h
e
ef
f
o
r
t
esti
m
ates
t
h
at
ar
e
g
en
er
all
y
ac
cu
r
ate
w
h
er
ea
s
th
e
MM
R
E
is
f
air
l
y
co
n
s
er
v
ati
v
e
with
a
b
ias
ag
ai
n
s
t
o
v
er
esti
m
ates
[
4
5
-
4
6
]
.
Fo
r
th
is
r
ea
s
o
n
,
Md
MRE
h
as
b
ee
n
also
u
s
ed
as
a
n
o
th
er
cr
iter
i
o
n
s
i
n
ce
it
is
le
s
s
s
en
s
iti
v
e
to
o
u
tlier
s
(
1
0
)
.
=
(
)
(
1
1
)
5
.
2
.
Da
t
a
s
et
s
Fo
r
th
is
s
tu
d
y
,
e
ig
h
t
d
ataset
s
,
co
llected
f
r
o
m
d
if
f
er
en
t
o
r
g
an
izatio
n
s
a
n
d
co
u
n
tr
ies,
w
er
e
s
elec
ted
to
ev
alu
a
te
th
e
p
er
f
o
r
m
a
n
ce
o
f
SVR
an
d
SV
R
-
R
F
tec
h
n
iq
u
es.
A
to
tal
o
f
1
1
1
9
p
r
o
j
ec
ts
w
er
e
u
s
ed
f
r
o
m
th
r
ee
s
o
u
r
ce
s
:
9
1
5
p
r
o
j
ec
ts
ca
m
e
f
r
o
m
s
ix
d
atasets
o
f
P
R
O
MI
SE
d
ata
r
ep
o
s
ito
r
y
w
h
ic
h
i
s
a
p
u
b
licl
y
a
v
ailab
le
o
n
li
n
e
d
ata
r
ep
o
s
ito
r
y
(
Me
n
zie
s
et
al
.
2
0
1
2
)
n
a
m
el
y
:
A
lb
r
ec
h
t,
C
O
C
OM
O8
1
,
C
h
in
a,
De
s
h
ar
n
ais,
Ke
m
er
er
an
d
Mi
y
az
a
k
i d
atasets
.
1
5
1
p
r
o
j
ec
ts
s
elec
ted
f
r
o
m
I
SB
SG
R
8
r
ep
o
s
ito
r
y
.
I
n
f
ac
t
,
th
is
r
ep
o
s
ito
r
y
co
n
ta
in
s
m
o
r
e
th
an
2
0
0
0
s
o
f
t
w
ar
e
p
r
o
j
ec
ts
d
escr
ib
e
d
b
y
m
o
r
e
th
a
n
5
0
n
u
m
er
ical
an
d
ca
teg
o
r
ical
attr
ib
u
tes.
T
h
e
s
el
ec
ted
p
r
o
j
ec
ts
ar
e
th
e
r
es
u
lt
s
o
f
a
d
ata
p
r
e
-
p
r
o
ce
s
s
in
g
s
tu
d
y
co
n
d
u
c
ted
b
y
[
4
7
]
,
th
e
o
b
j
ec
tiv
e
o
f
w
h
ich
w
a
s
to
s
elec
t
d
ata
(
p
r
o
j
ec
ts
an
d
attr
ib
u
te
s
)
,
in
o
r
d
er
to
r
etain
p
r
o
j
ec
ts
w
i
th
h
i
g
h
q
u
alit
y
.
T
h
e
f
ir
s
t
s
tep
o
f
t
h
i
s
s
t
u
d
y
w
a
s
to
s
elec
t
o
n
l
y
t
h
e
n
e
w
d
ev
elo
p
m
e
n
t
p
r
o
j
ec
ts
w
it
h
h
i
g
h
q
u
a
lit
y
d
ata
a
n
d
u
s
i
n
g
I
FP
UG
co
u
n
ti
n
g
ap
p
r
o
ac
h
.
T
h
e
s
ec
o
n
d
s
tep
w
a
s
co
n
ce
r
n
ed
b
y
s
elec
ti
n
g
a
n
o
p
ti
m
al
s
u
b
s
et
o
f
n
u
m
er
ical
attr
ib
u
tes
t
h
at
ar
e
r
elev
an
t to
e
f
f
o
r
t e
s
t
i
m
at
io
n
a
n
d
m
o
s
t a
p
p
r
o
p
r
iate
to
u
s
e
as e
f
f
o
r
t d
r
iv
er
s
i
n
e
m
p
ir
ic
a
l st
u
d
ies.
5
3
W
eb
p
r
o
j
ec
ts
f
r
o
m
T
u
k
u
t
u
k
u
d
atase
t
[
4
8
]
.
E
ac
h
W
eb
ap
p
licatio
n
is
d
escr
ib
ed
u
s
i
n
g
9
n
u
m
er
ical
attr
ib
u
tes
s
u
ch
as
:
th
e
n
u
m
b
er
o
f
h
t
m
l
o
r
s
h
t
m
l
f
ile
s
u
s
ed
,
th
e
n
u
m
b
er
o
f
m
ed
ia
f
il
es
an
d
tea
m
ex
p
er
ien
ce
.
Ho
w
e
v
er
,
ea
ch
p
r
o
j
ec
t
v
o
lu
n
teer
ed
to
t
h
e
T
u
k
u
t
u
k
u
d
atab
ase
w
a
s
i
n
itial
l
y
ch
ar
ac
ter
ized
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8938
IJ
-
A
I
Vo
l.
8
,
No
.
