I
AE
S
I
nte
rna
t
io
na
l J
o
urna
l o
f
Art
if
icia
l In
t
ellig
ence
(
I
J
-
AI
)
Vo
l.
9
,
No
.
1
,
Ma
r
ch
2
0
2
0
,
p
p
.
65
~
72
I
SS
N:
2252
-
8938
,
DOI
: 1
0
.
1
1
5
9
1
/i
j
ai.
v
9
.i
1
.
p
p
65
-
72
65
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
Cla
ss
ificatio
n of
m
ul
ticlas
s i
m
ba
la
nced data
usi
ng
c
o
st
-
sens
itive
decisio
n t
re
e C5.0
M
.
Aldi
k
i F
ebria
nto
no
1
,
Sh
o
leh H
a
di P
ra
m
o
no
2
,
Ra
h
m
a
dw
a
t
i
3
,
G
o
ls
ha
h Na
g
hd
y
4
1
De
p
a
rtme
n
t
o
f
El
e
c
tri
c
a
l
a
n
d
Co
m
p
u
ter E
n
g
in
e
e
rin
g
,
Bra
w
ij
a
y
a
Un
iv
e
rsity
,
In
d
o
n
e
sia
2,
3
F
a
c
u
lt
y
,
De
p
a
rtme
n
t
o
f
El
e
c
tri
c
a
l
a
n
d
C
o
m
p
u
ter E
n
g
in
e
e
rin
g
,
Br
a
w
ij
a
y
a
Un
iv
e
rsit
y
,
In
d
o
n
e
sia
4
S
c
h
o
o
l
o
f
El
e
c
tri
c
a
l,
Co
m
p
u
ter,
T
e
le
c
o
m
m
u
n
ica
ti
o
n
En
g
in
e
e
rin
g
.
Un
iv
e
rsit
y
o
f
W
o
ll
o
n
g
o
n
g
,
Ne
w
S
o
u
t
h
W
a
les
A
u
stra
li
a
Art
icle
I
nfo
AB
ST
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Sep
t
5
,
2
0
1
9
R
ev
i
s
ed
No
v
2
5
,
2
0
1
9
A
cc
ep
ted
Dec
1
0
,
2
0
1
9
T
h
e
m
u
lt
icla
ss
im
b
a
lan
c
e
d
d
a
ta
p
ro
b
lem
s
in
d
a
ta
m
in
in
g
w
e
r
e
in
tere
stin
g
c
a
se
s
to
stu
d
y
c
u
rre
n
tl
y
.
T
h
e
p
ro
b
lem
s
h
a
d
a
n
in
f
lu
e
n
c
e
o
n
th
e
c
l
a
ss
if
i
c
a
ti
o
n
p
ro
c
e
ss
in
m
a
c
h
in
e
lea
rn
in
g
p
r
o
c
e
ss
e
s.
S
o
m
e
c
a
s
e
s
sh
o
w
e
d
th
a
t
m
in
o
rit
y
c
las
s
in
th
e
d
a
tas
e
t
h
a
d
a
n
im
p
o
rtan
t
i
n
f
o
rm
a
ti
o
n
v
a
lu
e
c
o
m
p
a
re
d
to
t
h
e
m
a
jo
rit
y
c
la
ss
.
W
h
e
n
m
in
o
rit
y
c
la
ss
wa
s
m
isc
l
a
ss
i
f
ica
ti
o
n
,
it
w
o
u
ld
a
ff
e
c
t
th
e
a
c
c
u
ra
c
y
v
a
lu
e
a
n
d
c
las
sif
ier
p
e
rf
o
r
m
a
n
c
e
.
In
th
is
re
se
a
rc
h
,
c
o
st
se
n
siti
v
e
d
e
c
isio
n
tree
C5
.
0
w
a
s
u
se
d
to
so
lv
e
m
u
lt
icla
ss
i
m
b
a
lan
c
e
d
d
a
ta
p
ro
b
lem
s.
T
h
e
f
irst
sta
g
e
,
m
a
k
in
g
th
e
d
e
c
isi
o
n
tree
m
o
d
e
l
u
se
s
th
e
C5
.
0
a
lg
o
rit
h
m
th
e
n
th
e
c
o
st
se
n
siti
v
e
lea
rn
in
g
u
se
s
t
h
e
m
e
tac
o
st
m
e
th
o
d
t
o
o
b
tai
n
t
h
e
m
in
im
u
m
c
o
st
m
o
d
e
l.
T
h
e
r
e
su
lt
s
o
f
te
stin
g
th
e
C5
.
0
a
lg
o
rit
h
m
h
a
d
b
e
tt
e
r
p
e
rf
o
r
m
a
n
c
e
th
a
n
C
4
.
5
a
n
d
ID3
a
lg
o
rit
h
m
s.
T
h
e
p
e
rc
e
n
tag
e
o
f
a
lg
o
rit
h
m
p
e
rf
o
r
m
a
n
c
e
f
ro
m
C5
.
0
,
C
4
.
5
a
n
d
ID3
w
e
re
4
0
.
9
1
%
,
4
0
,
2
4
%
a
n
d
1
9
.
2
3
%
K
ey
w
o
r
d
s
:
C
las
s
i
f
icatio
n
Dec
is
io
n
T
r
ee
C
5
.
0
Me
taco
s
t
Mu
lticla
s
s
P
ar
ticle
S
w
ar
m
Op
ti
m
at
io
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
BY
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
M.
A
ld
i
k
i Feb
r
ian
to
n
o
,
R
esear
ch
Sch
o
lar
,
Dep
ar
t
m
en
t o
f
E
lectr
ical
an
d
C
o
m
p
u
ter
E
n
g
in
ee
r
i
n
g
,
B
r
a
w
ij
ay
a
U
n
i
v
er
s
it
y
,
I
n
d
o
n
e
s
ia
E
m
ail: a
ld
i
k
i
f
eb
r
ian
to
n
o
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
I
n
th
e
d
i
g
ital
er
a
n
o
w
ad
a
y
s
,
t
h
e
a
m
o
u
n
t
o
f
d
ata
a
v
ailab
le
i
n
th
e
d
atab
ase
is
in
cr
ea
s
in
g
ti
m
e
b
y
ti
m
e.
Ho
w
e
v
er
,
n
o
t
all
o
f
t
h
e
d
ata
c
an
b
e
u
s
ed
a
n
d
u
tili
ze
d
e
v
en
t
h
o
u
g
h
a
lo
t
o
f
d
ata
h
as
i
m
p
o
r
tan
t
i
n
f
o
r
m
atio
n
a
s
f
u
tu
r
e
k
n
o
w
led
g
e.
B
ased
o
n
class
es
’
d
is
tr
ib
u
t
io
n
,
t
h
e
d
ata
s
et
is
d
iv
id
ed
in
to
t
w
o
,
n
a
m
e
l
y
b
in
ar
y
d
ata
an
d
m
u
lti
-
clas
s
d
ata
[
1
]
.
B
in
ar
y
d
ata
o
n
l
y
h
as
t
w
o
clas
s
es
w
h
il
e
m
u
lti
-
cla
s
s
d
ata
h
as
m
o
r
e
t
h
an
t
w
o
class
e
s
[
2
]
.
B
in
ar
y
i
m
b
ala
n
ce
d
d
ata
an
d
m
u
lti
-
clas
s
i
m
b
alan
ce
d
d
ata
ar
e
p
r
o
b
lem
s
th
at
m
u
s
t
b
e
f
a
ce
d
n
o
w
ad
a
y
s
[
3
]
.
T
h
e
m
aj
o
r
ity
cla
s
s
i
s
a
cla
s
s
th
a
t
h
a
s
a
r
atio
n
u
m
b
er
o
f
in
s
ta
n
ce
s
m
o
r
e
th
a
n
o
t
h
e
r
class
es,
w
h
er
ea
s
th
e
m
in
o
r
it
y
cla
s
s
i
s
o
f
te
n
co
n
s
id
er
ed
n
o
t
in
f
l
u
en
t
ial
in
t
h
e
m
ac
h
i
n
e
lear
n
i
n
g
b
ec
au
s
e
it
o
n
l
y
h
as
les
s
r
atio
n
u
m
b
er
o
f
in
s
ta
n
ce
s
[
4
]
.
I
n
t
h
e
r
ea
l
w
o
r
ld
,
m
in
o
r
it
y
c
lass
h
as
an
i
m
p
o
r
ta
n
t
i
m
p
ac
t
b
ec
a
u
s
e
it
co
n
tai
n
s
u
s
e
f
u
l
k
n
o
w
led
g
e
i
n
f
o
r
m
atio
n
[
5
]
,
s
u
ch
a
s
d
iag
n
o
s
i
s
t
y
p
e
s
o
f
d
is
e
ase
[
6
]
,
p
r
ed
ictin
g
cr
ed
it
w
o
r
t
h
in
e
s
s
o
f
b
an
k
[
7
]
,
an
d
telec
o
m
cu
s
to
m
er
ch
u
r
n
p
r
ed
ictio
n
[
8
]
.
I
n
th
e
class
i
f
ica
tio
n
p
r
o
ce
s
s
u
s
in
g
m
ac
h
in
e
lear
n
in
g
,
m
i
n
o
r
it
y
cla
s
s
e
s
ar
e
o
f
ten
m
i
s
clas
s
i
f
i
ed
b
ec
au
s
e
m
ac
h
in
e
lear
n
i
n
g
p
r
io
r
itizes
m
aj
o
r
ity
cla
s
s
e
s
an
d
i
g
n
o
r
es
m
i
n
o
r
it
y
cla
s
s
es
[
9
]
.
T
h
er
e
h
av
e
b
ee
n
m
a
n
y
tech
n
iq
u
es
ca
r
r
ied
o
u
t
b
y
r
e
s
e
ar
ch
er
s
to
s
o
lv
e
p
r
o
b
le
m
s
in
i
m
b
a
lan
ce
d
d
ata
[
1
0
]
.
I
n
class
i
f
icatio
n
p
r
o
ce
s
s
,
th
e
m
u
lti
-
clas
s
i
m
b
ala
n
ce
d
d
ata
h
as
m
o
r
e
d
i
f
f
icu
lt
le
v
el
to
s
o
lv
e
r
at
h
er
t
h
an
b
in
ar
y
i
m
b
al
an
ce
d
d
ata
b
ec
au
s
e
m
u
lti
-
clas
s
i
m
b
ala
n
ce
d
d
ata
ca
n
h
a
v
e
m
o
r
e
t
h
a
n
o
n
e
m
i
n
o
r
it
y
cla
s
s
es
[
1
1
]
.
T
h
er
e
ar
e
m
an
y
w
a
y
s
a
n
d
tech
n
iq
u
es
t
h
at
ca
n
b
e
u
s
ed
f
o
r
m
u
lt
i
-
cla
s
s
i
m
b
a
lan
ce
d
d
ata
p
r
o
b
lem
s
[
1
2
]
.
Ho
w
e
v
er
,
to
o
b
tain
o
p
ti
m
al
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8938
I
n
t J
A
r
ti
f
I
n
tell
,
Vo
l.
9
,
No
.
1
,
Ma
r
ch
20
20
:
65
–
72
66
r
esu
lt
s
,
it
n
ee
d
s
n
e
w
d
ev
elo
p
m
en
t
tech
n
iq
u
e
s
.
