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.
1
4
,
No
.
6
,
Dec
em
b
er
2
0
2
5
,
p
p
.
4
7
8
7
~
4
8
0
1
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
14
.i
6
.
p
p
4
7
8
7
-
4
8
0
1
4787
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
Enha
ncing
so
ft
wa
re f
a
ult
predi
ctio
n t
hro
ug
h data
b
a
la
ncing
techniqu
es a
nd m
a
chine learning
Ak
s
ha
t
Ra
j
,
Durv
a
M
a
ha
de
o
Cha
v
a
n,
P
riy
a
l A
g
a
rwa
l,
J
estin
G
ig
i
,
Ma
d
h
u
r
i
Ra
o
,
Vin
a
y
a
k
Mu
sa
le
,
Aksh
ita
Ch
a
n
c
h
l
a
n
i,
Mu
r
ta
z
a
S
h
a
b
b
irbh
a
i
Dho
l
k
a
w
a
la
,
K
u
la
m
a
la
Vi
n
o
d
K
u
m
a
r
D
e
p
a
r
t
me
n
t
o
f
C
o
mp
u
t
e
r
E
n
g
i
n
e
e
r
i
n
g
a
n
d
Te
c
h
n
o
l
o
g
y
,
D
r
.
V
i
s
h
w
a
n
a
t
h
K
a
r
a
d
M
I
T
W
o
r
l
d
P
e
a
c
e
U
n
i
v
e
r
s
i
t
y
,
P
u
n
e
,
I
n
d
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ma
y
6
,
2
0
2
4
R
ev
is
ed
Oct
2
7
,
2
0
2
5
Acc
ep
ted
No
v
8
,
2
0
2
5
S
o
ftwa
re
fa
u
lt
p
re
d
ictio
n
is
e
ss
e
n
ti
a
l
fo
r
e
n
s
u
rin
g
th
e
re
li
a
b
i
li
ty
a
n
d
q
u
a
li
t
y
o
f
so
ftwa
re
sy
ste
m
s b
y
id
e
n
ti
f
y
in
g
p
o
te
n
ti
a
l
d
e
fe
c
ts ea
rly
in
th
e
d
e
v
e
lo
p
m
e
n
t
li
fe
c
y
c
le.
Ho
we
v
e
r,
t
h
e
p
re
se
n
c
e
o
f
imb
a
lan
c
e
d
d
a
tas
e
ts
p
o
se
s
a
sig
n
ifi
c
a
n
t
c
h
a
ll
e
n
g
e
t
o
t
h
e
e
ffe
c
ti
v
e
n
e
ss
o
f
fa
u
lt
p
re
d
icti
o
n
m
o
d
e
ls.
I
n
t
h
is
p
a
p
e
r,
we
in
v
e
stig
a
te
t
h
e
imp
a
c
t
o
f
d
iffer
e
n
t
d
a
ta
b
a
lan
c
in
g
tec
h
n
iq
u
e
s,
in
c
lu
d
in
g
g
e
n
e
ra
ti
v
e
a
d
v
e
rsa
rial
n
e
two
r
k
s
(G
AN
s),
sy
n
th
e
ti
c
m
in
o
r
it
y
o
v
e
r
-
sa
m
p
li
n
g
tec
h
n
iq
u
e
(S
M
OTE)
,
a
n
d
Ne
a
rM
iss,
o
n
m
a
c
h
in
e
lea
rn
in
g
(M
L)
m
o
d
e
l
p
e
rfo
rm
a
n
c
e
fo
r
s
o
ftwa
re
fa
u
lt
p
re
d
ictio
n
.
T
h
ro
u
g
h
a
c
o
m
p
a
ra
ti
v
e
a
n
a
ly
sis
a
c
ro
ss
m
u
lt
ip
le
d
a
tas
e
ts
c
o
m
m
o
n
ly
u
se
d
in
so
ftwa
re
e
n
g
in
e
e
rin
g
re
se
a
rc
h
,
we
e
v
a
lu
a
te
th
e
e
ffica
c
y
o
f
t
h
e
se
tec
h
n
iq
u
e
s
in
a
d
d
re
ss
in
g
c
las
s
imb
a
lan
c
e
a
n
d
imp
r
o
v
i
n
g
p
re
d
icti
v
e
a
c
c
u
ra
c
y
.
Ou
r
fi
n
d
i
n
g
s
p
ro
v
i
d
e
in
si
g
h
t
s
in
to
th
e
m
o
st
e
ffe
c
ti
v
e
a
p
p
ro
a
c
h
e
s
fo
r
h
a
n
d
li
n
g
imb
a
lan
c
e
d
d
a
ta
in
so
ft
wa
re
fa
u
lt
p
re
d
ictio
n
tas
k
s,
th
e
re
b
y
a
d
v
a
n
c
in
g
th
e
sta
te
-
of
-
t
h
e
-
a
rt
i
n
so
ftwa
re
e
n
g
in
e
e
rin
g
re
se
a
rc
h
a
n
d
p
ra
c
ti
c
e
.
An
e
x
ten
siv
e
e
x
p
e
rime
n
tatio
n
is
p
e
rfo
rm
e
d
a
n
d
a
n
a
ly
z
e
d
i
n
t
h
is
stu
d
y
h
e
re
th
a
t
in
c
l
u
d
e
s
8
d
a
tas
e
ts,
4
d
a
ta
b
a
lan
c
in
g
tec
h
n
iq
u
e
s
,
a
n
d
4
ML
tec
h
n
iq
u
e
s
in
o
rd
e
r
to
d
e
m
o
n
stra
te
th
e
e
ffica
c
y
o
f
v
a
rio
u
s m
o
d
e
ls i
n
s
o
ft
wa
re
fa
u
lt
p
re
d
icti
o
n
.
K
ey
w
o
r
d
s
:
G
e
n
e
r
a
t
i
v
e
a
d
v
e
r
s
a
r
i
al
n
e
t
w
o
r
k
s
I
m
b
alan
ce
d
d
ata
Nea
r
Miss
SMOT
E
So
f
twar
e
f
au
lt p
r
e
d
ictio
n
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Ma
d
h
u
r
i Rao
Dep
ar
tm
e
n
t
o
f
C
o
m
p
u
t
er
E
n
g
i
n
e
er
in
g
a
n
d
T
ec
h
n
o
l
o
g
y
,
Dr
.
Vis
h
w
a
n
at
h
Ka
r
a
d
M
I
T
W
o
r
l
d
Pe
ac
e
U
n
i
v
e
r
s
i
ty
Ko
th
r
u
d
,
Pu
n
e,
Ma
h
a
r
ash
tr
a,
I
n
d
ia
E
m
ail: m
ad
h
u
r
i.
r
ao
@
m
itwp
u
.
ed
u
.
in
1.