4
,
Dec
em
b
er
201
9
:
3
9
9
–
410
406
u
s
i
n
g
m
o
r
e
th
a
n
9
s
o
f
t
w
ar
e
attr
ib
u
tes,
b
u
t
s
o
m
e
o
f
t
h
e
m
w
er
e
g
r
o
u
p
ed
to
g
et
h
er
.
Fo
r
ex
a
m
p
le,
w
e
g
r
o
u
p
ed
to
g
et
h
er
th
e
f
o
llo
w
i
n
g
th
r
ee
at
tr
ib
u
tes:
n
u
m
b
er
o
f
n
e
w
W
eb
p
ag
es
d
e
v
elo
p
ed
b
y
th
e
tea
m
,
n
u
m
b
er
o
f
W
eb
p
ag
e
s
p
r
o
v
id
ed
b
y
th
e
c
u
s
to
m
er
a
n
d
th
e
n
u
m
b
er
o
f
W
eb
p
ag
es
d
ev
e
lo
p
ed
b
y
a
th
ir
d
p
ar
ty
(
o
u
ts
o
u
r
ce
d
)
in
o
n
e
attr
ib
u
te
r
ef
lecti
n
g
th
e
to
tal
n
u
m
b
er
o
f
W
eb
p
ag
es
in
th
e
ap
p
licatio
n
(
W
eb
p
ag
es).
T
ab
le
3
s
u
m
m
ar
izes
d
escr
ip
ti
v
e
s
tatis
tics
o
f
th
e
s
elec
ted
d
atasets
,
in
cl
u
d
i
n
g
s
ize
o
f
d
ataset,
e
f
f
o
r
t
u
n
i
t,
n
u
m
b
er
o
f
at
tr
ib
u
tes,
m
e
d
ian
,
m
ea
n
,
m
i
n
i
m
u
m
,
m
a
x
i
m
u
m
,
s
k
e
w
n
e
s
s
a
n
d
k
u
r
to
s
is
o
f
ef
f
o
r
t.
No
n
e
o
f
t
h
e
s
elec
ted
d
atasets
h
ad
a
n
o
r
m
al
l
y
d
is
tr
ib
u
ted
ef
f
o
r
t
as
s
k
e
w
n
e
s
s
v
al
u
es
r
an
g
ed
f
r
o
m
2
.
0
4
to
6
.
2
6
.
T
h
is
p
r
esen
ts
a
ch
allen
g
e
f
o
r
r
esear
ch
er
s
att
e
m
p
tin
g
to
b
u
ild
ac
cu
r
ate
S
D
E
E
tech
n
iq
u
es [
1
6
,
4
9
]
.
T
ab
le
3
.
Descr
ip
tiv
e
s
tatis
tic
s
o
f
th
e
ei
g
h
t d
atasets
D
a
t
a
se
t
#
o
f
so
f
t
w
a
r
p
r
o
j
e
c
t
U
n
i
t
#
o
f
f
e
a
t
u
r
e
s
D
i
st
r
i
b
u
t
i
o
n
o
f
Ef
f
o
r
t
Mi
n
Ma
x
Me
a
n
Me
d
i
a
n
S
k
e
w
n
e
ss
K
u
r
t
o
s
i
s
I
S
B
S
G
(
R
8
)
1
5
1
M
a
n
/
h
o
u
r
s
6
24
6
0
2
7
0
5
0
3
9
2
4
4
9
4
.
1
7
2
1
.
1
0
C
O
C
O
M
O
2
5
2
M
a
n
/
mo
n
t
h
s
13
6
1
1
4
0
0
6
8
3
.
4
98
4
.
3
9
2
0
.
5
0
T
U
K
U
T
U
K
U
53
M
a
n
/
mo
n
t
h
s
9
6
5
0
0
0
4
1
4
.
8
5
1
0
5
4
.
2
1
2
0
.
1
7
D
ESH
A
R
N
A
I
S
77
M
a
n
/
h
o
u
r
s
8
5
4
6
2
3
9
4
0
4
8
3
4
3
5
4
2
2
.
0
4
5
.
3
0
A
L
B
R
EC
H
T
24
M
a
n
/
mo
n
t
h
s
7
0
.
5
1
0
5
.
2
0
2
1
.
8
8
1
1
.
4
5
2
.
3
0
4
.
6
7
K
EM
ER
ER
15
M
a
n
/
mo
n
t
h
s
6
23
1
1
0
7
2
1
9
.
2
4
1
3
0
3
.
0
7
1
0
.
6
M
I
Y
A
ZA
K
I
48
M
a
n
/
mo
n
t
h
s
8
5
.
6
1
5
8
6
8
7
.
4
7
38
6
.
2
6
4
1
.
3
C
H
I
N
A
4
9
9
M
a
n
/
h
o
u
r
s
15
26
5
4
6
2
0
3
9
2
1
.
0
4
1
8
2
9
3
.
9
2
1
9
.
3
5
.
3
.
Va
lid
a
t
i
o
n
m
et
ho
d
A
3
0
%
h
o
ld
o
u
t
v
alid
atio
n
m
et
h
o
d
w
a
s
u
s
ed
to
ev
al
u
at
e
th
e
g
en
er
aliza
t
io
n
ab
ilit
y
o
f
th
e
e
s
ti
m
atio
n
m
o
d
el
s
.
So
,
th
e
d
atasets
w
er
e
s
p
lit
r
an
d
o
m
l
y
i
n
to
t
w
o
n
o
n
-
o
v
er
lap
p
in
g
s
ets:
tr
ain
i
n
g
s
et
co
n
tain
i
n
g
7
0
%
o
f
d
ata
an
d
test
in
g
s
et
co
m
p
o
s
ed
f
r
o
m
3
0
%
o
f
th
e
r
em
a
in
i
n
g
d
ata.