E
s
p
ec
iall
y
,
tech
n
iq
u
es
t
h
at
ar
e
f
o
cu
s
o
n
th
e
class
i
f
ica
tio
n
o
f
m
i
n
o
r
it
y
clas
s
es
w
i
th
o
u
t ig
n
o
r
in
g
t
h
e
m
aj
o
r
ity
cla
s
s
e
s
.
On
e
t
ec
h
n
iq
u
e
t
h
at
is
e
f
f
ec
ti
v
e
an
d
ef
f
icie
n
t
to
h
a
n
d
le
t
h
e
i
m
b
a
lan
ce
d
d
ata
p
r
o
b
le
m
i
s
co
s
t
s
e
n
s
i
tiv
e
d
ec
is
io
n
tr
ee
[
1
3
-
1
4
]
.
T
h
e
tech
n
iq
u
e
i
s
t
w
o
m
et
h
o
d
s
co
m
b
in
at
io
n
,
n
a
m
e
l
y
co
s
t
-
s
e
n
s
i
tiv
e
lear
n
i
n
g
a
n
d
d
ec
is
io
n
tr
ee
.
T
h
e
tech
n
iq
u
e
w
o
r
k
s
b
y
m
a
k
i
n
g
a
d
ec
is
io
n
t
r
e
e
m
o
d
el,
in
w
h
ich
ca
lc
u
late
s
th
e
m
o
s
t
m
i
n
i
m
u
m
co
s
t
o
f
m
i
s
clas
s
i
f
icatio
n
[
1
5
-
1
6
]
.
T
h
u
s
,
i
m
b
ala
n
ce
d
d
ata
m
is
c
lass
if
ica
tio
n
ca
n
b
e
m
in
i
m
ized
b
y
u
s
i
n
g
co
s
t
s
en
s
iti
v
e
lear
n
i
n
g
tec
h
n
iq
u
es
in
t
h
e
d
ec
is
io
n
tr
ee
m
o
d
el.
T
h
er
ef
o
r
e,
it
n
ee
d
s
n
e
w
d
ev
elo
p
m
e
n
t
in
m
a
k
i
n
g
a
d
ec
is
io
n
tr
ee
s
m
o
d
el.
T
h
e
C
5
.
0
alg
o
r
ith
m
is
a
d
ev
elo
p
m
en
t
o
f
th
e
C
4
.
5
an
d
I
D3
alg
o
r
ith
m
s
[
1
7
]
.
T
h
e
C
5
.
0
alg
o
r
ith
m
w
o
r
k
s
b
y
u
s
i
n
g
in
f
o
r
m
at
io
n
g
ai
n
.
I
t
ca
n
m
a
k
e
t
h
e
d
ec
is
io
n
tr
ee
m
o
d
el
m
o
r
e
ef
f
ec
ti
v
e
an
d
f
a
s
ter
in
m
ak
in
g
d
ec
is
io
n
s
[
1
8
-
2
0
]
.
I
n
th
is
r
esear
c
h
,
w
e
w
ill
f
o
cu
s
o
n
h
a
n
d
lin
g
m
u
lticlas
s
i
m
b
a
lan
ce
d
d
ata
p
r
o
b
lem
s
b
y
u
s
in
g
co
s
t
s
e
n
s
i
tiv
e
d
ec
is
i
o
n
tr
ee
C
5
.
0
.
T
h
en
,
co
s
t
s
en
s
i
tiv
e
lear
n
i
n
g
u
s
e
s
m
etac
o
s
t
m
eth
o
d
.
T
h
e
m
et
h
o
d
w
o
r
k
s
b
y
r
e
-
lab
elin
g
an
d
p
r
u
n
in
g
o
f
a
d
ec
is
io
n
tr
ee
m
o
d
el
to
b
u
ild
a
m
in
i
m
u
m
co
s
t
m
o
d
el.
T
h
e
d
ec
is
io
n
tr
ee
m
o
d
el
h
as
an
i
m
p
ac
t
to
th
e
p
er
f
o
r
m
an
ce
a
n
d
r
ate
o
f
th
e
f
o
r
w
ar
d
class
i
f
icatio
n
p
r
o
ce
s
s
.
T
h
is
p
ap
er
aim
s
to
co
m
p
ar
e
t
h
e
p
er
f
o
r
m
an
ce
o
f
C
5
.
0
,
C
4
.
5
an
d
I
D3
alg
o
r
ith
m
s
i
n
d
esig
n
i
n
g
th
e
d
ec
is
io
n
tr
ee
m
o
d
el.
2.
RE
S
E
ARCH
M
E
T
H
O
D
2
.
1
.
Da
t
a
s
et
T
h
is
r
esear
ch
is
u
s
in
g
d
at
asets
f
r
o
m
U
C
I
Ma
ch
i
n
e
L
ea
r
n
i
n
g
R
ep
o
s
ito
r
y
,
n
a
m
el
y
:
Gla
s
s
,
L
y
m
p
o
g
r
ap
h
y
,
Vec
h
ile,
T
h
y
r
o
id
,
an
d
W
in
e.
T
h
e
r
ea
s
o
n
o
f
ch
o
o
s
in
g
th
e
s
e
f
i
v
e
d
at
asets
b
ec
au
s
e
t
h
at
th
e
m
u
lti
-
clas
s
i
m
b
ala
n
ce
d
d
ata
is
o
f
ten
u
s
ed
i
n
t
h
e
p
r
o
b
lem
s
o
f
c
lass
if
ica
tio
n
r
esear
c
h
[
2
1
-
2
3
]
.
T
h
e
d
escr
ip
tio
n
o
f
th
ese
d
ata
s
et
is
li
s
ted
in
T
ab
le
1
.
10
-
f
o
ld
cr
o
s
s
v
alid
atio
n
i
s
u
s
ed
i
n
m
ac
h
i
n
e
lear
n
in
g
p
r
o
ce
s
s
b
ec
au
s
e
it
h
a
s
b
ec
o
m
e
a
s
tan
d
ar
d
f
o
r
in
cr
ea
s
i
n
g
cl
ass
i
f
ier
r
esu
lts
,
w
h
er
e
th
e
d
at
aset
is
d
iv
id
ed
in
to
t
w
o
,
n
a
m
el
y
9
0
% f
o
r
tr
ain
i
n
g
an
d
1
0
% f
o
r
test
in
g
[
9
].
T
ab
le
1
.
Descr
ip
tio
n
o
f
d
atasets
No
D
a
t
a
se
t
N
u
mb
e
r
o
f
I
n
st
a
n
c
e
s
N
u
mb
e
r
o
f
A
t
r
i
b
u
t
N
u
mb
e
r
o
f
C
l
a
sse
s
D
e
scri
p
t
i
o
n
1
G
l
a
ss
2
1
4
10
7
G
l
a
ss i
d
e
n
t
i
f
i
c
a
t
i
o
n
d
a
t
a
2
L
y
mp
o
g
r
a
f
i
1
4
8
18
4
H
e
a
l
t
h
R
e
so
u
r
c
e
D
a
t
a
se
t
3
V
e
c
h
i
l
e
9
4
6
18
4
V
e
c
h
i
l
e
i
d
e
n
t
i
f
i
c
a
t
i
o
n
d
a
t
a
4
T
h
y
r
o
i
d
2
1
5
5
4
H
e
a
l
t
h
R
e
so
u
r
c
e
D
a
t
a
se
t
5
W
i
n
e
1
7
8
13
3
W
i
n
e
i
d
e
n
t
i
f
i
c
a
t
i
o
n
d
a
t
a
2
.
2
.
P
re
pro
ce
s
s
ing
P
r
ep
r
o
ce
s
s
in
g
is
th
e
p
r
o
ce
s
s
in
g
o
f
d
atasets
b
y
a
n
al
y
zi
n
g
a
n
d
i
m
p
r
o
v
in
g
d
ata
s
o
t
h
at
it
ca
n
p
r
o
d
u
ce
n
e
w
d
atase
ts
t
h
at
ar
e
g
o
o
d
to
u
s
e
i
n
t
h
e
n
ex
t
p
r
o
ce
s
s
.
T
h
e
w
a
y
to
h
a
n
d
le
t
h
e
p
r
o
b
le
m
i
s
b
y
g
o
i
n
g
th
r
o
u
g
h
s
ev
er
al
s
ta
g
es,
i
n
clu
d
i
n
g
ch
a
n
g
in
g
/cle
a
n
i
n
g
d
ata,
r
ed
u
cin
g
d
ata
an
d
tr
an
s
f
o
r
m
in
g
d
ata
[
2
4
]
.
T
h
e
n
ex
t
s
tep
is
to
n
o
r
m
alize
m
i
n
-
m
ax
b
y
c
h
an
g
i
n
g
t
h
e
v
alu
e
o
f
t
h
e
d
ata
s
ets
b
y
s
ca
li
n
g
0
-
1
w
i
th
th
e
ai
m
o
f
w
ei
g
h
ti
n
g
th
e
i
n
f
o
r
m
atio
n
co
n
tai
n
ed
in
b
alan
ce
d
d
atasets
to
d
o
th
e
attr
i
b
u
te
s
elec
tio
n
p
r
o
ce
s
s
.
Stated
b
elo
w
is
M
in
-
m
ax
n
o
r
m
aliza
t
io
n
(
1
)
.
m
i
n
(
)
m
a
x
(
)
m
i
n
(
)
i
xx
y
xx
(
1
)
y
is
th
e
r
e
s
u
lt
o
f
s
ca
l
in
g
n
e
w
v
al
u
es
(
0
-
1
)
,
x
i
is
t
h
e
d
at
a
v
alu
e
th
at
w
an
t
s
to
b
e
ch
an
g
ed
.
Mi
n
(
x
)
is
th
e
m
in
i
m
u
m
v
al
u
e
o
f
t
h
e
attr
i
b
u
te
(
x
)
an
d
m
ax
(
x
)
m
a
x
i
m
u
m
v
al
u
e
o
f
t
h
e
attr
ib
u
te
(
x
).
2
.
3
.
P
a
rt
icle
s
wa
r
m
o
pti
m
iz
a
t
io
n
P
ar
ticle
s
w
ar
m
o
p
ti
m
izatio
n
(
P
SO)
is
a
m
eth
o
d
o
f
o
p
ti
m
izat
io
n
th
a
t
is
g
o
o
d
at
i
m
p
r
o
v
i
n
g
p
er
f
o
r
m
a
n
ce
b
y
s
elec
t
in
g
attr
ib
u
tes
b
ased
o
n
th
e
v
al
u
e
o
f
in
f
o
r
m
atio
n
w
ei
g
h
ts
i
n
th
e
d
a
tasets
.
I
n
f
o
r
m
ati
v
e
an
d
r
elev
an
t
attr
ib
u
te
s
w
ill
b
e
clo
s
er
t
o
1
w
h
ile
t
h
e
attr
ib
u
tes
th
at
ar
e
n
o
t
n
ee
d
ed
ar
e
clo
s
er
to
0
.
T
h
er
e
ar
e
t
w
o
p
ar
a
m
eter
s
in
d
eter
m
in
i
n
g
t
h
e
v
alu
e
o
f
in
f
o
r
m
atio
n
w
eig
h
t,
n
a
m
el
y
t
h
e
v
elo
cit
y
v
a
lu
e
a
n
d
p
o
s
itio
n
v
al
u
e
[
2
5
]
.