I
NT
RO
D
UCT
I
O
N
So
f
twar
e
f
au
lt
p
r
ed
ictio
n
aim
s
at
id
en
tify
in
g
im
p
e
n
d
in
g
f
a
u
lts
o
r
b
u
g
s
in
s
o
f
twar
e
m
o
d
u
les
b
ef
o
r
e
th
ey
ar
e
v
is
ib
le
o
p
e
r
atio
n
al
i
s
s
u
es,
th
er
eb
y
en
h
an
cin
g
th
e
q
u
ality
o
f
t
h
e
b
u
ilt
s
o
f
twar
e
[
1
]
,
[
2
]
.
W
ith
th
e
in
cr
ea
s
in
g
co
m
p
lex
it
y
o
f
s
o
f
t
war
e
s
y
s
tem
s
an
d
th
e
d
em
an
d
f
o
r
r
eliab
le
an
d
ef
f
icien
t
s
o
f
tw
ar
e,
th
e
im
p
o
r
ta
n
ce
o
f
ac
cu
r
ate
f
au
lt
p
r
ed
ictio
n
tech
n
iq
u
es
ca
n
n
o
t
b
e
o
v
er
s
tated
.
T
r
ad
itio
n
al
m
eth
o
d
s
o
f
f
au
lt
p
r
ed
ictio
n
o
f
te
n
r
ely
o
n
m
ac
h
in
e
lear
n
in
g
(
ML
)
m
o
d
els
tr
ain
ed
o
n
h
is
to
r
ical
d
ata
to
class
if
y
s
o
f
twar
e
m
o
d
u
les
as
d
ef
ec
tiv
e
o
r
non
-
d
ef
ec
tiv
e
b
ased
o
n
v
ar
i
o
u
s
co
d
e
m
etr
ics
an
d
ch
ar
ac
te
r
is
tics
[
3
]
–
[
6
]
.
Ho
wev
er
,
o
n
e
o
f
th
e
s
ig
n
if
ican
t
ch
allen
g
es
in
s
o
f
twar
e
f
a
u
lt
p
r
ed
ictio
n
is
d
ea
li
n
g
with
im
b
a
lan
ce
d
d
atasets
.
I
m
b
alan
ce
d
d
atasets
o
cc
u
r
wh
en
th
e
class
es
o
f
in
ter
est
(
d
ef
ec
t
iv
e
v
s
.
n
o
n
-
d
e
f
ec
tiv
e
m
o
d
u
le
s
)
ar
e
n
o
t
e
v
en
ly
d
is
tr
ib
u
ted
,
lead
in
g
to
b
iased
m
o
d
el
p
er
f
o
r
m
an
ce
[
7
]
,
[
8
]
.
T
h
e
r
esear
ch
aim
o
f
th
is
s
tu
d
y
i
s
to
in
v
esti
g
ate
th
e
e
f
f
ec
tiv
en
e
s
s
o
f
d
if
f
er
en
t
d
ata
b
alan
cin
g
tech
n
i
q
u
es in
im
p
r
o
v
in
g
ML
m
o
d
el
p
er
f
o
r
m
an
ce
f
o
r
s
o
f
twar
e
f
au
lt p
r
e
d
ictio
n
.
S
p
ec
if
ically
,
we
will
ex
p
lo
r
e
tech
n
iq
u
es
s
u
ch
as
g
en
er
ativ
e
ad
v
er
s
ar
ial
n
etwo
r
k
s
(
GANs)
[
9
]
–
[
1
1
]
,
s
y
n
th
e
tic
m
in
o
r
ity
o
v
er
-
s
am
p
lin
g
tech
n
iq
u
e
(
SMOT
E
)
[
1
2
]
,
[
1
3
]
,
a
n
d
Nea
r
Miss
[
1
4
]
t
o
ad
d
r
ess
th
e
im
b
alan
ce
in
th
e
d
ataset
.
B
y
co
m
p
ar
in
g
th
e
p
er
f
o
r
m
an
ce
o
f
ML
m
o
d
els
tr
ain
ed
o
n
b
alan
ce
d
an
d
u
n
b
alan
ce
d
d
a
tasets
,
we
s
ee
k
to
id
en
tify
th
e
m
o
s
t
ef
f
ec
tiv
e
ap
p
r
o
ac
h
f
o
r
h
a
n
d
lin
g
im
b
alan
ce
d
d
ata
in
s
o
f
twar
e
f
au
lt
p
r
ed
ictio
n
task
s
.
ML
tech
n
iq
u
es
co
u
ld
b
e
clas
s
if
ied
as
s
u
p
er
v
is
ed
,
u
n
s
u
p
er
v
is
ed
an
d
s
em
i
-
s
u
p
er
v
is
ed
.
ML
tech
n
iq
u
es
h
a
v
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
14
,
No
.
6
,
Dec
em
b
er
2
0
2
5
:
4
7
8
7
-
4
8
0
1
4788
b
ee
n
v
er
y
u
s
ef
u
l
in
d
etec
tin
g
i
s
s
u
es
an
d
b
u
g
s
in
s
o
f
twar
e
[
1
5
]
,
[
1
6
]
.
H
o
wev
er
,
t
h
e
ef
f
icac
y
o
f
ML
tech
n
iq
u
es
d
ep
en
d
s
o
n
th
e
n
at
u
r
e
o
f
d
atas
ets
th
at
ar
e
o
f
ten
lin
k
ed
to
b
e
im
b
alan
ce
d
.
Ad
d
r
ess
in
g
is
s
u
es
o
f
d
ata
im
b
alan
ce
is
th
er
ef
o
r
e
h
i
g
h
ly
ess
en
tial
[
1
7
]
in
ev
e
r
y
ar
ea
wh
er
e
d
ata
d
is
tr
ib
u
tio
n
h
as a
s
ig
n
i
f
ican
ce
in
d
ec
is
io
n
-
m
ak
in
g
.
2.
L
I
T
E
R
AT
U
RE
R
E
VI
E
W
A
ND
RE
S
E
ARCH
T
ARG
E
T
S
So
f
twar
e
f
au
lt
p
r
e
d
ictio
n
is
a
cr
itical
asp
ec
t
o
f
s
o
f
twar
e
e
n
g
in
ee
r
in
g
,
aim
i
n
g
to
i
d
en
tif
y
p
o
ten
tial
d
ef
ec
ts
in
s
o
f
twar
e
m
o
d
u
les
b
ef
o
r
e
th
ey
lead
to
o
p
e
r
atio
n
al
is
s
u
es.