T
h
e
p
u
r
p
o
s
e
o
f
h
o
ld
o
u
t
e
v
alu
a
tio
n
is
to
test
a
m
o
d
el
o
n
d
if
f
er
en
t
d
ata
to
t
h
at
f
r
o
m
w
h
ich
i
t
is
lear
n
ed
.
T
h
is
p
r
o
v
id
es
less
b
iased
e
s
ti
m
ate
o
f
lear
n
in
g
p
er
f
o
r
m
a
n
ce
t
h
an
al
l
-
i
n
e
v
alu
a
tio
n
m
et
h
o
d
.
6.
E
M
P
I
RICAL
R
E
SU
L
T
S
T
h
is
s
ec
tio
n
r
ep
o
r
ts
a
n
d
d
is
c
u
s
s
es
t
h
e
r
es
u
lt
s
o
f
e
m
p
ir
ical
e
x
p
er
i
m
e
n
ts
p
er
f
o
r
m
ed
u
s
in
g
S
VR
m
o
d
els
d
esig
n
ed
in
Sectio
n
I
V
an
d
f
o
l
lo
w
i
n
g
th
e
b
u
ild
in
g
p
r
o
ce
s
s
ill
u
s
tr
ated
i
n
Fi
g
u
r
e
1
.
T
o
ca
r
r
y
o
u
t th
e
s
e
e
m
p
ir
ical
ex
p
er
i
m
e
n
ts
,
d
if
f
er
en
t
R
p
ac
k
ag
es
w
er
e
u
s
ed
to
d
ev
elo
p
an
R
p
r
o
to
ty
p
e
e
m
p
lo
y
e
d
to
co
n
s
tr
u
ct
th
e
p
r
o
p
o
s
ed
m
o
d
el
s
.
I
n
t
h
is
w
a
y
,
e1
0
7
1
p
ac
k
ag
e
w
as
u
s
ed
to
b
u
ild
t
h
e
S
VR
m
o
d
els
a
n
d
r
an
d
o
m
Fo
r
est,
an
d
B
o
r
u
t
a
p
ac
k
ag
es
w
er
e
u
s
ed
f
o
r
f
ea
t
u
r
e
s
elec
tio
n
m
et
h
o
d
s
.
6
.
1
.
F
e
a
t
ure
s
elec
t
io
n r
esu
lt
s
T
h
is
s
u
b
s
ec
t
io
n
p
r
esen
ts
t
h
e
r
esu
lt
s
o
f
t
h
e
p
r
ep
r
o
ce
s
s
in
g
s
tep
.
T
ab
le
4
p
r
o
v
id
es
th
e
f
o
u
r
leas
t
i
m
p
o
r
tan
t
f
ea
tu
r
es
g
e
n
er
ated
b
y
r
an
d
o
m
f
o
r
est,
an
d
th
e
n
u
m
b
er
o
f
s
e
lecte
d
f
ea
t
u
r
es
an
d
r
em
o
v
ed
o
n
es
b
y
B
o
r
u
ta
m
e
th
o
d
in
ea
c
h
d
ataset
.
I
t
ca
n
b
e
s
ee
n
f
r
o
m
t
h
e
d
a
ta
in
T
ab
le
4
th
at
th
e
f
ea
tu
r
es
r
ej
ec
ted
b
y
B
o
r
u
ta
ar
e
g
en
er
all
y
a
m
o
n
g
t
h
e
f
o
u
r
lea
s
t
i
m
p
o
r
ta
n
t
f
ea
t
u
r
e
id
en
ti
f
ied
b
y
r
an
d
o
m
f
o
r
est,
w
h
ic
h
i
s
n
o
t
s
u
r
p
r
is
in
g
s
i
n
ce
B
o
r
u
ta
alg
o
r
ith
m
is
b
ased
o
n
R
F
v
ar
iab
le
i
m
p
o
r
tan
ce
.
Ho
w
ev
er
,
B
o
r
u
ta
m
eth
o
d
d
id
n
o
t
al
w
a
y
s
r
e
m
o
v
e
t
h
e
f
ir
s
t
leas
t
i
m
p
o
r
tan
t
f
ea
tu
r
e.
As
ex
a
m
p
le,
f
o
r
A
lb
r
ec
h
t
d
at
aset,
it
r
em
o
v
ed
th
e
s
ec
o
n
d
o
n
e
(
in
p
u
t)
w
h
ile
th
e
f
ir
s
t
least
i
m
p
o
r
ta
n
t
f
ea
t
u
r
e
is
FP
A
d
j.
C
o
n
ce
r
n
i
n
g
t
h
e
n
u
m
b
er
o
f
t
h
e
s
elec
ted
f
ea
t
u
r
es,
B
o
r
u
ta
m
eth
o
d
s
elec
ted
al
m
o
s
t
at
least
5
0
%
o
f
f
ea
t
u
r
es
a
v
ailab
le
in
ea
c
h
d
ataset.
T
h
e
o
n
ly
e
x
ce
p
tio
n
w
a
s
th
e
ca
s
e
o
f
T
u
k
u
t
u
k
u
d
ata
s
et
f
o
r
w
h
ic
h
o
u
t
o
f
n
i
n
e
f
ea
t
u
r
es,
B
o
r
u
ta
s
el
ec
ted
o
n
l
y
t
w
o
f
ea
t
u
r
es.