B
elo
w
ar
e
th
e
s
tep
s
o
f
s
elec
t
i
n
g
at
tr
ib
u
te
s
u
s
in
g
p
ar
ticle
s
w
ar
m
o
p
ti
m
iz
atio
n
:
Step
1
.
I
n
p
u
t th
e
d
ataset.
Step
2
.
T
h
e
v
elo
cit
y
a
n
d
in
itia
l p
o
s
itio
n
o
f
th
e
at
tr
ib
u
te
ar
e
d
eter
m
i
n
ed
r
an
d
o
m
l
y
.
Step
3
.
I
n
itial v
elo
cit
y
a
t p
o
s
it
io
n
0
to
w
ar
d
s
o
p
ti
m
al
p
o
in
t 1
.
Step
4
.
C
alcu
lati
n
g
th
e
v
elo
cit
y
o
f
attr
ib
u
te
i
i
n
d
d
i
m
e
n
s
io
n
u
s
i
n
g
th
e
f
o
llo
w
in
g
(
2
)
:
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
A
r
ti
f
I
n
tell
I
SS
N:
2252
-
8938
C
la
s
s
i
fica
tio
n
o
f m
u
lticla
s
s
im
b
a
la
n
ce
d
d
a
t
a
u
s
in
g
c
o
s
t
-
s
en
s
itive
d
ec
is
io
n
tr
ee
C
5
.
0
(
M.
A
l
d
iki
F
eb
r
ia
n
to
n
o
)
67
1
1
1
2
2
kk
i
d
i
d
i
d
i
d
i
d
i
d
V
w
V
c
r
a
n
d
P
X
c
r
a
n
d
G
X
(
2
)
Step
5
.
C
alcu
lati
n
g
th
e
p
o
s
itio
n
o
f
i
n
d
iv
id
u
al
attr
ib
u
te
s
i
in
d
d
im
e
n
s
io
n
s
u
s
in
g
t
h
e
f
o
llo
w
i
n
g
(
3
)
:
11
k
k
k
i
d
i
d
i
d
X
X
V
(
3
)
Step
6
.
Up
d
atin
g
P
b
est an
d
Gb
est
to
g
et
th
e
m
o
s
t o
p
ti
m
a
l v
e
lo
cit
y
an
d
p
o
s
itio
n
o
n
ea
ch
att
r
ib
u
te.
Step
7
.
A
ttrib
u
tes
w
h
o
s
e
p
o
s
it
io
n
an
d
i
n
itial
v
elo
cit
y
d
o
n
o
t
ch
an
g
e/ar
e
w
o
r
th
0
(
ze
r
o
)
o
r
h
av
e
s
m
all
v
al
u
e
ar
e
r
e
m
o
v
ed
af
ter
t
h
e
u
p
d
ate
p
r
o
ce
s
s
is
co
m
p
le
te.
w
h
er
e,
V
id
=
I
n
d
iv
id
u
al
v
elo
cit
y
co
m
p
o
n
en
t
i
in
d
d
i
m
e
n
s
io
n
X
id
=
P
o
s
itio
n
o
f
in
d
i
v
id
u
al
i
i
n
d
d
i
m
en
s
io
n
=
I
n
er
tia
w
ei
g
h
t p
ar
a
m
eter
12
cc
=
C
o
n
s
ta
n
t a
cc
eler
atio
n
(
lear
n
in
g
r
ate)
,
th
e
v
al
u
e
is
b
et
w
ee
n
0
an
d
1
1
,
2
r
a
n
d
=
R
an
d
o
m
p
ar
a
m
eter
b
et
w
ee
n
0
an
d
1
id
P
=
P
b
est (
lo
ca
l
b
est)
in
d
iv
id
u
a
l
i
in
d
d
i
m
en
s
io
n
id
G
=
Gb
est (
g
lo
b
al
b
est)
in
d
iv
id
u
al
i
in
d
d
i
m
en
s
io
n
2
.
4
.
Dec
is
io
n
t
ree
T
h
e
Dec
is
io
n
T
r
ee
is
a
class
if
icatio
n
m
e
th
o
d
in
d
ata
m
in
in
g
class
i
f
icatio
n
th
at
ca
n
b
e
u
s
e
d
to
s
o
lv
e
class
i
f
icatio
n
p
r
o
b
lem
s
b
o
th
b
in
ar
y
clas
s
an
d
m
u
lt
i
-
cla
s
s
[
2
6
]
.
T
h
e
co
n
ce
p
t
o
f
d
ec
is
io
n
tr
ee
m
o
d
el
is
a
to
p
d
o
w
n
s
tr
u
ct
u
r
e
th
at
tu
r
n
d
ata
in
to
a
d
ec
is
io
n
tr
ee
t
h
en
p
r
o
d
u
ce
r
u
les.
T
h
e
d
ec
is
io
n
tr
ee
r
eq
u
ir
es
alg
o
r
it
h
m
f
o
r
m
u
la
f
o
r
d
eter
m
i
n
i
n
g
t
h
e
r
o
o
t
n
o
d
e,
in
ter
n
a
l
n
o
d
e
an
d
leaf
n
o
d
e
to
m
a
k
e
a
d
ec
is
io
n
tr
ee
m
o
d
el
[
2
7
]
.
A
l
g
o
r
ith
m
s
th
at
ar
e
o
f
te
n
u
s
e
d
to
s
o
lv
e
class
i
f
icatio
n
p
r
o
b
le
m
s
ar
e
I
D3
,
C
4
.
5
an
d
C
5
.
0
[
2
8
]
.
C
o
m
p
ar
is
o
n
s
o
f
ea
ch
alg
o
r
it
h
m
ar
e
s
h
o
w
n
in
T
ab
le
2
.
T
ab
le
2
.
C
o
m
p
ar
is
o
n
s
b
et
w
ee
n
d
if
f
er
en
t d
ec
is
io
n
tr
ee
alg
o
r
i
th
m
I
D
3
C
4
.
5
C
5
.
0
Ty
p
e
o
f
d
a
t
a
C
a
t
e
g
o
r
i
c
a
l
C
o
n
t
i
n
u
o
u
s a
n
d
C
a
t
e
g
o
r
i
c
a
l
C
o
n
t
i
n
u
o
u
s a
n
d
C
a
t
e
g
o
r
i
c
a
l
,
D
a
t
e
s,
T
i
me
s,
T
i
me
st
a
m
p
s
S
p
e
e
d
L
o
w
F
a
st
e
r
t
h
a
n
I
D
3
H
i
g
h
e
st
P
r
u
n
i
n
g
No
Pre
-
p
r
u
n
i
n
g
Pre
-
p
r
u
n
i
n
g
B
o
o
st
i
n
g
N
o
t
su
p
p
o
r
t
e
d
N
o
t
su
p
p
o
r
t
e
d
S
u
p
p
o
r
t
e
d
M
i
ss
i
n
g
V
a
l
u
e
C
a
n
t
’
t
d
e
a
l
w
i
t
h
C
a
n
t
’
t
d
e
a
l
w
i
t
h
C
a
n
d
e
a
l
w
i
t
h
F
o
r
mu
l
a
U
se
i
n
f
o
r
mat
i
o
n
e
n
t
r
o
p
y
U
se
sp
l
i
t
i
n
f
o
a
n
d
g
a
i
n
r
a
t
i
o
U
se
i
n
f
o
r
mat
i
o
n
G
a
i
n
2
.
5
.
C5
.
0
Alg
o
ri
t
h
m
C
5
.
0
alg
o
r
ith
m
i
s
a
d
ev
elo
p
m
en
t
o
f
th
e
p
r
ev
io
u
s
al
g
o
r
ith
m
s
,
I
D3
an
d
C
4
.
5
[
1
7
]
.
T
h
e
ad
v
an
tag
e
s
o
f
C
5
.
0
ar
e
th
at
it
is
lo
w
m
e
m
o
r
y
u
s
a
g
e,
f
aster
in
m
ak
in
g
d
ec
i
s
io
n
s
an
d
m
o
r
e
e
f
f
ec
ti
v
e
i
n
b
u
ild
in
g
d
ec
is
io
n
tr
ee
m
o
d
el
s
[
1
8
,
2
0
]
.
C
5
.
0
m
o
d
el
w
o
r
k
s
b
y
ca
lc
u
lati
n
g
t
h
e
e
n
tr
o
p
y
v
alu
e
an
d
in
f
o
r
m
atio
n
g
ai
n
o
n
ea
ch
attr
ib
u
te
[
1
9
]
.
T
h
e
attr
ib
u
te
th
at
h
a
s
t
h
e
lar
g
est
in
f
o
r
m
a
t
io
n
g
a
in
v
al
u
e
t
h
e
n
b
ec
o
m
e
s
t
h
e
r
o
o
t n
o
d
e.
T
h
en
th
e
r
o
o
t
n
o
d
e
f
o
r
m
s
a
n
e
w
cr
i
t
er
ia
b
r
an
ch
ca
lled
th
e
in
ter
n
al
n
o
d
e.
T
h
e
p
r
o
ce
s
s
w
ill
s
to
p
if
n
o
n
e
w
cr
iter
io
n
is
g
en
er
ated
.
T
h
e
d
o
w
n
n
o
d
e
t
h
en
f
u
n
ctio
n
s
as
an
o
u
tp
u
t
ca
l
l
ed
leaf
n
o
d
e.
T
h
e
L
ea
f
n
o
d
e
w
o
r
k
to
p
r
ed
ict
th
e
class
o
f
th
e
clas
s
i
f
ier
[
2
6
]
.
B
elo
w
is
t
h
e
p
r
o
ce
d
u
r
e
o
f
h
o
w
to
m
a
k
e
th
e
d
ec
is
io
n
tr
ee
m
o
d
el
u
s
i
n
g
th
e
C
5
.
0
alg
o
r
i
th
m
:
Step
1
.
I
n
p
u
t d
ataset
tr
ain
i
n
g
Step
2
.
C
alcu
lati
n
g
th
e
to
tal
e
n
tr
o
p
y
v
al
u
e
o
f
t
h
e
d
ataset
u
s
i
n
g
t
h
e
f
o
llo
w
in
g
(
4
)
:
(
4
)
Step
3
.
C
alcu
lati
n
g
th
e
e
n
tr
o
p
y
v
al
u
e
o
f
ea
c
h
cr
iter
io
n
.
Step
4
.
C
alcu
lati
n
g
th
e
i
n
f
o
r
m
atio
n
g
ai
n
v
a
lu
e
f
o
r
ea
ch
attr
ib
u
te
u
s
i
n
g
t
h
e
f
o
llo
w
i
n
g
(
5
)
:
2
1
(
)
(
)
l
o
g
(
(
)
)
n
ii
i
H
S
p
s
p
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8938
I
n
t J
A
r
ti
f
I
n
tell
,
Vo
l.
9
,
No
.
1
,
Ma
r
ch
20
20
:
65
–
72
68
(
,
)
(
)
(
)
i
n
i
A
S
I
G
S
A
H
S
H
S
S
(
5
)
Step
5
.
T
h
e
attr
ib
u
te
th
at
h
as t
h
e
lar
g
es
t in
f
o
r
m
at
io
n
g
a
in
v
al
u
e
as t
h
e
r
o
o
t n
o
d
e.
Step
6
.
C
r
ea
tin
g
a
b
r
an
c
h
f
r
o
m
th
e
n
o
d
e’
s
cr
iter
ia.
Step
7
.