Ad
d
r
ess
in
g
th
e
c
h
allen
g
es
p
o
s
ed
b
y
im
b
alan
ce
d
d
atasets
is
cr
u
cial
f
o
r
im
p
r
o
v
in
g
th
e
ac
c
u
r
ac
y
an
d
ef
f
ec
tiv
e
n
ess
o
f
f
au
lt
p
r
ed
ict
io
n
m
o
d
els.
I
n
th
is
liter
atu
r
e
s
u
r
v
ey
,
we
r
ev
iew
p
r
ev
io
u
s
r
esear
ch
in
s
o
f
twar
e
f
au
lt
p
r
ed
ictio
n
,
with
a
f
o
cu
s
o
n
d
ata
b
alan
cin
g
tech
n
iq
u
es
s
u
ch
as
GANs,
SMOT
E
,
an
d
Nea
r
Miss
.
W
e
al
s
o
d
is
cu
s
s
th
e
lim
itatio
n
s
o
f
e
ar
lier
ap
p
r
o
ac
h
es
an
d
h
o
w
th
e
p
r
o
p
o
s
ed
r
esear
ch
ai
m
s
to
ad
d
r
ess
th
em
.
I
n
th
e
r
e
alm
o
f
s
o
f
twar
e
f
au
lt
p
r
ed
icti
o
n
,
a
d
d
r
ess
in
g
th
e
ch
allen
g
e
o
f
im
b
alan
ce
d
d
atasets
is
p
ar
am
o
u
n
t
f
o
r
ac
h
iev
in
g
ac
cu
r
ate
an
d
r
eliab
le
p
r
ed
icti
v
e
m
o
d
els.
Var
io
u
s
d
ata
b
alan
cin
g
tech
n
iq
u
es
h
av
e
b
ee
n
p
r
o
p
o
s
ed
an
d
e
x
p
lo
r
ed
in
th
e
liter
atu
r
e
to
m
itig
ate
t
h
e
im
p
ac
t
o
f
class
im
b
alan
ce
o
n
m
o
d
el
p
er
f
o
r
m
an
ce
.
I
n
t
h
is
s
ec
tio
n
,
we
r
ev
i
ew
p
r
ev
io
u
s
r
esear
c
h
f
o
c
u
s
in
g
o
n
d
ata
b
alan
cin
g
tech
n
iq
u
es
s
u
ch
as
GANs,
SMOT
E
,
an
d
Nea
r
Miss
,
h
ig
h
lig
h
t
in
g
th
eir
s
tr
en
g
th
s
,
lim
itatio
n
s
,
an
d
g
a
p
s
,
as
well
as
d
is
cu
s
s
in
g
h
o
w
th
e
p
r
o
p
o
s
ed
r
esear
ch
aim
s
to
ad
d
r
ess
t
h
ese
s
h
o
r
tco
m
in
g
s
.
E
n
g
elm
an
an
d
L
ess
m
an
n
[
1
8
]
p
r
esen
t
c
o
n
d
itio
n
al
W
ass
er
s
t
ein
GAN
-
b
ased
o
v
er
s
am
p
lin
g
o
f
tab
u
lar
d
ata
f
o
r
im
b
al
an
ce
d
lear
n
in
g
f
o
r
o
v
er
s
am
p
lin
g
im
b
ala
n
ce
d
d
at
a
in
cr
ed
it
s
co
r
in
g
.
Alq
ar
n
i
an
d
Aljam
aa
n
[
1
9
]
p
r
o
p
o
s
e
a
n
o
v
el
ap
p
r
o
ac
h
th
at
co
m
b
in
es
GAN
-
b
ased
m
eth
o
d
s
with
b
o
o
s
tin
g
en
s
em
b
les
to
im
p
r
o
v
e
s
o
f
twar
e
d
ef
ec
t
p
r
ed
i
ctio
n
p
er
f
o
r
m
an
ce
.
W
h
ile
th
e
s
tu
d
y
o
f
f
e
r
s
a
p
r
o
m
is
in
g
s
o
lu
tio
n
,
it
lack
s
s
p
ec
if
i
c
v
alid
atio
n
o
n
th
e
q
u
ality
o
f
g
en
er
ated
s
y
n
th
etic
d
ata
an
d
th
e
s
ca
lab
ilit
y
o
f
th
e
p
r
o
p
o
s
ed
a
p
p
r
o
ac
h
.
Ou
r
r
es
ea
r
ch
aim
s
to
ad
d
r
ess
th
ese
g
ap
s
b
y
co
n
d
u
ctin
g
r
ig
o
r
o
u
s
p
er
f
o
r
m
a
n
ce
e
v
alu
a
tio
n
an
d
s
ca
lab
ilit
y
test
in
g
o
f
GAN
-
b
ased
o
v
e
r
s
am
p
lin
g
tech
n
iq
u
es,
th
u
s
p
r
o
v
id
i
n
g
em
p
ir
ical
ev
id
en
ce
o
f
th
eir
ef
f
ec
tiv
en
ess
an
d
p
r
ac
tical
f
ea
s
ib
ilit
y
.
Sev
er
al
s
tu
d
ies
h
av
e
in
v
esti
g
ated
th
e
u
s
e
o
f
d
if
f
er
en
t
d
ata
b
alan
cin
g
tech
n
iq
u
es
to
im
p
r
o
v
e
s
o
f
twar
e
f
au
lt
p
r
ed
ictio
n
m
o
d
els.
Fo
r
e
x
am
p
le,
Ku
m
a
r
an
d
Ven
k
ates
an
[
2
0
]
ex
p
lo
r
ed
th
e
u
s
e
o
f
G
ANs
f
o
r
ad
d
r
ess
in
g
d
ata
im
b
alan
ce
in
s
o
f
twar
e
d
e
f
ec
t
p
r
ed
ictio
n
.
T
h
e
au
th
o
r
s
p
r
o
p
o
s
ed
a
n
o
v
el
ap
p
r
o
ac
h
t
h
at
lev
er
ag
es
GANs
to
g
en
er
ate
s
y
n
t
h
etic
d
ata
s
am
p
l
es,
th
er
eb
y
b
alan
cin
g
th
e
d
is
tr
ib
u
tio
n
o
f
d
ef
ec
tiv
e
a
n
d
n
o
n
-
d
ef
ec
tiv
e
in
s
tan
ce
s
in
th
e
d
ataset.