T
h
e
s
in
g
le
m
o
s
t
s
tr
ik
i
n
g
r
esu
lt
to
e
m
er
g
e
f
r
o
m
t
h
e
d
at
a
is
t
h
at
all
f
ea
t
u
r
es
o
f
C
O
C
OM
O
d
ataset
w
er
e
d
ee
m
e
d
r
elev
an
t
a
n
d
n
o
n
e
o
f
th
e
m
w
as r
ej
ec
ted
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
-
AI
I
SS
N:
2252
-
8938
I
mp
r
o
vin
g
s
o
ftw
a
r
e
d
ev
elo
p
m
en
t e
ffo
r
t e
s
tima
tio
n
u
s
in
g
s
u
p
p
o
r
t v
ec
to
r
r
eg
r
ess
io
n
a
n
d
... (
A
b
d
ela
li Za
kra
n
i
)
407
T
ab
le
4
.
Nu
m
b
er
o
f
s
elec
ted
f
ea
tu
r
es a
n
d
r
e
m
o
v
ed
o
n
es i
n
e
ac
h
d
ataset
D
a
t
a
se
t
s
#
o
f
F
e
a
t
.
F
o
u
r
l
e
a
st
i
m
p
o
r
t
a
n
t
f
e
a
t
u
r
e
R
F
(
1
,
2
,
3
,
4
)
B
o
r
u
t
a
#
o
f
se
l
e
c
t
e
d
F
e
a
t
u
r
e
s
R
e
mo
v
e
d
f
e
a
t
u
r
e
I
S
B
S
G
(
R
8
)
6
B
u
s
i
n
e
ss,
L
o
c
a
t
i
o
n
s,
F
a
c
t
o
r
,
C
o
n
c
u
r
r
e
n
t
3
F
a
c
t
o
r
s,
B
u
si
n
e
ss,
L
o
c
a
t
i
o
n
s
C
O
C
O
M
O
13
V
EX
P
,
V
I
R
T
maj
e
u
r
,
L
EX
P
,
V
I
R
T
mi
n
e
u
r
13
-
T
U
K
U
T
U
K
U
9
A
u
d
i
o
,
T
e
a
me
x
p
,
t
o
t
_
n
h
i
g
h
,
A
N
I
M
2
d
e
v
Te
a
m,
t
e
a
mEx
p
,
t
e
x
t
P
,
i
mag
,
a
n
i
m,
a
u
d
i
o
,
t
o
t
_
n
h
i
g
h
D
ESH
A
R
N
A
IS
8
T
e
a
mEx
p
,
M
a
n
a
g
e
r
Ex
p
,
L
a
n
g
u
a
g
e
,
E
n
v
e
r
g
u
r
e
6
T
e
a
mEx
p
,
M
a
n
g
e
r
Ex
A
L
B
R
EC
H
T
7
F
P
A
d
j
,
I
n
p
u
t
,
I
n
q
u
i
r
y
,
F
i
l
e
6
I
n
p
u
t
K
EM
ER
ER
6
L
a
n
g
u
a
g
e
,
H
a
r
d
w
a
r
e
,
D
u
r
a
t
i
o
n
,
R
A
W
F
P
3
L
a
n
g
u
a
g
e
,
H
a
r
d
w
a
r
e
,
D
u
r
a
t
i
o
n
,
M
I
Y
A
ZA
K
I
7
EF
O
R
M
,
ES
C
R
N
,
F
I
L
E,
S
C
R
N
6
EF
O
R
M
C
H
I
N
A
15
D
e
v
.
Ty
p
e
,
D
e
l
e
t
e
d
,
C
h
a
n
g
e
d
,
R
e
so
u
r
c
e
13
D
e
l
e
t
e
d
,
D
e
v
.
Ty
p
e
6
.
2
.
E
v
a
lua
t
io
n
o
f
SVR
w
it
h
F
SS
T
h
e
s
ec
o
n
d
s
tep
o
f
th
e
m
o
d
el
b
u
ild
in
g
p
r
o
ce
s
s
u
s
es
t
h
e
o
r
ig
in
al
a
n
d
th
e
r
ed
u
ce
d
d
atasets
to
d
eter
m
in
e
t
h
e
b
est
s
etu
p
o
f
t
h
e
p
r
o
p
o
s
ed
SVR
m
o
d
els.
T
h
e
b
est
co
n
f
ig
u
r
atio
n
i
s
d
eter
m
i
n
ed
,
as
ex
p
lain
ed
ea
r
lier
,
b
y
a
s
ea
r
c
h
g
r
id
to
m
i
n
i
m
ize
t
h
e
m
ea
n
s
q
u
ar
e
er
r
o
r
(
MSE
)
.
On
ce
th
e
f
i
v
e
S
VR
m
o
d
els
w
er
e
tr
ai
n
ed
u
s
i
n
g
tr
ain
i
n
g
s
e
ts
(
7
0
%
o
f
d
ata)
,
w
e
ev
al
u
ated
th
e
g
e
n
er
a
lizatio
n
ca
p
ab
ilit
y
o
f
t
h
e
f
i
v
e
co
n
f
i
g
u
r
atio
n
s
o
f
SVR
m
o
d
els
u
s
in
g
test
i
n
g
s
e
ts
(
3
0
%)
o
v
er
t
h
e
ei
g
h
t
d
ata
s
ets.
T
h
e
e
v
al
u
atio
n
w
as
b
as
ed
o
n
th
e
MM
R
E
,
Md
MR
E
,
an
d
P
r
ed
(
0
.
2
5
)
cr
it
er
ia.
T
h
e
co
m
p
lete
e
m
p
ir
ical
r
esu
lt
s
o
b
tain
ed
ar
e
s
h
o
w
n
i
n
T
ab
les
5
-
8
.
Fro
m
d
ata
in
T
ab
le
5
,
w
e
n
o
tice
th
a
t
n
o
SV
R
co
n
f
i
g
u
r
atio
n
g
av
e
t
h
e
b
est
P
r
ed
(
0
.