R
ep
ea
tin
g
th
e
p
r
o
ce
s
s
o
f
ca
lcu
lati
n
g
th
e
en
tr
o
p
y
v
al
u
e
a
n
d
in
f
o
r
m
a
tio
n
g
ai
n
.
T
h
e
p
r
o
ce
s
s
w
ill
s
to
p
if
all
th
e
attr
ib
u
te
cr
iter
ia
h
av
e
p
r
o
d
u
ce
d
th
e
p
r
ed
icted
class
.
w
h
er
e,
S
=
Set o
f
ca
s
es
N
=
Nu
m
b
er
o
f
p
ar
titi
o
n
at
tr
ib
u
t
es A
Pi
=
P
r
o
p
o
r
tio
n
o
f
Si to
S
A
=
A
ttrib
u
te
|
S
|
=
Nu
m
b
er
o
f
ca
s
e
s
o
n
t
h
e
i p
ar
titi
o
n
|
S
|
=
Nu
m
b
er
o
f
ca
s
e
s
i
n
S
2
.
6
.
Co
s
t
-
s
ens
it
iv
e
lea
rning
C
o
s
t
-
s
e
n
s
it
iv
e
lear
n
in
g
is
a
lear
n
in
g
m
et
h
o
d
th
at
ca
n
s
o
lv
e
i
m
b
ala
n
ce
d
d
ata
p
r
o
b
le
m
s
[
1
4
]
.
T
h
e
co
n
ce
p
t
o
f
t
h
e
m
e
th
o
d
is
to
r
ed
u
ce
m
is
cla
s
s
i
f
icatio
n
co
s
t
s
o
th
a
t
ca
n
i
m
p
r
o
v
e
t
h
e
p
er
f
o
r
m
a
n
ce
o
f
th
e
cla
s
s
i
f
ier
.
T
h
is
m
et
h
o
d
ass
u
m
e
s
t
h
at
t
h
e
m
i
n
o
r
it
y
class
h
as
a
h
i
g
h
er
m
is
cla
s
s
if
icatio
n
co
s
t
t
h
a
n
th
e
m
aj
o
r
ity
clas
s
.
T
h
er
ef
o
r
e,
m
ac
h
in
e
lear
n
i
n
g
w
ill
f
o
c
u
s
o
n
ac
c
u
r
atel
y
clas
s
i
f
y
i
n
g
m
i
n
o
r
it
y
cla
s
s
.
C
o
s
t
s
en
s
iti
v
e
ca
n
b
e
ca
te
g
o
r
ized
in
to
t
w
o
cr
iter
ia.
T
h
e
f
ir
s
t
cr
it
er
io
n
is
th
e
d
ir
ec
t
m
e
th
o
d
w
h
ich
it
s
cla
s
s
i
f
ier
i
s
co
s
t
s
en
s
iti
v
e
to
t
h
e
m
s
u
ch
a
s
I
C
E
T
an
d
co
s
t
-
s
e
n
s
i
tiv
e
d
ec
i
s
io
n
tr
ee
.
T
h
e
o
th
er
ca
te
g
o
r
y
i
s
th
e
co
s
t
-
s
en
s
iti
v
e
m
eta
-
lear
n
i
n
g
m
et
h
o
d
.
I
t
i
s
ca
te
g
o
r
ized
in
to
t
w
o
tec
h
n
iq
u
es,
n
a
m
el
y
t
h
r
es
h
o
ld
in
g
an
d
s
a
m
p
lin
g
.
T
h
e
th
r
esh
o
ld
in
g
tech
n
iq
u
e
h
as
m
et
h
o
d
s
,
n
a
m
el
y
Me
ta
C
o
s
t,
C
o
s
tSen
s
iti
v
e
C
la
s
s
i
f
ier
,
C
o
s
t
-
s
e
n
s
iti
v
e
n
aï
v
e
B
ay
e
s
an
d
E
m
p
ir
ical
T
h
r
es
h
o
ld
in
g
[
1
8
]
.
Fig
u
r
e
1
s
h
o
w
s
tr
u
c
tu
r
e
o
f
co
s
t se
n
s
i
tiv
e
lear
n
i
n
g
Fig
u
r
e
1
.
Stru
ct
u
r
e
o
f
co
s
t se
n
s
iti
v
e
lear
n
i
n
g
2
.
7
.
M
et
a
Co
s
t
Me
taco
s
t
is
a
co
s
t
-
s
e
n
s
iti
v
e
m
e
th
o
d
th
at
u
s
es
t
h
e
th
r
esh
o
ld
in
g
m
eta
-
lear
n
i
n
g
ap
p
r
o
ac
h
to
m
i
n
i
m
ize
co
s
ts
[
1
5
]
.
T
h
e
w
o
r
k
i
n
g
p
r
in
cip
le
o
f
m
eta
co
s
t
is
ca
lcu
lati
n
g
th
e
p
r
o
b
ab
ilit
y
v
al
u
e
o
f
ea
ch
lea
f
n
o
d
e
[
1
6
]
.
W
h
en
th
e
p
r
o
b
ab
ilit
y
v
a
lu
e
o
f
lea
f
n
o
d
e
is
>
1
,
r
e
-
lear
n
i
n
g
o
f
ea
ch
n
o
d
e
is
d
o
n
e
b
y
p
r
u
n
i
n
g
o
r
r
elab
elin
g
u
n
ti
l
t
h
e
m
i
n
i
m
u
m
co
s
t
i
s
o
b
tain
ed
.
T
h
e
co
s
t
is
d
e
n
o
ted
as
C
(
i,
j
)
,
w
h
er
e
i
i
s
t
h
e
ac
t
u
al
class
b
u
t
i
s
p
r
ed
icted
to
b
e
cla
s
s
j
w
h
ich
ca
u
s
e
s
m
i
s
clas
s
i
f
ic
atio
n
[
2
2
]
.
T
h
e
co
s
t
w
ith
t
h
e
m
i
n
i
m
u
m
v
alu
e
w
i
ll
b
e
u
s
ed
as
t
h
e
p
r
ed
icted
clas
s
o
r
n
e
w
lea
f
n
o
d
e
to
g
et
a
n
o
p
tim
a
l
p
er
f
o
r
m
a
n
ce
.
B
elo
w
is
t
h
e
p
r
o
ce
d
u
r
e
o
f
ap
p
l
y
i
n
g
th
e
m
etac
o
s
t a
l
g
o
r
ith
m
.
M
e
t
h
o
d
:
M
e
t
a
c
o
st
Ou
t
p
u
t
:
D
e
c
i
si
o
n
T
r
e
e
S
t
e
p
:
Fo
r
i
=
1
t
o
m
L
e
t
S
i
b
e
a
l
e
a
f
n
o
d
e
o
f
S
w
i
t
h
n
n
o
d
e
.
L
e
t
M
i
=
M
o
d
e
l
p
r
o
d
u
c
e
d
b
y
a
p
p
l
y
i
n
g
L
t
o
S
i
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
A
r
ti
f
I
n
tell
I
SS
N:
2252
-
8938
C
la
s
s
i
fica
tio
n
o
f m
u
lticla
s
s
im
b
a
la
n
ce
d
d
a
t
a
u
s
in
g
c
o
s
t
-
s
en
s
itive
d
ec
is
io
n
tr
ee
C
5
.
0
(
M.
A
l
d
iki
F
eb
r
ia
n
to
n
o
)
69
Fo
r
e
a
c
h
n
o
d
e
x
i
n
S
Fo
r
e
a
c
h
c
l
a
ss
j
L
e
t
()
P
j
x
1
(
,
)
1
i
i
i
P
j
x
M
i
f
(
,
)
0
i
P
j
x
M
i
s p
r
o
d
u
c
e
d
b
y
M
i
E
l
se
(
,
)
1
i
P
j
x
M
L
e
t
x
’
s c
l
a
ss =
a
r
g
mi
n
i
(
)
(
,
)
j
P
j
x
C
i
j
L
e
t
M
=
M
o
d
e
l
p
r
o
d
u
c
e
d
b
y
a
p
p
l
y
i
n
g
L
t
o
S.
R
e
t
u
r
n
M
w
h
er
e,
S
=
T
r
ain
in
g
s
et
L
=
C
lass
if
ica
tio
n
lear
n
in
g
al
g
o
r
ith
m
C
=
C
o
s
t
m
atr
i
x
M
=
Nu
m
b
er
o
f
lea
f
n
o
d
e
to
g
e
n
er
ate
N
=
Nu
m
b
er
o
f
n
o
d
e
in
ea
ch
lea
f
n
o
d
e
2
.
8
.
Va
lid
a
t
i
o
n
m
u
lt
icla
s
s
C
o
s
t
m
atr
i
x
i
s
a
r
ep
r
esen
t
atio
n
o
f
t
h
e
r
es
u
lt
s
o
f
p
r
ed
ictio
n
er
r
o
r
s
(
m
i
s
clas
s
i
f
ic
atio
n
)
i
n
m
ac
h
in
e
lear
n
i
n
g
[
8
]
.
C
o
s
t
m
atr
i
x
is
d
e
f
i
n
ed
as
C
(
i:
j
)
w
h
er
e
i
is
th
e
ac
t
u
al
class
an
d
j
is
th
e
p
r
ed
icted
class
[
2
2
]
.
Misclass
if
icatio
n
p
r
ed
ictio
n
s
co
s
t
o
f
m
u
lti
-
clas
s
ar
e
p
r
esen
ted
s
u
c
h
as
C
(
2
,
1
)
,
C
(
3
,
1
)
,
C
(
1
,
2
)
,
C
(
3
,
2
)
,
C
(
1
,
3
)
,
C
(
2
,
3
)
,
w
h
ile
co
r
r
ec
t
class
if
icatio
n
co
s
t
o
f
m
u
lt
i
-
c
lass
ar
e
p
r
esen
t
ed
s
u
ch
as
C
(
1
,
1
)
,
C
(
2
,
2
)
,
C
(
3
,
3
)
th
at
co
s
t
v
alu
e
ar
e
0
.
T
ab
le
o
f
co
s
t
m
atr
i
x
is
s
h
o
w
n
in
t
h
e
T
ab
le
3
.
Pr
ed
icted
d
ata
o
f
th
e
class
i
f
icatio
n
m
o
d
el
ar
e
p
r
esen
ted
u
s
in
g
t
h
e
C
o
n
f
u
s
io
n
m
a
t
r
ix
tab
le,
w
h
ic
h
co
n
tai
n
s
i
n
f
o
r
m
atio
n
ab
o
u
t
t
h
e
ac
tu
al
cl
a
s
s
p
r
ese
n
ted
i
n
t
h
e
r
o
w
o
f
th
e
m
a
tr
ix
a
n
d
p
r
ed
icted
class
o
n
th
e
co
l
u
m
n
[
1
5
]
.
C
o
n
f
u
s
io
n
m
atr
ix
is
u
s
ed
to
m
ea
s
u
r
e
th
e
p
er
f
o
r
m
a
n
ce
o
f
a
class
if
ier
m
o
d
el
b
ase
d
o
n
T
P
,
T
N,
FP
an
d
FN.
T
r
u
e
p
o
s
itiv
e
(
T
P
)
an
d
tr
u
e
n
eg
a
tiv
e
(
T
N)
w
h
e
n
th
e
p
r
ed
ictio
n
r
esu
lts
ar
e
t
r
u
e
[
8
]
.