T
h
eir
r
esu
lts
s
h
o
wed
p
r
o
m
is
in
g
im
p
r
o
v
em
e
n
ts
in
m
o
d
el
ac
cu
r
ac
y
an
d
p
er
f
o
r
m
an
ce
c
o
m
p
ar
e
d
to
tr
ad
itio
n
al
tech
n
iq
u
es.
Similar
ly
,
Fen
g
et
a
l.
[
2
1
]
f
o
c
u
s
es
o
n
u
s
in
g
SMOT
E
,
a
p
o
p
u
lar
o
v
e
r
s
am
p
lin
g
tech
n
iq
u
e,
t
o
ad
d
r
ess
class
i
m
b
alan
ce
in
s
o
f
twar
e
f
au
lt
p
r
ed
ictio
n
.
T
h
e
s
tu
d
y
co
m
p
ar
ed
th
e
p
er
f
o
r
m
an
ce
o
f
ML
m
o
d
els
tr
ain
ed
o
n
b
alan
ce
d
an
d
u
n
b
alan
ce
d
d
atasets
,
d
em
o
n
s
tr
atin
g
th
e
ef
f
ec
tiv
e
n
ess
o
f
SMOT
E
in
im
p
r
o
v
in
g
m
o
d
el
p
er
f
o
r
m
a
n
ce
.
Ho
wev
er
,
th
e
s
tu
d
y
also
h
ig
h
lig
h
ted
lim
itatio
n
s
in
th
e
s
ca
lab
ilit
y
an
d
co
m
p
u
tatio
n
al
ef
f
icien
c
y
o
f
SMOT
E
,
in
d
icatin
g
t
h
e
n
ee
d
f
o
r
f
u
r
th
er
r
esear
ch
in
th
is
ar
ea
.
I
n
ad
d
itio
n
to
o
v
er
s
am
p
lin
g
tech
n
iq
u
es
lik
e
SMOT
E
an
d
GANs,
u
n
d
e
r
s
a
m
p
lin
g
m
et
h
o
d
s
s
u
ch
as
Nea
r
Miss
h
av
e
also
b
ee
n
ex
p
lo
r
ed
i
n
th
e
co
n
te
x
t
o
f
s
o
f
twar
e
f
au
lt
p
r
ed
ictio
n
.
Fo
r
ex
am
p
le,
Mq
ad
i
et
a
l.
[
1
4
]
in
v
e
s
tig
ates
th
e
u
s
e
o
f
Nea
r
Miss
f
o
r
h
a
n
d
lin
g
class
im
b
alan
ce
i
n
d
e
f
ec
t
p
r
ed
ictio
n
d
atasets
.
T
h
e
au
t
h
o
r
s
p
r
o
p
o
s
ed
a
h
y
b
r
id
ap
p
r
o
ac
h
th
at
co
m
b
in
es
u
n
d
er
s
am
p
lin
g
with
f
ea
tu
r
e
s
elec
tio
n
to
i
m
p
r
o
v
e
th
e
p
er
f
o
r
m
a
n
ce
o
f
ML
m
o
d
els.
T
h
eir
ex
p
er
im
en
tal
r
esu
lts
s
h
o
wed
p
r
o
m
is
in
g
im
p
r
o
v
e
m
en
ts
in
m
o
d
el
ac
cu
r
a
cy
an
d
g
e
n
er
ali
za
tio
n
ca
p
ab
ilit
ies.
Fu
r
th
er
m
o
r
e
,
in
[
2
2
]
,
a
s
u
r
v
e
y
o
f
s
o
f
twar
e
f
a
u
lt
p
r
ed
ictio
n
tech
n
iq
u
es
an
d
r
ec
en
t
d
ev
el
o
p
m
en
ts
is
p
r
o
v
id
ed
,
wh
ich
h
ig
h
lig
h
ts
th
e
n
ee
d
to
ad
d
r
ess
class
im
b
alan
ce
is
s
u
e
s
.
W
h
ile
th
e
s
u
r
v
ey
o
f
f
e
r
s
v
alu
ab
le
in
s
ig
h
ts
in
to
v
ar
io
u
s
s
u
p
er
v
is
ed
ML
tech
n
iq
u
es
an
d
s
am
p
lin
g
m
eth
o
d
s
,
it
lack
s
in
-
d
ep
th
ex
p
lo
r
atio
n
o
f
em
er
g
in
g
tech
n
iq
u
es
b
e
y
o
n
d
SMOT
E
.
Ou
r
r
esear
ch
aim
s
to
ex
te
n
d
th
is
s
u
r
v
ey
b
y
in
v
esti
g
atin
g
t
h
e
ef
f
ec
tiv
e
n
ess
o
f
GANs
an
d
o
th
e
r
a
d
v
an
ce
d
d
at
a
b
alan
cin
g
tec
h
n
iq
u
es
in
s
o
f
twar
e
f
au
lt
p
r
e
d
ictio
n
,
th
u
s
e
n
r
ich
in
g
t
h
e
ex
is
tin
g
liter
atu
r
e
with
n
ew
in
s
ig
h
ts
an
d
em
p
ir
ical
ev
id
en
ce
.
Ou
r
p
r
im
ar
y
r
esear
ch
tar
g
et
is
to
co
n
d
u
ct
ex
ten
s
iv
e
e
x
p
e
r
im
en
ts
to
e
v
alu
ate
t
h
e
p
e
r
f
o
r
m
an
ce
o
f
s
o
f
twar
e
d
ef
ec
t
p
r
e
d
ictio
n
i
n
s
ce
n
ar
io
s
wh
er
e
d
ata
is
im
b
ala
n
ce
d
.
Desp
ite
t
h
e
a
d
v
an
ce
m
e
n
ts
in
d
ata
b
alan
cin
g
tech
n
iq
u
es
f
o
r
s
o
f
twar
e
f
au
l
t
p
r
ed
ictio
n
,
s
ev
e
r
al
lim
itatio
n
s
ex
is
t
in
ea
r
lier
ap
p
r
o
ac
h
es.
On
e
co
m
m
o
n
lim
itatio
n
is
th
e
lack
o
f
co
m
p
r
eh
en
s
iv
e
ev
alu
atio
n
an
d
co
m
p
ar
is
o
n
o
f
d
if
f
e
r
en
t
s
am
p
lin
g
m
eth
o
d
s
ac
r
o
s
s
d
iv
er
s
e
d
atasets
an
d
ML
alg
o
r
ith
m
s
.