2
5
)
v
alu
e
i
n
all
d
atasets
.
Ho
w
e
v
er
,
SVR
-
B
FE_
1
(
r
e
m
o
v
i
n
g
o
n
l
y
th
e
lea
s
t
i
m
p
o
r
tan
t
f
ea
tu
r
e)
g
en
er
ated
th
e
b
est
P
r
ed
in
6
o
u
t
o
f
8
d
ataset
s
a
n
d
SVR
-
B
FE_
4
o
n
l
y
ca
m
e
s
ec
o
n
d
b
y
g
i
v
in
g
b
est
v
a
lu
e
o
f
P
r
ed
in
5
d
atasets
.
T
h
e
SVR
w
i
th
o
u
t
F
S
an
d
SV
R
w
it
h
B
o
r
u
ta
m
eth
o
d
p
r
o
d
u
ce
d
b
est
v
al
u
e
o
f
P
r
ed
o
n
l
y
in
o
n
e
d
ataset
:
Ke
m
er
er
a
n
d
C
h
i
n
a
r
esp
ec
tiv
el
y
.
T
h
e
b
est v
al
u
e
o
f
P
r
ed
(
0
.
2
5
)
w
as o
b
tain
ed
b
y
SVR
-
B
FE_
4
in
C
h
i
n
a
d
ataset
(
8
3
.
3
3
)
.
T
h
e
r
esu
lt
s
r
ep
o
r
ted
in
T
ab
le
6
an
d
T
ab
le
7
r
elate
d
to
MM
R
E
a
n
d
Md
MRE
m
ea
s
u
r
e
s
c
o
n
f
ir
m
t
h
e
f
ac
t
t
h
at
n
o
SV
R
co
n
f
i
g
u
r
atio
n
p
er
f
o
r
m
ed
b
etter
t
h
an
th
e
o
th
er
i
n
all
s
itu
a
tio
n
.
Ne
v
e
r
th
e
less
,
w
e
ca
n
ea
s
il
y
o
b
s
er
v
e
th
at
t
h
e
b
est
v
alu
e
s
o
f
MM
R
E
an
d
Md
MRE
ar
e
o
b
tain
ed
w
it
h
s
a
m
e
d
ataset
s
as
t
h
o
s
e
o
f
P
r
ed
.
So
,
th
e
lo
w
es
t
er
r
o
r
s
w
er
e
o
b
tain
ed
with
C
h
in
a
d
ataset
a
n
d
h
ig
h
es
t
er
r
o
r
s
w
er
e
g
e
n
er
ated
w
it
h
I
S
B
SG
d
ataset.
W
h
at
is
i
n
ter
est
in
g
ab
o
u
t
t
h
e
d
ata
i
n
t
h
ese
tab
les
is
t
h
at
t
h
e
v
al
u
e
s
o
f
Md
MRE
ar
e
f
ar
lo
w
er
t
h
an
t
h
o
s
e
o
f
MM
R
E
esp
ec
iall
y
in
C
OC
OM
O,
I
SB
SG,
T
u
k
u
tu
k
u
a
n
d
Ke
m
er
er
d
atasets
.
T
h
ese
latter
f
i
n
d
in
g
s
a
g
r
ee
w
it
h
t
h
e
v
al
u
e
s
o
f
s
k
e
w
n
e
s
s
a
n
d
k
u
r
to
s
is
o
f
t
h
ese
d
atasets
t
h
at
ex
h
ib
it
h
ig
h
l
ev
el
o
f
as
y
m
m
etr
y
a
n
d
o
f
n
o
n
n
o
r
m
alit
y
.
T
ab
le
5
.
T
h
e
r
esu
lts
o
b
tain
ed
i
n
ter
m
s
o
f
p
r
ed
(
0
.
2
5
)
o
v
er
th
e
eig
h
t d
atasets
T
e
c
h
n
i
q
u
e
s
F
S
S
me
t
h
o
d
/
#
r
e
mo
v
e
d
f
e
a
t
u
r
e
s
C
O
C
O
M
O
I
S
B
S
G
T
U
K
U
T
U
K
U
A
B
R
EC
H
T
D
ESH
A
R
N
A
I
S
K
EM
ER
ER
M
I
Y
A
ZA
K
I
C
H
I
N
A
S
V
R
0
3
0
,
2
6
3
2
6
,
6
6
7
3
1
,
2
5
2
8
,
5
7
1
2
1
,
7
3
9
20
3
5
,
7
1
4
1
5
,
3
3
3
S
V
R
-
B
F
E
1
3
6
,
8
4
2
3
1
,
1
1
1
3
7
,
5
4
2
,
8
5
7
3
0
,
4
3
5
20
4
2
,
8
5
7
8
1
,
3
3
3
2
3
2
,
8
9
5
2
4
,
4
4
4
3
7
,
5
4
2
,
8
5
7
3
4
,
7
8
3
20
2
1
,
4
2
9
8
1
,
3
3
3
3
3
1
,
5
7
9
2
4
,
4
4
4
3
7
,
5
4
2
,
8
5
7
3
4
,
7
8
3
0
2
1
,
4
2
9
80
4
3
6
,
8
4
2
3
1
,
1
1
1
1
2
,
5
4
2
,
8
5
7
3
9
,
1
3
0
2
8
,
5
7
1
8
3
,
3
3
3
S
V
R
-
B
o
r
u
t
a
3
4
,
2
1
1
2
4
,
4
4
4
25
2
8
,
5
7
1
3
4
,
7
8
3
0
4
2
,
8
5
7
7
5
,
3
3
3
T
ab
le
6
.