Fals
e
p
o
s
itiv
e
(
FP
)
w
h
e
n
th
e
ac
tu
al
p
r
ed
ictio
n
r
esu
lt
s
ar
e
(
-
)
b
u
t
ac
t
u
all
y
(
+)
.
Fals
e
n
e
g
ati
v
e
(
FN)
w
h
e
n
t
h
e
p
r
ed
ictio
n
r
esu
l
ts
ar
e
(
+)
b
u
t
ac
tu
all
y
(
-
).
T
ab
le
o
f
co
n
f
u
s
io
n
m
atr
i
x
is
s
h
o
w
n
i
n
t
h
e
T
ab
le
4
.
T
ab
le
3
.
C
o
s
t m
a
tr
ix
P
r
e
d
i
c
t
e
d
C
l
a
ss
1
2
3
1
C
(
1
,
1
)
C
(
2
,
1
)
C
(
3
,
1
)
A
c
t
u
a
l
2
C
(
1
,
2
)
C
(
2
,
2
)
C
(
3
,
2
)
C
l
a
ss
3
C
(
1
,
3
)
C
(
2
,
3
)
C
(
3
,
3
)
T
ab
le
4
.
C
o
n
f
u
n
ctio
n
m
atr
i
x
P
r
e
d
i
c
t
e
d
C
l
a
ss
1
2
3
1
N
11
N
12
N
13
A
c
t
u
a
l
2
N
21
N
22
N
23
C
l
a
ss
3
N
31
N
32
N
33
T
h
e
p
er
f
o
r
m
a
n
ce
o
f
cla
s
s
i
f
ie
r
ca
n
b
e
s
ee
n
f
r
o
m
t
h
e
ac
c
u
r
ac
y
o
f
clas
s
if
icatio
n
,
w
h
er
e
th
e
h
i
g
h
e
r
class
i
f
icatio
n
ac
cu
r
ac
y
t
h
is
m
ea
n
s
th
e
b
etter
p
er
f
o
r
m
an
ce
o
f
th
e
clas
s
i
f
icatio
n
m
o
d
el
[
1
5
]
.
T
h
e
ac
cu
r
ac
y
is
a
test
i
n
g
m
e
th
o
d
b
ased
o
n
th
e
n
u
m
b
er
o
f
co
r
r
ec
t
r
esu
lt
p
r
ed
ictio
n
cla
s
s
i
f
icatio
n
.
T
h
e
eq
u
a
tio
n
f
o
r
ca
lcu
la
tin
g
th
e
v
al
u
e
o
f
ac
c
u
r
ac
y
i
s
s
h
o
w
n
as
f
o
llo
w
s
:
TP
A
c
c
u
ra
c
y
TP
TN
F
P
F
N
(
6
)
Misclas
s
i
f
icatio
n
co
s
t
i
s
t
h
e
v
alu
e
o
f
clas
s
p
r
ed
ictio
n
er
r
o
r
s
b
ec
au
s
e
t
h
e
cla
s
s
i
f
ier
ca
n
n
o
t
p
r
ed
ict
th
e
class
co
r
r
ec
tl
y
[
8
]
.
I
t
is
in
v
er
s
el
y
p
r
o
p
o
r
tio
n
al
to
th
e
a
cc
u
r
ac
y
v
a
lu
e
i
f
ac
cu
r
ac
y
v
alu
e
is
lar
g
e,
t
h
e
p
er
f
o
r
m
a
n
ce
o
f
t
h
e
clas
s
i
f
ier
is
g
o
o
d
class
i
f
icatio
n
b
u
t
i
f
th
e
co
s
t
h
a
s
lar
g
e
v
a
lu
e,
t
h
e
p
er
f
o
r
m
a
n
ce
o
f
t
h
e
class
i
f
ier
is
b
ad
class
i
f
icat
io
n
.
T
h
e
eq
u
atio
n
to
ca
lcu
late
t
h
e
co
s
t is s
h
o
w
n
as
f
o
llo
w
s
:
C
o
s
*
*
F
N
F
P
t
C
F
N
C
F
P
(
7
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8938
I
n
t J
A
r
ti
f
I
n
tell
,
Vo
l.
9
,
No
.
1
,
Ma
r
ch
20
20
:
65
–
72
70
2
.
9
.
T
ra
ini
ng
p
ro
ce
s
s
I
n
th
is
r
esear
c
h
,
t
h
er
e
ar
e
s
ev
er
al
s
tep
s
o
r
s
ta
g
es
o
f
r
es
ea
r
ch
,
s
u
ch
a
s
(
i)
d
ata
co
lle
ctio
n
,
(
ii)
p
r
ep
r
o
ce
s
s
in
g
,
(
iii)
attr
ib
u
tes
s
elec
tio
n
,
(
iv
)
tr
ain
i
n
g
an
d
test
i
n
g
p
r
o
ce
s
s
,
(
v
)
v
alid
atio
n
.
T
h
ese
a
r
e
th
e
m
e
th
o
d
s
o
lu
tio
n
s
tep
s
t
h
at
w
ill b
e
ca
r
r
ied
o
u
t to
s
o
lv
e
th
e
p
r
o
b
le
m
o
n
i
m
b
ala
n
ce
d
m
u
ltic
lass
d
ata.
Step
1
.
I
n
p
u
t th
e
d
ataset
f
o
r
p
r
o
ce
s
s
in
g
.
Step
2
.
R
ep
lacin
g
th
e
m
is
s
in
g
v
al
u
e
an
d
r
e
m
o
v
e
t
h
e
o
u
tli
n
er
f
r
o
m
th
e
d
ataset
a
n
d
p
er
f
o
r
m
th
e
m
i
n
-
m
a
x
n
o
r
m
aliza
t
io
n
p
r
o
ce
s
s
.
Step
3
.
Selectin
g
t
h
e
attr
ib
u
te
th
at
w
ill b
e
u
s
ed
u
s
in
g
P
SO.
Step
4
.
C
r
ea
tin
g
t
h
e
d
ec
is
io
n
t
r
ee
m
o
d
el
u
s
i
n
g
t
h
e
I
D3
,
C
4
.
5
,
C
5
.
0
alg
o
r
ith
m
.
Step
5
.
P
e
r
f
o
r
m
in
g
a
r
e
m
o
d
eli
n
g
o
f
th
e
d
ec
is
io
n
tr
ee
b
y
m
in
i
m
izi
n
g
t
h
e
co
s
t o
f
u
s
in
g
t
h
e
m
etac
o
s
t
m
et
h
o
d
.
Step
6
.
P
e
r
f
o
r
m
in
g
v
a
lid
atio
n
b
y
ca
lc
u
lati
n
g
th
e
v
al
u
e
o
f
ac
c
u
r
ac
y
a
n
d
to
tal
co
s
t.
3.
RE
SU
L
T
AND
DI
SCUS
SI
O
N
T
h
e
class
if
ier
m
o
d
el
test
i
n
g
i
s
ca
r
r
ied
o
u
t
u
s
i
n
g
a
n
I
n
tel
C
o
r
e
i5
p
e
r
s
o
n
al
co
m
p
u
ter
,
1
6
GB
R
AM
,
W
in
d
o
w
s
1
0
o
p
er
atin
g
s
y
s
te
m
a
n
d
R
ap
id
Mi
n
er
v
er
s
io
n
7
.
3
.
0
test
in
g
t
h
e
clas
s
if
icatio
n
m
o
d
el
is
ca
r
r
ied
o
u
t
u
s
i
n
g
5
d
ataset
s
o
f
UC
I
R
ep
o
s
ito
r
y
L
ea
r
n
in
g
R
ep
o
s
ito
r
y
(
Glass
,
L
y
m
p
o
g
r
af
i,
Vec
h
i
le,
T
h
y
r
o
id
an
d
W
i
n
e)
.
T
h
er
e
ar
e
th
r
ee
s
ce
n
ar
io
o
f
test
in
g
:
1.
P
er
f
o
r
m
a
n
ce
o
f
I
D3
,
C
4
.
5
,
an
d
C
5
.
0
.
2.
P
er
f
o
r
m
a
n
ce
of
I
D3
+
P
SO,
C
4
.
5
+
P
SO,
C
5
.
0
+
P
SO.
3.
P
er
f
o
r
m
a
n
ce
o
f
I
D3
+
P
SO +
Me
taC
o
s
t,
C
4
.
5
+
P
SO +
Me
taC
o
s
t,
C
5
.
0
+
P
SO +
Me
taC
o
s
t.
T
ab
le
5
is
th
e
r
esu
lt
o
f
test
i
n
g
p
er
f
o
r
m
an
ce
ac
c
u
r
ac
y
u
s
i
n
g
m
ac
h
in
e
lear
n
i
n
g
"
r
m
i
n
n
er
"
.
T
h
e
o
u
tp
u
t
o
f
m
ac
h
in
e
lear
n
i
n
g
is
a
tab
le
co
n
f
u
n
c
tio
n
m
atr
i
x
th
at
s
h
o
ws
th
e
p
er
f
o
r
m
an
ce
o
f
t
h
e
class
if
ier
.
T
h
is
r
esear
ch
u
s
e
s
3
test
i
n
g
s
ce
n
ar
io
s
.
T
h
e
f
ir
s
t
s
ce
n
ar
io
,
test
in
g
u
s
es
th
e
d
ec
is
io
n
tr
ee
alg
o
r
it
h
m
(
I
D3
,
C
4
.
5
,
an
d
C
5
.
0
)
.
T
h
e
s
ec
o
n
d
s
ce
n
ar
io
,
test
i
n
g
u
s
e
s
s
ce
n
ar
io
1
an
d
P
SO.
T
h
e
th
ir
d
s
ce
n
ar
io
,
test
i
n
g
u
s
es
s
ce
n
ar
io
2
an
d
t
h
e
m
etac
o
s
t
m
et
h
o
d
.
T
ab
le
5
.
R
esu
lt o
f
ac
c
u
ar
y
m
e
asu
r
e
m
en
t
D
A
TA
S
ET
I
D
3
C
4
.
5
C
5
.
0
I
D
3
I
D
3
+
P
S
O
I
D
3
+
P
S
O
+
M
ET
A
C
4
.
5
C
4
.
5
+
P
S
O
C
4
.
5
+
P
S
O
+
M
ET
A
C
5
.
0
C
5
.
0
+
P
S
O
C
5
.
0
+
P
S
O
+
M
ET
A
V
e
c
h
i
l
e
2
5
.
5
3
%
2
5
.
5
3
%
2
5
.
5
3
%
7
1
.
0
1
%
7
2
.
3
4
%
7
6
.
8
6
%
7
1
.
0
1
%
7
1
.
5
4
%
7
5
.
2
7
%
G
l
a
ss
3
2
.
7
1
%
3
2
.
7
1
%
3
2
.
7
1
%
6
8
.
2
2
%
6
8
.
2
2
%
7
0
.
0
9
%
6
7
.
2
9
%
7
0
.
0
9
%
7
6
.
1
7
%
L
y
mp
o
g
r
a
f
i
4
1
.
2
2
%
4
1
.
2
2
%
4
1
.
2
2
%
7
5
.
6
8
%
7
9
.
7
1
%
8
3
.
3
3
%
7
5
.
0
0
%
7
8
.
9
9
%
8
3
.
3
3
%
W
i
n
e
3
3
.
1
5
%
3
3
.
1
5
%
3
3
.