Ma
n
y
s
tu
d
ies
f
o
cu
s
o
n
a
lim
ited
s
et
o
f
tech
n
iq
u
es
o
r
d
atasets
,
wh
ich
m
ay
n
o
t
f
u
lly
ca
p
tu
r
e
th
e
v
ar
iab
ilit
y
an
d
c
o
m
p
lex
ity
o
f
r
ea
l
-
wo
r
ld
s
o
f
twar
e
d
ev
e
lo
p
m
en
t
s
ce
n
a
r
io
s
.
Fu
r
th
er
m
o
r
e
,
ex
is
tin
g
r
esear
c
h
o
f
ten
o
v
er
lo
o
k
s
th
e
im
p
ac
t
o
f
d
ata
b
ias
an
d
m
o
d
el
in
te
r
p
r
etab
ilit
y
o
n
th
e
ef
f
ec
tiv
en
ess
o
f
d
ata
b
ala
n
cin
g
tech
n
iq
u
es.
I
m
b
alan
ce
d
d
ata
s
ets
m
ay
co
n
tain
b
iased
r
ep
r
esen
tatio
n
s
o
f
ce
r
tain
class
es,
lead
in
g
to
s
k
ewe
d
m
o
d
el
p
r
e
d
ictio
n
s
an
d
er
r
o
n
eo
u
s
co
n
clu
s
io
n
s
.
Mo
r
eo
v
e
r
,
th
e
in
ter
p
r
etab
ilit
y
o
f
ML
m
o
d
els tr
ain
ed
o
n
b
alan
ce
d
o
r
s
y
n
th
etic
d
ata
is
o
f
ten
o
v
er
lo
o
k
ed
,
m
ak
in
g
it c
h
allen
g
in
g
to
u
n
d
e
r
s
tan
d
th
e
u
n
d
er
ly
i
n
g
f
ac
to
r
s
d
r
i
v
in
g
m
o
d
el
p
r
ed
ictio
n
s
.
Her
e
,
we
s
tu
d
y
th
e
im
p
ac
t
o
f
d
ata
im
b
al
an
ce
o
n
8
d
if
f
er
en
t
d
atasets
:
J
M1
,
A
R
1
,
C
M1
,
K
C
2
,
MW1
,
PC
1
,
MC2
,
an
d
KC
1
.
Ou
r
r
esear
ch
aim
s
to
b
u
il
d
u
p
o
n
th
e
p
r
o
b
lem
s
id
en
tifie
d
in
liter
atu
r
e
r
ev
ie
w
d
u
e
to
d
ata
im
b
alan
ce
an
d
b
y
ev
alu
atin
g
th
e
ap
p
licab
ilit
y
o
f
GAN
-
b
ased
o
v
er
s
am
p
lin
g
ac
r
o
s
s
m
u
ltip
le
s
o
f
twar
e
f
au
lt
p
r
ed
ictio
n
d
atasets
.
W
e
th
u
s
ex
ten
d
th
e
s
co
p
e
b
ey
o
n
d
s
p
ec
if
ic
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
E
n
h
a
n
ci
n
g
s
o
ftw
a
r
e
fa
u
lt p
r
ed
ictio
n
th
r
o
u
g
h
d
a
ta
b
a
la
n
cin
g
tech
n
iq
u
es a
n
d
ma
ch
i
n
e
lea
r
n
in
g
(
A
ksh
a
t R
a
j)
4789
d
o
m
ain
s
an
d
ad
d
r
ess
th
e
lim
itatio
n
s
o
f
s
in
g
le
-
class
f
o
c
u
s
.
Ou
r
r
esear
ch
s
ee
k
s
to
b
r
id
g
e
th
is
g
ap
b
y
in
co
r
p
o
r
atin
g
GAN
-
b
ased
o
v
e
r
s
am
p
lin
g
in
to
th
e
c
o
m
p
ar
ativ
e
an
aly
s
is
,
th
u
s
p
r
o
v
id
in
g
a
m
o
r
e
co
m
p
r
eh
en
s
iv
e
ev
alu
atio
n
o
f
s
am
p
lin
g
tec
h
n
i
q
u
es'
ef
f
icac
y
in
s
o
f
twar
e
f
au
l
t p
r
ed
ictio
n
.
3.
P
RO
P
O
SE
D
M
E
T
H
O
DO
L
O
G
Y
T
o
o
v
er
co
m
e
th
e
lim
itatio
n
s
o
f
ex
is
tin
g
r
esear
ch
,
we
ad
o
p
ted
a
s
y
s
tem
atic
ap
p
r
o
ac
h
th
at
in
clu
d
es
th
o
r
o
u
g
h
ex
p
e
r
im
en
tatio
n
,
r
i
g
o
r
o
u
s
ev
alu
atio
n
m
etr
ics,
a
n
d
e
x
ten
s
iv
e
v
alid
atio
n
ac
r
o
s
s
d
iv
er
s
e
s
o
f
twar
e
d
ev
elo
p
m
e
n
t
s
ce
n
ar
i
o
s
ex
p
l
o
r
i
n
g
GAN,
SMOT
E
,
an
d
N
ea
r
Miss
ac
r
o
s
s
m
u
ltip
le
d
atasets
an
d
ML
alg
o
r
ith
m
s
.
Ad
d
itio
n
ally
,
we
h
av
e
e
x
p
lo
r
ed
th
e
u
s
e
o
f
ad
v
an
ce
d
s
am
p
lin
g
tech
n
iq
u
es
an
d
en
s
em
b
le
m
eth
o
d
s
to
f
u
r
th
er
en
h
an
ce
th
e
m
o
d
el
p
er
f
o
r
m
an
ce
an
d
r
eliab
ilit
y
.
He
n
ce
,
th
e
p
r
o
p
o
s
ed
r
esear
ch
s
ee
k
s
to
a
d
v
an
ce
th
e
f
ield
o
f
s
o
f
twar
e
f
au
lt
p
r
ed
ictio
n
b
y
p
r
o
v
id
in
g
in
s
ig
h
ts
i
n
to
th
e
m
o
s
t
ef
f
ec
tiv
e
d
ata
b
ala
n
cin
g
tech
n
iq
u
es
f
o
r
im
p
r
o
v
in
g
m
o
d
el
ac
cu
r
ac
y
a
n
d
r
eliab
ilit
y
.
B
y
a
d
d
r
ess
in
g
th
e
lim
itatio
n
s
o
f
ea
r
lier
ap
p
r
o
ac
h
es,
we
aim
to
co
n
tr
ib
u
te
to
th
e
d
ev
el
o
p
m
en
t
o
f
m
o
r
e
r
o
b
u
s
t
an
d
d
ep
e
n
d
ab
le
f
au
lt
p
r
ed
ictio
n
m
o
d
els,
u
ltima
tely
en
h
an
ci
n
g
th
e
q
u
ality
an
d
r
eliab
ilit
y
o
f
s
o
f
twar
e
s
y
s
tem
s
.