T
h
e
r
esu
lts
o
b
tain
ed
i
n
ter
m
s
o
f
MM
R
E
o
v
er
th
e
ei
g
h
t d
ata
s
ets
T
e
c
h
n
i
q
u
e
s
F
S
S
me
t
h
o
d
/
#
r
e
mo
v
e
d
f
e
a
t
u
r
e
C
O
C
O
M
O
I
S
B
S
G
T
U
K
U
T
U
K
U
A
B
R
EC
H
T
D
ESH
A
R
N
A
I
S
K
EM
ER
ER
M
I
Y
A
ZA
K
I
C
H
I
N
A
S
V
R
0
1
,
3
6
7
1
,
7
0
3
1
,
0
6
5
0
,
5
8
3
0
,
4
6
4
1
,
3
7
0
,
5
5
6
1
,
3
3
7
S
V
R
-
B
F
E
1
1
,
3
7
5
1
,
4
7
8
0
,
8
5
6
0
,
5
8
3
0
,
4
6
7
1
,
2
3
5
0
,
5
5
3
0
,
1
8
7
2
1
,
2
6
2
1
,
1
8
7
0
,
8
0
,
5
4
8
0
,
4
5
6
1
,
3
3
4
1
,
5
0
3
0
,
1
9
1
3
1
,
3
0
4
1
,
4
0
7
0
,
8
1
4
0
,
6
6
8
0
,
5
2
1
1
,
6
2
7
1
,
3
9
4
0
,
1
9
5
4
1
,
2
4
2
1
,
0
9
2
1
,
0
2
8
0
,
6
8
5
0
,
4
6
2
1
,
5
9
1
,
3
6
7
0
,
1
7
S
V
R
-
B
o
r
u
t
a
1
,
5
2
4
1
,
5
5
9
0
,
5
0
7
0
,
5
6
6
0
,
4
5
7
1
,
6
7
1
0
,
5
2
7
0
,
2
4
2
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8938
IJ
-
A
I
Vo
l.
8
,
No
.
4
,
Dec
em
b
er
201
9
:
3
9
9
–
410
408
T
ab
le
7
.
T
h
e
r
esu
lts
o
b
tain
ed
i
n
ter
m
s
o
f
Md
MRE
o
v
er
t
h
e
e
ig
h
t d
atasets
T
e
c
h
n
i
q
u
e
s
F
S
S
me
t
h
o
d
/
#
r
e
mo
v
e
d
f
e
a
t
u
r
e
C
O
C
O
M
O
I
S
B
S
G
T
U
K
U
T
U
K
U
A
B
R
EC
H
T
D
ESH
A
R
N
A
I
S
K
EM
ER
ER
M
I
Y
A
ZA
K
I
C
H
I
N
A
S
V
R
0
0
,
4
5
1
0
,
5
4
0
,
3
1
6
0
,
3
2
3
0
,
3
7
6
0
,
5
7
9
0
,
3
7
5
0
,
7
2
6
S
V
R
-
B
F
E
1
0
,
4
5
6
0
,
5
2
6
0
,
3
3
8
0
,
4
2
2
0
,
3
9
7
0
,
8
4
4
0
,
2
9
5
0
,
0
5
4
2
0
,
4
7
1
0
,
5
6
8
0
,
3
3
9
0
,
3
0
8
0
,
3
2
4
0
,
4
6
9
0
,
5
8
8
0
,
0
5
2
3
0
,
5
2
1
0
,
4
8
6
0
,
4
0
7
0
,
7
2
5
0
,
3
9
6
0
,
6
4
6
0
,
5
2
6
0
,
0
5
5
4
0
,
5
0
8
0
,
4
2
9
0
,
4
7
0
,
7
0
9
0
,
3
6
1
0
,
6
3
6
0
,
5
7
1
0
,
0
4
7
S
V
R
-
B
o
r
u
t
a
0
,
4
1
0
,
4
3
3
0
,
5
1
6
0
,
3
4
3
0
,
3
3
2
0
,
6
6
8
0
,
3
1
1
0
,
0
8
8
T
ab
le
8
.
T
h
e
R
esu
lt
s
o
b
tain
ed
in
ter
m
s
o
f
p
r
ed
(
0
.
2
5
)
,
Md
MR
E
an
d
Md
MRE
o
v
er
th
e
eig
h
t
d
atasets
T
e
c
h
n
i
q
u
e
s
F
S
S
me
t
h
o
d
/
#
r
e
mo
v
e
d
f
e
a
t
u
r
e
P
r
e
d
(
0
.
2
5
)
M
M
R
E
M
d
M
R
E
S
V
R
0
2
6
,
1
9
2
1
,
0
5
6
0
,
4
6
1
S
V
R
-
B
F
E
1
4
0
,
3
6
7
0
,
8
4
2
0
,
4
1
7
2
3
6
,
9
0
5
0
,
9
1
0
0
,
3
9
0
3
3
4
,
0
7
4
0
,
9
9
1
0
,
4
7
0
4
3
4
,
2
9
3
0
,
9
5
5
0
,
4
6
6
S
V
R
-
B
o
r
u
t
a
3
3
,
1
5
0
0
,
8
8
2
0
,
3
8
8
T
o
s
u
m
u
p
,
th
e
f
i
n
d
i
n
g
s
o
f
th
is
s
t
u
d
y
s
u
g
g
est
t
h
at
t
h
e
u
s
e
o
f
f
ea
t
u
r
e
s
elec
tio
n
m
et
h
o
d
in
th
e
p
r
ep
r
o
ce
s
s
in
g
p
h
ase
o
f
t
h
e
SV
R
m
o
d
el
b
u
ild
in
g
ca
n
co
n
tr
ib
u
te
s
ig
n
i
f
ican
tl
y
to
i
m
p
r
o
v
e
t
h
e
ac
cu
r
ac
y
o
f
ef
f
o
r
t
esti
m
ates.