1
5
%
9
3
.
8
2
%
9
4
.
0
5
%
9
7
.
6
2
%
9
2
.
8
6
%
9
4
.
3
8
%
9
5
.
8
3
%
T
h
y
r
o
i
d
6
9
.
7
7
%
6
9
.
7
7
%
6
9
.
7
7
%
9
3
.
9
5
%
9
3
.
9
5
%
9
3
.
4
9
%
9
4
.
4
2
%
9
4
.
4
2
%
9
5
.
8
1
%
I
n
t
h
e
f
ir
s
t
s
ce
n
ar
io
,
(
1
)
I
D3
h
as
th
e
s
m
alle
s
t
p
er
f
o
r
m
an
ce
b
ec
au
s
e
it
is
n
o
t
ab
le
to
ce
r
ti
f
y
co
n
tin
u
e
d
ata.
(
2
)
T
esti
n
g
u
s
in
g
C
4
.
5
h
as
g
o
o
d
p
er
f
o
r
m
a
n
ce
i
n
d
ata
g
la
s
s
,
l
y
m
p
o
g
r
af
i
,
an
d
win
e
w
i
th
a
v
a
lu
e
o
f
6
8
.
2
2
%,
7
5
.
6
8
%,
an
d
9
3
.
8
2
%.
(
3
)
T
esti
n
g
u
s
i
n
g
C
5
.
0
h
a
s
g
o
o
d
p
er
f
o
r
m
an
ce
o
n
t
h
y
r
o
id
d
ata
w
it
h
a
n
ac
cu
r
ac
y
v
alu
e
o
f
9
4
.
4
2
%.
I
n
th
e
s
ec
o
n
d
s
ce
n
ar
io
,
(
1
)
P
SO
ca
n
n
o
t
i
m
p
r
o
v
e
I
D3
p
er
f
o
r
m
a
n
ce
.
(
2
)
PS
O
ca
n
i
m
p
r
o
v
e
th
e
p
er
f
o
r
m
a
n
ce
o
f
C
4
.
5
in
4
d
atasets
n
a
m
el
y
Vec
h
ile,
l
y
m
p
o
g
r
af
i,
w
i
n
e,
t
h
y
r
o
id
w
it
h
a
v
al
u
e
o
f
7
2
.
3
4
%,
7
9
.
7
1
%,
9
4
.
0
5
%,
9
3
.
9
5
%.
(
3
)
P
SO
ca
n
i
m
p
r
o
v
e
t
h
e
p
er
f
o
r
m
a
n
ce
o
f
C
5
.
0
in
all
d
atas
ets
w
i
th
a
v
al
u
e
o
f
7
1
.
5
4
%,
7
0
.
0
9
%,
7
8
.
9
9
%,
9
4
.
3
8
%,
9
4
.
4
2
%.
I
n
t
h
e
t
h
ir
d
s
c
en
ar
io
,
(
1
)
co
m
b
i
n
i
n
g
I
D3
an
d
Me
taC
o
s
t
ca
n
n
o
t
i
m
p
r
o
v
e
th
e
p
er
f
o
r
m
a
n
ce
o
f
t
h
e
class
i
f
ier
.
(
2
)
C
o
m
b
in
in
g
C
4
.
5
an
d
Me
taco
s
t
ca
n
i
m
p
r
o
v
e
th
e
p
er
f
o
r
m
an
ce
o
f
4
d
atasets
n
a
m
el
y
Vec
h
ile,
g
l
ass
,
l
y
m
p
o
g
r
ap
h
y
,
w
i
n
e
w
it
h
a
v
al
u
e
o
f
7
6
.
8
6
%,
7
0
.
0
9
%,
8
3
.
3
3
%,
an
d
9
7
.
6
2
%.
(
3
)
C
o
m
b
in
i
n
g
C
5
.
0
a
n
d
Me
ta
co
s
t c
an
i
m
p
r
o
v
e
all
d
ataset
s
w
it
h
v
al
u
e
s
o
f
7
5
.
2
7
%,
7
6
.
1
7
%,
8
3
.
3
3
%,
9
5
.
8
3
%,
9
5
.
8
1
% r
esp
ec
tiv
ely
.
Fig
u
r
e
2
is
a
r
ep
r
esen
tatio
n
o
f
T
a
b
le
5
w
h
er
e
t
h
e
b
est
p
er
f
o
r
m
an
ce
i
s
u
s
in
g
s
ce
n
ar
io
3
.
P
e
r
f
o
r
m
a
n
ce
C
4
.
5
h
as
t
h
e
h
ig
h
e
s
t
ac
cu
r
ac
y
v
a
lu
e
i
n
v
ec
h
ile
a
n
d
w
in
e
d
ata
w
it
h
a
v
al
u
e
o
f
7
6
.
8
6
%
an
d
9
7
.
6
2
%.
C
5
.
0
p
er
f
o
r
m
a
n
ce
h
a
s
th
e
h
ig
h
e
s
t
ac
cu
r
ac
y
v
a
lu
e
i
n
g
la
s
s
an
d
th
y
r
o
id
d
ata
w
it
h
a
v
alu
e
o
f
7
6
.
1
7
%,
9
5
.
8
1
%.
W
h
ile
th
e
l
y
m
p
o
g
r
ap
h
ic
d
ata
o
n
C
4
.
5
an
d
C
5
.
0
h
av
e
th
e
s
a
m
e
ac
cu
r
ac
y
v
alu
e
t
h
at
i
s
eq
u
al
to
8
3
.
3
3
%.
Fig
u
r
e
3
is
th
e
p
er
ce
n
ta
g
e
o
f
t
h
e
to
tal
v
al
u
e
o
f
all
d
ata
s
ets
o
f
th
e
ac
c
u
r
ac
y
v
al
u
e
o
f
s
ce
n
ar
io
3
.
Fro
m
th
e
p
er
ce
n
ta
g
e
o
f
1
0
0
%,
I
D3
h
as
a
p
er
f
o
r
m
a
n
c
e
p
er
ce
n
tag
e
o
f
1
9
.
2
3
%,
C
4
.
5
h
as
a
p
er
f
o
r
m
an
ce
p
er
ce
n
ta
g
e
o
f
4
0
.
2
4
%,
an
d
C
5
.
0
h
as a
p
er
f
o
r
m
an
ce
p
er
ce
n
tag
e
o
f
4
0
.
9
1
%.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
A
r
ti
f
I
n
tell
I
SS
N:
2252
-
8938
C
la
s
s
i
fica
tio
n
o
f m
u
lticla
s
s
im
b
a
la
n
ce
d
d
a
t
a
u
s
in
g
c
o
s
t
-
s
en
s
itive
d
ec
is
io
n
tr
ee
C
5
.
0
(
M.
A
l
d
iki
F
eb
r
ia
n
to
n
o
)
71
Fig
u
r
e
2
.
Diag
r
a
m
o
f
ac
cu
r
ac
y
co
m
p
ar
is
o
n
Fig
u
r
e
3
.
P
er
f
o
r
m
a
n
ce
Diag
r
a
m
T
ab
le
6
is
th
e
r
esu
lt
ca
lc
u
lati
n
g
co
s
t
o
f
th
e
clas
s
i
f
ier
.
C
o
s
t
is
th
e
v
a
lu
e
o
f
m
i
s
clas
s
if
icat
io
n
,
if
t
h
e
class
i
f
ier
h
a
s
lar
g
e
co
s
t
v
al
u
e,
th
e
class
i
f
ier
is
ca
lled
a
b
ad
c
lass
i
f
icatio
n
b
u
t
if
t
h
e
clas
s
i
f
i
er
h
as
a
s
m
all
co
s
t,
th
e
cla
s
s
i
f
ier
is
ca
lled
a
g
o
o
d
class
i
f
icatio
n
.
I
n
t
h
e
f
ir
s
t
s
ce
n
ar
io
,
I
D3
h
as
th
e
lar
g
e
s
t
co
s
t
v
al
u
e
o
f
all
alg
o
r
ith
m
s
.
C
4
.
5
h
a
s
s
m
al
l
c
o
s
t
v
al
u
e
i
n
d
ata
g
la
s
s
,
l
y
m
p
o
g
r
ap
h
y
a
n
d
w
i
n
e.
C
5
.
0
h
as
s
m
al
l
co
s
t
v
al
u
e
i
n
v
ec
h
i
le
an
d
th
y
r
o
id
d
ata.
I
n
t
h
e
s
ec
o
n
d
s
ce
n
ar
io
,
I
D3
is
t
h
e
w
o
r
s
t
c
lass
if
i
ca
tio
n
.
C
4
.
5
h
as
s
m
al
l
co
s
t
v
al
u
e
i
n
v
ec
h
i
le
a
n
d
l
y
m
p
o
g
r
ap
h
y
d
ata
.
C
5
.
0
h
as
s
m
all
co
s
t
v
al
u
e
i
n
d
ata
g
la
s
s
a
n
d
t
h
y
r
o
id
.
W
h
er
ea
s
C
4
.
5
an
d
C
5
.
0
h
av
e
t
h
e
s
a
m
e
co
s
t
i
n
w
in
e
d
a
ta
w
it
h
v
alu
e
s
o
f
2
0
0
.
I
n
th
e
th
ir
d
s
ce
n
ar
io
,
I
D3
is
th
e
w
o
r
s
t
class
i
f
icatio
n
.
C
4
.
5
h
as
s
m
all
co
s
t
v
al
u
e
i
n
d
ata
v
ec
h
ile
a
n
d
w
i
n
e.
C
5
.
0
h
a
s
s
m
all
co
s
t
v
alu
e
i
n
d
ata
g
lass
an
d
th
y
r
o
id
.
W
h
er
ea
s
C
4
.
5
an
d
C
5
.
0
h
av
e
th
e
s
a
m
e
co
s
t o
n
th
e
l
y
m
p
o
g
r
ap
h
ic
d
ata
w
it
h
v
alu
e
s
o
f
1
0
5
8
.
T
ab
le
7
s
h
o
w
s
t
h
e
ti
m
e
n
ee
d
ed
d
u
r
in
g
t
h
e
test
i
n
g
p
r
o
ce
s
s
in
m
ac
h
i
n
e
le
ar
n
i
n
g
.
T
h
e
d
a
ta
is
ta
k
e
n
d
u
r
in
g
th
e
test
i
n
g
p
r
o
ce
s
s
f
r
o
m
s
ce
n
ar
io
3
.
B
ased
o
n
th
e
test
in
g
o
f
th
e
5
d
atasets
,
C
5
.
0
h
as
a
f
aster
p
r
o
ce
s
s
in
g
ti
m
e
i
n
m
a
k
i
n
g
d
ec
is
io
n
s
th
a
n
C
4
.
5
.
Fig
u
r
e
4
s
h
o
w
s
a
g
r
ap
h
o
f
th
e
tr
ain
in
g
p
r
o
ce
s
s
ti
m
e.
B
lu
e
lin
e
i
s
th
e
li
n
e
f
r
o
m
t
h
e
C
4
.
5
p
r
o
ce
s
s
w
h
ile
r
ed
lin
e
is
t
h
e
li
n
e
f
r
o
m
th
e
C
5
.
0
p
r
o
ce
s
s
.
B
ased
o
n
th
e
g
r
ap
h
C
5
.
0
h
as
a
f
as
ter
p
r
o
ce
s
s
in
g
t
i
m
e
t
h
a
n
C
4
.