Fig
u
r
e
1
d
e
p
icts
th
e
lo
g
ical
s
tep
s
tak
en
in
t
h
e
p
r
o
p
o
s
ed
m
o
d
el
.
Fig
u
r
e
1
.
L
o
g
ical
s
tep
s
o
f
th
e
p
r
o
p
o
s
ed
s
o
f
twar
e
f
au
lt
p
r
ed
ic
tio
n
m
o
d
el
3
.
1
.
Da
t
a
p
re
pro
ce
s
s
ing
W
e
b
eg
in
b
y
co
llectin
g
s
o
f
twar
e
f
au
lt
p
r
e
d
ictio
n
d
atasets
f
r
o
m
r
ep
u
tab
le
s
o
u
r
ce
s
s
u
ch
a
s
NASA
's
s
o
f
twar
e
en
g
in
ee
r
in
g
lab
o
r
at
o
r
y
an
d
th
e
PR
OM
I
SE
s
o
f
twar
e
en
g
in
ee
r
i
n
g
r
ep
o
s
ito
r
y
.
Da
ta
p
r
ep
r
o
ce
s
s
in
g
is
co
n
d
u
cte
d
to
en
s
u
r
e
d
ata
q
u
ality
an
d
co
n
s
is
ten
cy
.
T
h
is
in
v
o
lv
es
h
a
n
d
lin
g
m
is
s
in
g
v
a
lu
es,
o
u
tlier
s
,
an
d
in
co
n
s
is
ten
cies
in
th
e
d
ataset
s
.
W
e
em
p
lo
y
tec
h
n
iq
u
es
s
u
ch
as
m
ea
n
im
p
u
tatio
n
o
r
d
eletio
n
f
o
r
m
is
s
in
g
v
alu
es,
o
u
tlier
d
etec
tio
n
an
d
r
em
o
v
al
u
s
in
g
s
tatis
tical
m
eth
o
d
s
o
r
d
o
m
ain
k
n
o
wled
g
e,
an
d
s
tan
d
ar
d
izatio
n
o
r
n
o
r
m
aliza
tio
n
to
s
ca
le
th
e
f
ea
tu
r
es
ap
p
r
o
p
r
iately
.
Ad
d
itio
n
a
lly
,
we
p
er
f
o
r
m
ex
p
l
o
r
ato
r
y
d
ata
an
aly
s
is
(
E
DA)
[
1
8
]
to
g
ain
in
s
ig
h
ts
in
to
th
e
d
ata
d
is
tr
ib
u
tio
n
an
d
ch
ar
ac
ter
is
tics
,
id
en
tify
in
g
p
o
ten
tial
p
a
tter
n
s
o
r
tr
en
d
s
th
at
m
ay
aid
in
m
o
d
el
d
ev
elo
p
m
en
t
an
d
in
ter
p
r
etatio
n
.
E
DA
also
h
elp
s
u
s
u
n
d
er
s
tan
d
th
e
ex
ten
t
o
f
class
im
b
alan
ce
in
th
e
d
atasets
,
wh
ich
is
cr
u
cial
f
o
r
s
elec
tin
g
ap
p
r
o
p
r
iate
d
ata
b
ala
n
cin
g
tec
h
n
iq
u
es.
3.
2
.
Da
t
a
g
ener
a
t
io
n us
ing
g
ener
a
t
iv
e
a
dv
er
s
a
ria
l net
wo
r
k
Dee
p
lear
n
in
g
m
o
d
els
s
u
ch
as
GAN
co
m
p
r
is
e
two
n
eu
r
al
n
etwo
r
k
s
,
th
e
g
en
e
r
ato
r
an
d
th
e
d
is
cr
im
in
ato
r
,
th
at
ar
e
e
n
g
ag
e
d
in
a
m
in
im
ax
g
am
e.
T
h
e
g
e
n
er
ato
r
attem
p
ts
to
g
e
n
er
ate
s
y
n
th
etic
d
ata
th
at
is
alm
o
s
t
lik
e
th
e
r
ea
l
d
ata,
wh
il
e
th
e
d
is
cr
im
in
ato
r
aim
s
to
id
en
tify
th
e
r
ea
l
an
d
g
e
n
er
ated
d
ata.
T
h
e
g
e
n
er
ato
r
aim
s
to
g
e
n
er
ate
in
c
r
ea
s
in
g
ly
r
ea
lis
tic
s
am
p
les
b
y
m
in
im
i
zin
g
its
o
wn
lo
s
s
(
G
l
o
s
s
)
,
wh
ich
m
ea
s
u
r
es
t
h
e
d
is
cr
ep
an
cy
b
etwe
en
t
h
e
d
is
cr
im
in
ato
r
'
s
p
r
ed
ictio
n
s
a
n
d
a
lab
el
in
d
icatin
g
th
e
g
en
er
ated
d
ata
is
r
ea
l.
C
o
n
v
er
s
ely
,
th
e
d
is
cr
im
in
ato
r
s
ee
k
s
to
c
o
r
r
ec
tly
class
if
y
r
ea
l
an
d
f
ak
e
s
am
p
les,
th
u
s
m
in
im
izin
g
its
lo
s
s
(
D
lo
s
s
)
.
G
lo
s
s
an
d
D
lo
s
s
ar
e
f
u
n
d
am
e
n
tal
co
m
p
o
n
en
ts
in
tr
ain
in
g
GANs,
r
ep
r
esen
tin
g
th
e
o
b
jectiv
es
o
f
th
e
g
en
er
ato
r
a
n
d
d
is
cr
im
in
ato
r
,
r
esp
ec
tiv
ely
,
in
ac
h
iev
in
g
th
eir
co
m
p
etin
g
g
o
als.
T
h
e
g
en
er
a
to
r
lo
s
s
is
ty
p
ically
d
ef
in
ed
u
s
in
g
b
in
a
r
y
cr
o
s
s
-
e
n
tr
o
p
y
,
m
ea
s
u
r
in
g
th
e
d
if
f
e
r
en
ce
b
etwe
en
th
e
d
is
cr
im
in
ato
r
'
s
p
r
ed
ictio
n
s
o
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
14
,
No
.
6
,
Dec
em
b
er
2
0
2
5
:
4
7
8
7
-
4
8
0
1
4790
g
en
er
ated
d
ata
an
d
a
m
atr
ix
o
f
o
n
es
(
s
in
ce
th
e
g
en
er
at
o
r
wan
ts
th
e
d
is
cr
im
in
ato
r
to
cl
ass
if
y
its
o
u
tp
u
ts
as
r
ea
l)
.