I
n
ad
d
itio
n
,
th
e
b
ac
k
w
ar
d
f
ea
t
u
r
e
s
elec
tio
n
ca
n
g
e
n
er
ate
b
etter
ef
f
o
r
t
esti
m
ates
t
h
a
n
B
o
r
u
ta
m
e
th
o
d
.
7.
CO
NCLU
SI
O
N
AND
F
U
T
U
RE
WO
RK
T
h
is
e
m
p
ir
ical
s
tu
d
y
as
s
es
s
e
d
th
e
i
m
p
ac
t
o
f
f
ea
t
u
r
e
s
elec
tio
n
m
eth
o
d
s
o
n
t
h
e
ac
cu
r
ac
y
o
f
SV
R
m
o
d
el
s
i
n
S
DE
E
.
Fo
r
th
i
s
p
u
r
p
o
s
e,
t
w
o
w
r
ap
p
er
f
ea
t
u
r
e
s
elec
tio
n
m
et
h
o
d
s
w
e
r
e
u
s
ed
t
o
p
r
e
-
p
r
o
ce
s
s
eig
h
t
w
ell
-
k
n
o
w
n
d
atase
ts
.
T
h
e
SV
R
m
o
d
els
b
ased
o
n
p
r
e
-
p
r
o
ce
s
s
ed
d
atasets
w
er
e
co
m
p
ar
ed
to
th
o
s
e
b
u
i
lt
w
it
h
o
u
t
f
ea
t
u
r
e
s
elec
tio
n
.
T
h
e
SVR
m
o
d
el
s
w
er
e
o
p
ti
m
ized
u
s
i
n
g
a
g
r
id
s
ea
r
ch
p
r
o
ce
d
u
r
e.
T
h
e
p
er
f
o
r
m
a
n
ce
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
els
w
as
a
s
s
e
s
s
e
d
u
s
i
n
g
th
r
ee
ac
c
u
r
ac
y
m
ea
s
u
r
es
th
r
o
u
g
h
3
0
%h
o
ld
o
u
t
v
alid
atio
n
m
eth
o
d
.
T
h
e
r
esu
lt
s
o
b
tain
ed
s
h
o
w
ed
t
h
at
t
h
e
SV
R
m
o
d
els
w
it
h
f
ea
t
u
r
e
s
elec
tio
n
g
e
n
er
ated
b
etter
esti
m
atio
n
th
a
n
t
h
e
SV
R
co
n
s
tr
u
cted
w
it
h
o
u
t
f
ea
tu
r
e
s
e
lectio
n
m
et
h
o
d
s
.
I
n
ad
d
itio
n
,
u
s
i
n
g
th
e
p
r
o
p
o
s
ed
b
ac
k
w
ar
d
f
ea
t
u
r
e
eli
m
i
n
atio
n
b
ased
o
n
R
F
f
ea
t
u
r
e
i
m
p
o
r
tan
ce
ca
n
lead
s
to
b
etter
ac
cu
r
ac
y
th
a
n
B
o
r
u
ta
m
et
h
o
d
.
Ho
w
e
v
er
,
th
i
s
s
t
u
d
y
h
as
o
n
l
y
ex
a
m
i
n
ed
t
h
e
S
VR
m
o
d
els
b
ased
o
n
o
n
e
t
y
p
e
o
f
f
ea
tu
r
e
s
elec
tio
n
m
et
h
o
d
.
T
h
er
ef
o
r
e,
it
w
o
u
ld
b
e
in
ter
esti
n
g
to
ass
e
s
s
t
h
e
i
m
p
ac
t o
f
o
th
er
s
f
ea
tu
r
e
s
elec
tio
n
m
eth
o
d
s
o
n
th
e
ac
c
u
r
ac
y
o
f
SV
R
m
o
d
els i
n
SDEE
.
RE
F
E
R
E
NC
E
S
[1
]
B.
W
.
Bo
e
h
m
,
S
o
f
tw
a
r
e
En
g
in
e
e
rin
g
Eco
n
o
m
ics
.
P
re
n
ti
c
e
Ha
ll
P
T
R,
1
9
8
1
,
p
.
7
6
8
.
[2
]
R.
d
.
A
.
A
ra
ú
jo
,
A
.
L
.
I.
Oliv
e
ir
a
,
a
n
d
S
.
R.
d
.
L
.
M
e
ira,
"
A
c
l
a
ss
o
f
h
y
b
rid
m
u
lt
il
a
y
e
r
p
e
rc
e
p
tro
n
s
f
o
r
so
f
t
w
a
re
d
e
v
e
lo
p
m
e
n
t
e
ff
o
rt
e
sti
m
a
ti
o
n
p
ro
b
lem
s,
"
Ex
p
e
rt
S
y
st.
A
p
p
l.
,
v
o
l.
9
0
,
p
p
.
1
-
1
2
,
/
2
0
1
7
.
[3
]
A
.
Za
k
ra
n
i
a
n
d
A
.
Id
ri,
"
A
p
p
l
y
in
g
ra
d
ial
b
a
sis
f
u
n
c
ti
o
n
n
e
u
ra
l
n
e
t
w
o
rk
s
b
a
se
d
o
n
f
u
z
z
y
c
lu
ste
rin
g
to
e
stim
a
te
we
b
a
p
p
li
c
a
ti
o
n
s e
f
f
o
rt"
,
In
tern
a
ti
o
n
a
l
Re
v
ie
w
o
n
Co
m
p
u
ters
a
n
d
S
o
f
twa
re
,
A
rti
c
le
v
o
l.
5
,
n
o
.