5
.
T
ab
le
6
.
T
h
e
m
ea
n
v
al
u
e
o
f
co
s
t o
f
t
h
e
s
u
m
m
ar
y
g
e
n
er
ated
.
D
A
TA
S
ET
I
D
3
C
4
.
5
C
5
.
0
I
D
3
I
D
3
+
P
S
O
I
D
3
+
P
S
O
+
M
ET
A
C
4
.
5
C
4
.
5
+
P
S
O
C
4
.
5
+
P
S
O
+
M
ET
A
C
5
.
0
C
5
.
0
+
P
S
O
C
5
.
0
+
P
S
O
+
M
ET
A
V
e
c
h
i
l
e
1
5
6
8
0
0
1
5
6
8
0
0
1
5
6
8
0
0
2
3
7
6
2
2
1
6
3
2
1
5
1
3
8
2
2
8
9
8
2
3
7
6
2
1
7
2
9
8
G
l
a
ss
4
1
4
7
2
4
1
4
7
2
4
1
4
7
2
9
2
4
8
9
2
4
8
8
1
9
2
9
8
0
0
8
1
9
2
5
2
0
2
L
y
mp
o
g
r
a
f
i
1
5
1
3
8
1
5
1
3
8
1
5
1
3
8
2
5
9
2
1
5
6
8
1
0
5
8
2
7
3
8
1
6
8
2
1
0
5
8
W
i
n
e
2
8
3
2
2
2
8
3
2
2
2
8
3
2
2
2
4
2
2
0
0
32
2
8
8
2
0
0
98
t
h
y
r
o
i
d
8
4
5
0
8
4
5
0
8
4
5
0
3
3
8
3
3
8
3
9
2
2
8
8
2
8
8
1
6
2
T
ab
le
7
.
T
im
e
o
f
p
r
o
ce
s
s
tr
ain
in
g
C
4
.
5
+
P
S
O
+
M
ET
A
C
5
.
0
+
P
S
O
+
M
ET
A
V
e
c
h
i
l
e
1
2
4
s
1
0
1
s
G
l
a
ss
3
1
s
2
2
s
L
y
mp
o
g
r
a
f
i
2
8
s
1
8
s
W
i
n
e
6
s
6
s
T
h
y
r
o
i
d
7
s
5
s
Fig
u
r
e
4
.
C
o
m
p
ar
is
o
n
ti
m
e
o
f
tr
ain
i
n
g
p
r
o
ce
s
s
4.
CO
NCLU
SI
O
N
T
h
is
p
ap
er
is
co
m
p
i
led
b
ased
o
n
t
h
e
liter
at
u
r
e
a
n
d
t
h
eo
r
etica
l
co
n
ce
p
ts
o
f
d
ata
m
in
i
n
g
a
n
d
m
ac
h
in
e
lear
n
in
g
to
f
in
d
a
s
o
lu
tio
n
to
h
an
d
le
m
u
lti
-
clas
s
i
m
b
ala
n
ce
d
d
ata
p
r
o
b
lem
s
.
I
n
t
h
i
s
r
ese
ar
ch
3
p
r
o
o
f
s
w
er
e
co
n
d
u
cted
.
First
p
r
o
o
f
,
u
s
in
g
s
ce
n
ar
io
2
,
P
SO
ca
n
im
p
r
o
v
e
th
e
p
er
f
o
r
m
an
ce
o
f
C
5
.
0
as
s
h
o
w
n
i
n
T
ab
le
5
.
T
h
e
s
ec
o
n
d
p
r
o
o
f
,
u
s
i
n
g
s
ce
n
a
r
io
3
,
C
5
.
0
h
as
a
b
etter
p
er
f
o
r
m
an
ce
th
a
n
I
D3
a
n
d
C
4
.
5
.
C
5
.
0
h
as
a
p
er
ce
n
ta
g
e
o
f
p
er
f
o
r
m
a
n
ce
o
f
4
0
.
9
1
%
w
h
ile
C
4
.
5
an
d
I
D3
ar
e
4
0
.
2
4
%
an
d
1
9
.
2
3
%
as
s
h
o
w
n
i
n
Fig
u
r
e
3
.
T
h
e
th
ir
d
p
r
o
o
f
,
b
ased
o
n
T
ab
le
7
,
C
5
.
0
h
as
a
f
aster
t
i
m
e
i
n
m
a
k
in
g
d
e
cisi
o
n
s
th
a
n
C
4
.
5
.
B
ased
o
n
th
ese
th
r
ee
p
r
o
o
f
s
,
it
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8938
I
n
t J
A
r
ti
f
I
n
tell
,
Vo
l.
9
,
No
.
1
,
Ma
r
ch
20
20
:
65
–
72
72
ca
n
b
e
co
n
cl
u
d
ed
th
at
co
s
t
s
e
n
s
it
iv
e
d
ec
i
s
io
n
tr
ee
C
5
.
0
m
e
t
h
o
d
h
as
a
b
etter
p
er
f
o
r
m
a
n
ce
th
an
u
s
i
n
g
I
D3
an
d
C
4
.
5
f
o
r
s
o
lv
in
g
m
u
lticla
s
s
i
m
b
alan
ce
d
d
ata
p
r
o
b
lem
s
.
I
n
f
u
t
u
r
e
r
esear
ch
,
C
5
.
0
ca
n
b
e
co
m
b
in
ed
w
i
th
t
h
e
co
s
t
s
en
s
iti
v
e
lear
n
in
g
m
et
h
o
d
u
s
in
g
Me
ta
lear
n
i
n
g
s
a
m
p
li
n
g
tech
n
iq
u
e
s
s
u
c
h
as
co
s
ti
n
g
an
d
w
e
ig
h
tin
g
to
o
b
tain
a
m
o
r
e
o
p
ti
m
al
clas
s
if
ier
p
er
f
o
r
m
an
ce
.
RE
F
E
R
E
NC
E
S
[1
]
H.
A
li
,
M
.
N.
M
.
S
a
ll
e
h
,
R.
S
a
e
d
u
d
i
n
,
a
n
d
K.
Hu
ss
a
in
,
“
Im
b
a
l
a
n
c
e
c
las
s
p
ro
b
lem
s
in
d
a
ta
m
in
in
g
:
a
re
v
ie
w
,
”
In
d
o
n
e
s.
J
.
E
lec
tr.
En
g
.
C
o
mp
u
t.
S
c
i.
,
v
o
l
.
1
4
,
n
o
.
3
,
p
p
.
1
5
6
0
-
1
5
7
1
,
2
0
1
9
.
[2
]
H.
He
a
n
d
E.
A
.
G
a
rc
ia,
“
L
e
a
rn
in
g
f
ro
m
i
m
b
a
lan
c
e
d
d
a
ta,”
IEE
E
T
ra
n
s.
Kn
o
wl.
Da
ta
E
n
g
.
,
v
o
l.
2
1
,
n
o
.
9
,
p
p
.
1
2
6
3
-
1
2
8
4
,
2
0
0
9
.
[3
]
S
.
W
a
n
g
a
n
d
X
.
Ya
o
,
“
M
u
lt
icla
ss
I
m
b
a
lan
c
e
P
ro
b
lem
s:
A
n
a
l
y
sis
a
n
d
P
o
ten
ti
a
l
S
o
lu
ti
o
n
s,”
IEE
E
T
r
a
n
s.
S
y
st.
M
a
n
.
Cy
b
e
rn
,
v
o
l.
4
2
,
n
o
.
4
,
p
p
.
1
1
1
9
–
1
1
3
0
,
2
0
1
2
.
[4
]
R.
L
o
n
g
a
d
g
e
,
S
.
S
.
Do
n
g
re
,
a
n
d
L
.
M
a
li
k
,
“
Clas
s
i
m
b
a
lan
c
e
p
ro
b
l
e
m
in
d
a
ta
m
in
in
g
:
re
v
ie
w
,
”
In
t.
J
.
Co
mp
u
t
.
S
c
i
.
Ne
tw.
,
v
o
l.
2
,
n
o
.
1
,
p
p
.
8
3
–
8
7
,
2
0
1
3
.
[5
]
N.
V
.
Ch
a
w
la,
A
.
L
a
z
a
r
e
v
ic,
L
.
O.
Ha
ll
,
a
n
d
K.
W
.
Bo
wy
e
r,
“
S
M
OT
EBo
o
st:
Im
p
ro
v
in
g
P
re
d
ictio
n
o
f
th
e
M
in
o
rit
y
Clas
s in
Bo
o
stin
g
,
”
Pro
c
e
e
d
i
n
g
.
Kn
o
wl.
b
a
se
Disc
o
v
.
Da
t
a
b
a
se
s,
p
p
.
1
0
7
–
1
1
9
,
2
0
0
3
.
[6
]
A
.
F
re
it
a
s,
A
.
d
a
Co
sta
P
e
re
ira,
a
n
d
P
.
Bra
z
d
il
,
“
C
o
st
-
S
e
n
siti
v
e
De
c
isio
n
T
re
e
s
A
p
p
li
e
d
t
o
M
e
d
ica
l
Da
ta.,
”
Da
W
a
K
,
v
o
l.
4
6
5
4
,
p
p
.
3
0
3
–
3
1
2
,
2
0
0
7
.
[7
]
A
.
M
o
tw
a
n
i,
P
.
C
h
a
u
ra
siy
a
,
G
.
Ba
jaj.
“
P
re
d
icti
n
g
Cre
d
it
W
o
rth
i
n
e
ss
o
f
Ba
n
k
Cu
sto
m
e
r
w
it
h
M
a
c
h
in
e
L
e
a
rn
in
g
o
v
e
r
Clo
u
d
,
”
IJCSE
,
v
o
l.
6
,
No
.
7
p
p
.
2
3
4
7
–
2
6
9
3
.
2
0
1
8
.
[8
]
C.
W
a
n
g
,
R.
L
i
a
n
d
Z.
C
h
e
n
.
“
P
a
rti
ti
o
n
c
o
st
-
se
n
siti
v
e
CA
RT
b
a
se
d
o
n
c
u
sto
m
e
r
v
a
lu
e
f
o
r
Tele
c
o
m
c
u
sto
m
e
r
c
h
u
rn
p
re
d
ictio
n
,
”
Pr
o
c
e
e
d
in
g
s
o
f
th
e
3
6
th
C
h
i
n
e
se
Co
n
tro
l
Co
n
fer
e
n
c
e
,
2
0
1
7
.
[9
]
N.
S
a
n
t
o
so
,
W
.
W
ib
o
w
o
,
H.
Hil
m
a
w
a
ti
,
“
In
teg
ra
ti
o
n
o
f
sy
n
th
e
ti
c
m
in
o
rit
y
o
v
e
rsa
m
p
li
n
g
tec
h
n
iq
u
e
f
o
r
im
b
a
lan
c
e
d
c
las
s,”
In
d
o
n
e
s.
J
.
El
e
c
tr.
E
n
g
.
C
o
mp
u
t
.
S
c
i
.
,
v
o
l.
1
3
,
n
o
.
1
,
p
p
.
1
0
2
–
1
0
8
,
2
0
1
9
.
[1
0
]
S
.
Ba
ru
a
,
M
.
Isla
m
,
X
.
Ya
o
,
a
n
d
K.