Ma
th
em
atica
lly
,
it c
an
b
e
ex
p
r
ess
ed
as
(
1
):
=
−
1
∑
=
1
(
(
(
)
)
)
(
1
)
Her
e,
G
(
)
is
th
e
o
u
tp
u
t
o
f
t
h
e
g
en
e
r
ato
r
f
o
r
th
e
i
-
th
i
n
p
u
t
n
o
is
e
s
am
p
le
an
d
D(
G
(
)
)
is
th
e
d
is
cr
im
in
ato
r
’
s
o
u
t
p
u
t
wh
en
t
h
e
p
r
o
b
ab
ilit
y
o
f
th
e
g
en
er
ate
d
s
am
p
le
G
(
)
is
r
ea
l
,
an
d
N
is
th
e
b
atch
s
ize
co
n
s
id
er
ed
.
T
h
e
d
is
cr
im
in
ato
r
lo
s
s
is
co
m
p
u
ted
u
s
in
g
b
in
ar
y
cr
o
s
s
-
en
tr
o
p
y
lo
s
s
s
ep
ar
ately
f
o
r
r
ea
l
a
n
d
f
a
k
e
d
ata,
an
d
t
h
en
av
e
r
ag
ed
.
Ma
th
em
atica
lly
,
it c
an
b
e
ex
p
r
ess
ed
as
(
2
)
:
=
−
1
2
∑
=
1
(
(
)
)
+
(
1
−
(
(
(
)
)
)
)
(
2
)
Her
e,
is
th
e
ith
r
ea
l
d
ata
p
o
in
t
an
d
(
)
th
e
d
is
cr
im
in
ato
r
is
o
u
tp
u
t
wh
en
is
r
ea
l.
G
(
)
is
th
e
o
u
tp
u
t
o
f
th
e
g
en
e
r
ato
r
f
o
r
th
e
i
-
th
in
p
u
t
n
o
is
e
s
am
p
le
an
d
D(
G
(
)
)
.
T
h
e
d
is
cr
im
in
ato
r
’
s
o
u
tp
u
t
wh
en
th
e
p
r
o
b
a
b
ilit
y
o
f
t
h
e
g
en
e
r
ated
s
am
p
le
G
(
)
is
r
ea
l
,
an
d
N
is
th
e
b
atch
s
ize
co
n
s
id
er
e
d
.
B
in
ar
y
cr
o
s
s
-
en
tr
o
p
y
lo
s
s
m
ea
s
u
r
es
th
e
d
if
f
er
en
ce
b
etwe
en
p
r
o
b
a
b
ilit
y
d
is
tr
ib
u
tio
n
s
,
p
ar
ticu
lar
ly
u
s
ef
u
l
f
o
r
b
i
n
ar
y
class
if
icatio
n
p
r
o
b
lem
s
.
I
t is d
e
f
in
ed
as
(
3
)
.
(
)
=
−
1
∑
=
1
(
1
−
(
)
)
)
(
3
)
W
h
er
e
i
s
th
e
tr
u
e
lab
el
(
0
o
r
1
)
a
n
d
(
)
is
th
e
p
r
ed
ict
ed
p
r
o
b
ab
ilit
y
o
f
th
e
p
o
s
itiv
e
class
.
T
h
e
GA
N
o
b
jectiv
e
f
u
n
ctio
n
is
ess
en
tially
a
m
in
im
a
x
g
am
e
b
etwe
en
t
h
e
g
en
er
ato
r
an
d
th
e
d
is
cr
im
in
a
to
r
.
T
h
e
g
en
e
r
ato
r
tr
ies
to
m
in
im
ize
th
e
d
is
cr
im
in
ato
r
'
s
ab
ilit
y
to
d
is
tin
g
u
is
h
b
etwe
en
r
ea
l
an
d
f
a
k
e
d
ata,
wh
ile
th
e
d
is
cr
im
in
ato
r
tr
ies to
m
ax
im
ize
it.
Ma
th
em
a
tically
,
it is
g
iv
en
as
(
4
)
.
(
,
)
=
~
(
)
[
l
og
(
)
]
+
~
(
)
[
l
og
(
1
−
(
(
)
)
)
]
(
4
)
Her
e
,
D(
x
)
is
th
e
d
is
cr
im
in
ato
r
’
s
o
u
tp
u
t
an
d
G
(
)
is
th
e
g
en
er
ato
r
'
s
o
u
tp
u
t
(
f
ak
e
d
ata)
an
d
(
)
(
)
ar
e
th
e
d
ata
a
n
d
n
o
is
e
d
is
tr
ib
u
tio
n
s
.
3.
3
.
Da
t
a
ba
la
ncing
us
ing
s
y
nthet
ic
m
ino
rit
y
o
v
e
r
-
s
a
m
pli
ng
t
ec
hn
iqu
e
SMOT
E
is
a
wid
ely
-
u
s
ed
d
ata
au
g
m
en
tatio
n
tech
n
iq
u
e
th
at
p
r
o
d
u
ce
s
u
n
r
ea
l sam
p
les f
o
r
t
h
e
m
in
o
r
ity
class
b
y
s
y
n
th
esizin
g
b
etwe
e
n
ex
is
tin
g
m
in
o
r
ity
class
s
am
p
les.
Sp
ec
if
ically
,
SMOT
E
c
o
n
s
id
er
s
a
m
in
o
r
ity
class
s
am
p
le
an
d
its
k
n
ea
r
est
n
eig
h
b
o
r
s
,
th
en
c
r
ea
tes
n
ew
s
y
n
th
etic
s
am
p
les
alo
n
g
th
e
lin
e
s
eg
m
en
ts
jo
in
in
g
th
em
.
B
y
ef
f
ec
tiv
ely
in
cr
ea
s
in
g
th
e
r
ep
r
esen
tatio
n
o
f
th
e
m
in
o
r
ity
class
,
SMOT
E
h
elp
s
to
b
alan
ce
th
e
class
d
is
tr
ib
u
tio
n
in
th
e
d
atasets
,
th
u
s
m
itig
atin
g
th
e
b
ias
to
wa
r
d
s
th
e
m
ajo
r
ity
class
.