5
,
p
p
.
5
1
6
-
5
2
4
,
2
0
1
0
.
[4
]
A
.
Zak
r
a
n
i,
A
.
Na
m
ir,
a
n
d
M
.
Ha
in
,
"
In
v
e
stig
a
ti
n
g
th
e
u
se
o
f
r
a
n
d
o
m
f
o
re
st
in
so
f
t
w
a
re
c
o
st
e
stim
a
ti
o
n
"
,
T
h
e
S
e
c
o
n
d
I
n
tern
a
ti
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
I
n
telli
g
e
n
t
C
o
m
p
u
ti
n
g
i
n
Da
ta S
c
ien
c
e
s,
F
è
s,
3
-
5
o
c
t
o
b
e
r,
2
0
1
8
.
[5
]
S
.
M
.
S
a
tap
a
th
y
,
B.
P
.
A
c
h
a
r
y
a
,
a
n
d
S
.
K.
Ra
th
,
"
Early
sta
g
e
so
f
t
w
a
re
e
ff
o
rt
e
sti
m
a
ti
o
n
u
si
n
g
ra
n
d
o
m
f
o
re
st
tec
h
n
iq
u
e
b
a
se
d
o
n
u
se
c
a
se
p
o
i
n
t
s"
,
IET
S
o
f
t
w
a
re
,
A
rti
c
le v
o
l.
1
0
,
n
o
.
1
,
p
p
.
1
0
-
1
7
,
2
0
1
6
.
[6
]
A
.
Id
ri
a
n
d
I.
A
b
n
a
n
e
,
"
F
u
z
z
y
An
a
lo
g
y
Ba
s
e
d
Eff
o
rt
Esti
m
a
ti
o
n
:
A
n
E
m
p
iri
c
a
l
Co
m
p
a
ra
ti
v
e
S
tu
d
y
,
"
in
1
7
th
I
EE
E
In
tern
a
ti
o
n
a
l
Co
n
f
e
re
n
c
e
o
n
Co
m
p
u
ter
a
n
d
In
f
o
rm
a
ti
o
n
T
e
c
h
n
o
l
o
g
y
,
CIT
2
0
1
7
,
2
0
1
7
,
p
p
.
1
1
4
-
1
2
1
:
In
st
it
u
te
o
f
El
e
c
tri
c
a
l
a
n
d
El
e
c
tro
n
ics
En
g
i
n
e
e
rs In
c
.
[7
]
A
.
L
.
I.
Oliv
e
ira,
"
Esti
m
a
ti
o
n
o
f
so
f
t
w
a
re
p
ro
jec
t
e
ff
o
rt
w
it
h
su
p
p
o
rt
v
e
c
to
r
re
g
re
ss
io
n
"
,
Ne
u
ro
c
o
m
p
u
ti
n
g
,
A
rti
c
le
v
o
l.
6
9
,
n
o
.
1
3
-
1
5
,
p
p
.
1
7
4
9
-
1
7
5
3
,
2
0
0
6
.
[8
]
Q.
L
iu
,
J.
X
iao
,
a
n
d
H.
Z
h
u
,
"
F
e
a
tu
re
se
lec
ti
o
n
f
o
r
so
f
tw
a
re
e
ff
o
r
t
e
stim
a
ti
o
n
w
it
h
lo
c
a
li
z
e
d
n
e
ig
h
b
o
r
h
o
o
d
m
u
tu
a
l
in
f
o
rm
a
ti
o
n
"
,
Clu
ste
r
Co
m
p
u
ti
n
g
,
A
rti
c
le i
n
P
re
ss
p
p
.
1
-
9
,
2
0
1
8
.
[9
]
Z.
Ch
e
n
,
T
.
M
e
n
z
ies
,
D.
P
o
rt,
a
n
d
B.
Bo
e
h
m
,
"
F
e
a
tu
re
su
b
se
t
se
lec
ti
o
n
c
a
n
im
p
ro
v
e
so
f
tw
a
r
e
c
o
st
e
stim
a
ti
o
n
a
c
c
u
ra
c
y
,
"
in
2
0
0
5
W
o
rk
sh
o
p
o
n
P
re
d
icto
r
M
o
d
e
ls
in
S
o
f
tw
a
r
e
En
g
in
e
e
rin
g
,
P
ROMIS
E
2
0
0
5
,
2
0
0
5
:
A
ss
o
c
iatio
n
f
o
r
Co
m
p
u
ti
n
g
M
a
c
h
in
e
ry
,
In
c
.
[1
0
]
M
.
A
z
z
e
h
,
D.
Ne
a
g
u
,
a
n
d
P
.
Co
w
li
n
g
,
"
I
m
p
ro
v
in
g
a
n
a
lo
g
y
so
f
t
wa
re
e
ff
o
rt
e
sti
m
a
ti
o
n
u
sin
g
f
u
z
z
y
fe
a
tu
re
su
b
se
t
se
lec
ti
o
n
a
lg
o
rit
h
m
,
"
in
3
0
th
In
t
e
rn
a
ti
o
n
a
l
Co
n
f
e
re
n
c
e
o
n
S
o
f
t
wa
re
En
g
in
e
e
rin
g
,
ICS
E
2
0
0
8
-
4
th
In
tern
a
ti
o
n
a
l
W
o
rk
sh
o
p
o
n
P
re
d
icto
r
M
o
d
e
ls i
n
S
o
f
tw
a
r
e
En
g
in
e
e
rin
g
,
P
ROMIS
E
2
0
0
8
,
L
e
ip
z
ig
,
2
0
0
8
,
p
p
.
7
1
-
7
8
.
Evaluation Warning : The document was created with Spire.PDF for Python.