M
u
ra
se
,
“
M
W
M
OT
E
-
M
a
jo
rit
y
W
e
ig
h
ted
M
in
o
rit
y
Ov
e
rsa
m
p
li
n
g
T
e
c
h
n
iq
u
e
f
o
r
I
m
b
a
lan
c
e
d
Da
ta S
e
t
L
e
a
rn
in
g
,
”
IEE
E
T
r
a
n
s.
K
n
o
w
l.
Da
ta
E
n
g
.
,
v
o
l.
2
6
,
n
o
.
2
,
p
p
.
4
0
5
-
4
2
5
,
2
0
1
4
.
[1
1
]
G
U
Qio
n
g
,
e
t
a
l.
“
A
n
im
p
ro
v
e
d
S
M
OT
E
a
lg
o
rit
h
m
b
a
se
d
o
n
g
e
n
e
ti
c
a
lg
o
rit
h
m
f
o
r
im
b
a
lan
c
e
d
d
a
ta
c
las
sif
ic
a
ti
o
n
,
”
J
Dig
In
f
M
a
n
a
g
,
v
o
l.
1
4
,
n
o
.
2
,
p
p
.
9
2
–
1
0
3
,
2
0
1
6
.
[1
2
]
K.
P
o
lata
n
d
S
.
G
u
n
e
s.
“
A
n
o
v
e
l
h
y
b
rid
in
telli
g
e
n
t
m
e
th
o
d
b
a
se
d
o
n
C4
.
5
d
e
c
isio
n
tree
c
las
sif
i
e
r
a
n
d
o
n
e
-
a
g
a
in
st
-
a
ll
a
p
p
ro
a
c
h
f
o
r
m
u
lt
i
-
c
las
sc
las
si
f
ica
ti
o
n
p
r
o
b
lem
s,”
Exp
e
rt S
y
ste
ms
wit
h
A
p
p
l
ica
ti
o
n
s
,
v
o
l.
36
,
p
p
.
1
5
8
7
-
1
5
9
2
,
2
0
0
9
.
[1
3
]
H.
He
a
n
d
E.
A
.
G
a
rc
ia,
“
L
e
a
rn
in
g
f
ro
m
i
m
b
a
lan
c
e
d
d
a
ta,”
IEE
E
T
ra
n
s.
Kn
o
wl.
Da
ta
E
n
g
.
,
v
o
l.
2
1
,
n
o
.
9
,
p
p
.
1
2
6
3
-
1
2
8
4
,
2
0
0
9
.
[1
4
]
S
.
A
.
S
h
il
b
a
y
e
h
.
“
Co
st S
e
n
siti
v
e
M
e
ta L
e
a
rn
in
g
,
”
S
u
b
m
it
ted
in
Pa
rtia
l
Fu
lf
il
me
n
t
o
f
t
h
e
Re
q
u
ire
me
n
ts o
f
t
h
e
De
g
re
e
o
f
Do
c
t
o
r o
f
Ph
i
lo
so
p
h
y
,
2
0
1
5
.
[1
5
]
A
.
N.
H
a
ld
a
n
k
a
r.
“
A
Co
st
S
e
n
si
ti
v
e
c
las
sif
i
e
r
f
o
r
Big
D
a
ta.”
I
EE
E
In
ter
n
a
ti
o
n
a
l
C
o
n
fer
e
n
c
e
o
n
Ad
v
a
n
c
e
s
in
El
e
c
tro
n
ics
,
Co
mm
u
n
ica
ti
o
n
a
n
d
Co
mp
u
ter
T
e
c
h
n
o
lo
g
y
(
ICAE
CCT
),
2
0
1
6
.
[1
6
]
C.
X.
L
in
g
,
V.
S
.
S
h
e
n
g
,
a
n
d
Q.
Ya
n
g
,
“
T
e
st
stra
teg
ies
f
o
r
c
o
st
-
se
n
siti
v
e
d
e
c
isio
n
tree
s,”
IEE
E
T
ra
n
s.
Kn
o
wl.
D
a
t
a
En
g
.
,
v
o
l.
1
8
,
n
o
.
8
,
p
p
.
1
0
5
5
-
1
0
6
7
,
2
0
0
6
.
[1
7
]
F
.
Ja
u
h
a
ri,
A
.
A
.
S
u
p
ia
n
to
,
“
Bu
il
d
in
g
st
u
d
e
n
t’s
p
e
rf
o
rm
a
n
c
e
d
e
c
isio
n
tree
c
las
sif
ier
u
sin
g
b
o
o
st
in
g
a
lg
o
rit
h
m
,
”
In
d
o
n
e
s.
J
.
E
lec
tr.
En
g
.
C
o
mp
u
t.
S
c
i.
,
v
o
l
.
1
4
,
n
o
.
3
,
p
p
.
1
2
9
8
–
1
3
0
4
,
2
0
1
9
.
[1
8
]
C.
X.
L
in
g
,
V.
S
.
S
h
e
n
g
,
“
Co
st
S
e
n
siti
v
e
L
e
a
rn
in
g
a
n
d
th
e
Clas
s
Im
b
a
lan
c
e
p
ro
b
lem
,
”
En
c
y
c
lo
p
e
d
ia
o
f
m
a
c
h
in
e
lea
rn
in
g
.
S
p
i
n
g
e
r
,
2
0
0
8
.
[1
9
]
B.
Ba
y
k
a
ra
.
“
I
m
p
a
c
t
o
f
Ev
a
lu
a
ti
o
n
M
e
th
o
d
s
o
n
De
c
isio
n
T
re
e
A
c
c
u
ra
c
y
,
”
M
.
S
c
.
th
e
sis
,
p
p
.
7
2
,
6
in
d
e
x
p
a
g
e
s,
2
0
1
5
.
[2
0
]
A
.
S
.
G
a
lath
i
y
a
.
“
I
m
p
ro
v
e
d
De
c
i
sio
n
T
re
e
In
d
u
c
ti
o
n
A
lg
o
rit
h
m
w
it
h
F
e
a
tu
re
S
e
lec
ti
o
n
,
Cr
o
ss
Va
li
d
a
ti
o
n
,
M
o
d
e
l
Co
m
p
lex
it
y
a
n
d
Re
d
u
c
e
d
Err
o
r
P
ru
n
i
n
g
,
”
(
IJ
CS
IT
)
Vo
l.
3
(2
)
,
p
p
.
3
4
2
7
-
3
4
3
1
,
2
0
1
6
.
[2
1
]
R.
Ba
tu
w
it
a
a
n
d
V
.
P
a
lad
e
,
“
A
n
e
w
p
e
r
f
o
r
m
a
n
c
e
m
e
a
su
re
f
o
r
c
las
s
i
m
b
a
lan
c
e
le
a
rn
in
g
.
A
p
p
li
c
a
ti
o
n
t
o
b
io
i
n
f
o
rm
a
ti
c
s p
ro
b
lem
s,”
8
th
In
t.
Co
n
f
.
M
a
c
h
.
L
e
a
rn
.
A
p
p
l.
ICM
L
A
2
0
0
9
,
p
p
.
5
4
5
–
5
5
0
,
2
0
0
9
.
[2
2
]
P.
Do
m
in
g
o
s.
“
M
e
tac
o
st:
A
g
e
n
e
ra
l
m
e
th
o
d
f
o
r
m
a
k
in
g
c
la
ss
i
fiers
c
o
st
-
se
n
siti
v
e
.
”
In
Pro
c
.
o
f
th
e
fi
ft
h
ACM
S
IGKD
D
in
ter
n
a
ti
o
n
a
l
c
o
n
fer
e
n
c
e
o
n
K
n
o
wled
g
e
d
isc
o
v
e
ry
a
n
d
d
a
ta
mi
n
i
n
g
,
ACM
Ne
w
Y
o
rk
,
NY
,
US
A
,
p
p
1
5
5
–
1
6
4
,
1
9
9
9
.
[2
3
]
S
.
W
a
n
g
a
n
d
X
.
Ya
o
,
“
M
u
lt
icla
ss
I
m
b
a
lan
c
e
P
ro
b
lem
s:
A
n
a
l
y
sis
a
n
d
P
o
ten
ti
a
l
S
o
lu
ti
o
n
s,”
IEE
E
T
r
a
n
s.
S
y
st.
M
a
n
.
Cy
b
e
rn
.
,
v
o
l.
4
2
,
n
o
.
4
,
p
p
.
1
1
1
9
–
1
1
3
0
,
2
0
1
2
.
[2
4
]
S
.
Zh
a
n
g
,
C.
Zh
a
n
g
a
n
d
Q.
Ya
n
g
.
“
Da
ta
p
re
p
a
ra
ti
o
n
f
o
r
d
a
ta
m
in
in
g
.
”
Ap
p
li
e
d
Arti
fi
c
i
a
l
In
telli
g
e
n
c
e
a
n
In
ter
n
a
t
io
n
a
l
J
o
u
rn
a
l,
Vo
l
.
1
7
,
p
p
.
5
-
6
p
,
2
0
1
0
.
[2
5
]
N.
T
.
Ng
h
e
,
Z.
G
a
n
tn
e
r
,
a
n
d
L
.
S
.
T
h
iem
e
.
“
Co
st
-
se
n
siti
v
e
lea
rn
in
g
m
e
th
o
d
s
f
o
r
im
b
a
lan
c
e
d
d
a
ta
,”
In
t
h
e
2
0
1
0
In
ter
n
a
t
io
n
a
l
jo
i
n
t
c
o
n
fer
e
n
c
e
o
n
l
n
e
two
rk
s (
IJ
CNN)
,
p
p
.
1
–
8
,
1
8
–
2
3
,
2
0
1
0
.
[2
6
]
L
.
Yin
,
Y.
Ge
,
K.
X
iao
,
X.
W
a
n
g
,
a
n
d
X
.
Q
u
a
n
,
“
F
e
a
tu
re
se
lec
ti
o
n
f
o
r
h
ig
h
-
d
im
e
n
sio
n
a
l
im
b
a
lan
c
e
d
d
a
ta,”
Ne
u
ro
c
o
mp
u
ti
n
g
,
v
o
l
.
1
0
5
,
p
p
.
3
-
1
1
,
2
0
1
3
.
[2
7
]
B.
N.
P
a
tel,
S
.
G
.
P
ra
jap
a
ti
a
n
d
K.I.
L
a
k
h
taria
,
“
Eff
icie
n
t
Clas
si
fica
ti
o
n
o
f
Da
ta
Us
in
g
De
c
isio
n
T
re
e
,
”
Bo
n
frin
g
in
ter
n
a
t
io
n
a
l
jo
u
rn
a
l
o
f
d
a
ta
min
i
n
g
,
Vo
l.
2
,
N
o
.
1
,
2
0
1
2
.
[2
8
]
N.
S
a
n
t
o
so
,
W
.
W
ib
o
w
o
,
H.
Hi
l
m
a
w
a
ti
,
“
In
teg
ra
ti
o
n
o
f
sy
n
th
e
ti
c
m
in
o
rit
y
o
v
e
rsa
m
p
li
n
g
tec
h
n
iq
u
e
f
o
r
im
b
a
lan
c
e
d
c
las
s,”
In
d
o
n
e
s.
J
.
El
e
c
tr.
E
n
g
.
C
o
mp
u
t
.
S
c
i
.
,
v
o
l.
1
3
,
n
o
.
1
,
p
p
.
1
0
2
-
1
0
8
,
2
0
1
9
.
Evaluation Warning : The document was created with Spire.PDF for Python.