T
h
is
tech
n
iq
u
e
aim
s
to
im
p
r
o
v
e
t
h
e
p
er
f
o
r
m
a
n
ce
o
f
ML
m
o
d
els
b
y
p
r
o
v
id
in
g
th
e
m
with
m
o
r
e
d
iv
er
s
e
an
d
r
ep
r
esen
tativ
e
tr
ain
in
g
d
ata,
th
er
eb
y
r
ed
u
cin
g
t
h
e
r
is
k
o
f
m
o
d
el
o
v
e
r
f
itti
n
g
an
d
im
p
r
o
v
in
g
p
r
ed
ictiv
e
ac
c
u
r
ac
y
f
o
r
t
h
e
m
in
o
r
ity
class
.
3.
4
.
Da
t
a
ba
la
ncing
us
ing
Nea
rMi
s
s
An
o
th
er
d
ata
b
alan
cin
g
tech
n
i
q
u
e
we
u
tili
ze
is
Nea
r
Mis
s
,
w
h
ich
is
s
p
ec
if
ically
d
esig
n
ed
t
o
ad
d
r
ess
class
im
b
alan
ce
b
y
u
n
d
er
s
am
p
lin
g
th
e
m
ajo
r
ity
class
.
Nea
r
Miss
s
e
lects
a
s
u
b
s
et
o
f
m
a
jo
r
ity
class
s
am
p
les
th
at
ar
e
clo
s
est
to
th
e
m
in
o
r
ity
class
s
am
p
les
in
f
ea
tu
r
e
s
p
ac
e,
ef
f
ec
tiv
ely
r
ed
u
cin
g
t
h
e
im
b
alan
ce
r
atio
b
etwe
en
th
e
m
ajo
r
ity
a
n
d
m
i
n
o
r
ity
class
es.
Nea
r
Miss
o
f
f
er
s
th
r
ee
v
ar
ian
ts
:
Nea
r
Miss
-
1
,
Nea
r
Miss
-
2
,
an
d
Nea
r
Miss
-
3
,
ea
ch
em
p
lo
y
in
g
d
if
f
er
en
t
s
tr
ateg
ies
to
s
elec
t
th
e
m
ajo
r
ity
class
s
am
p
les
f
o
r
r
e
m
o
v
al.
Nea
r
Miss
-
1
co
n
s
id
er
s
s
am
p
les
f
r
o
m
th
e
m
ajo
r
ity
class
with
th
e
s
m
all
est
av
er
ag
e
d
is
tan
ce
to
th
e
th
r
ee
n
ea
r
est
m
in
o
r
ity
class
s
am
p
le
s
.
I
n
Nea
r
Miss
-
2
,
s
am
p
les
f
r
o
m
th
e
m
ajo
r
ity
cl
ass
with
th
e
f
ar
th
est
av
er
ag
e
d
is
tan
ce
to
th
e
th
r
ee
n
ea
r
est
m
in
o
r
ity
class
s
am
p
les
ar
e
co
n
s
id
er
e
d
.
Nea
r
Miss
-
3
o
n
th
e
o
th
e
r
h
a
n
d
is
a
two
-
s
te
p
p
r
o
ce
s
s
th
at
f
ir
s
t
s
elec
ts
m
ajo
r
ity
class
s
am
p
les
u
s
in
g
Nea
r
Miss
-
1
o
r
Nea
r
Miss
-
2
an
d
t
h
en
f
u
r
th
e
r
r
e
f
in
es
t
h
e
s
elec
tio
n
b
ased
o
n
th
e
m
ajo
r
ity
class
s
am
p
les
th
at
ar
e
m
is
class
if
ied
b
y
a
k
-
n
ea
r
est
n
eig
h
b
o
r
(
KNN)
class
if
ier
tr
ain
ed
o
n
t
h
e
o
r
ig
in
al
d
ataset.
B
y
s
tr
ateg
ically
r
em
o
v
in
g
m
ajo
r
ity
class
s
am
p
les,
Nea
r
Miss
aim
s
to
e
n
h
an
ce
th
e
b
ala
n
ce
b
etwe
en
th
e
class
es a
n
d
im
p
r
o
v
e
th
e
p
er
f
o
r
m
a
n
ce
o
f
ML
m
o
d
els o
n
im
b
alan
ce
d
d
atasets
.
3.
5
.
M
a
chine le
a
rning
a
lg
o
ri
t
hm
s
T
h
e
im
p
lem
e
n
tatio
n
will
u
ti
lize
a
v
a
r
iety
o
f
ML
alg
o
r
i
th
m
s
s
u
itab
le
f
o
r
class
if
icatio
n
task
s
,
in
clu
d
in
g
b
u
t
n
o
t
lim
ited
to
:
r
an
d
o
m
f
o
r
ests
(
R
F)
[
2
3
]
,
KNN
[
2
4
]
,
lo
g
is
tic
r
eg
r
ess
io
n
(
L
R
)
[
2
5
]
,
[
2
6
]
,
a
n
d
g
r
ad
ien
t
b
o
o
s
tin
g
(
GB
)
[
2
7
]
a
lg
o
r
ith
m
s
.
I
n
o
u
r
a
n
aly
s
is
o
f
v
ar
io
u
s
d
atasets
ac
r
o
s
s
d
if
f
e
r
en
t
d
ata
b
alan
cin
g
tech
n
iq
u
es,
we
ev
alu
ate
d
th
e
p
er
f
o
r
m
an
ce
o
f
f
o
u
r
co
m
m
o
n
l
y
u
s
ed
m
ac
h
i
n
e
-
lear
n
in
g
m
o
d
els:
RF
,
LR
,
KNN,
an
d
GB
.
T
h
ese
m
o
d
els
wer
e
ch
o
s
en
f
o
r
th
eir
v
er
s
atility
an
d
wid
esp
r
ea
d
ap
p
licab
ilit
y
in
class
if
icatio
n
task
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
E
n
h
a
n
ci
n
g
s
o
ftw
a
r
e
fa
u
lt p
r
ed
ictio
n
th
r
o
u
g
h
d
a
ta
b
a
la
n
cin
g
tech
n
iq
u
es a
n
d
ma
ch
i
n
e
lea
r
n
in
g
(
A
ksh
a
t R
a
j)
4791
T
h
e
f
ir
s
t
m
o
d
el
em
p
lo
y
e
d
was
K
NN
,
wh
ich
is
a
n
o
n
-
p
ar
am
e
tr
ic
class
if
icatio
n
alg
o
r
ith
m
.
Giv
en
a
q
u
er
y
p
o
in
t
,
th
e
alg
o
r
ith
m
f
in
d
s
th
e
n
ea
r
est
n
eig
h
b
o
r
s
in
th
e
tr
ai
n
in
g
s
et
an
d
ass
ig
n
s
th
e
m
ajo
r
ity
class
am
o
n
g
th
ese
n
eig
h
b
o
r
s
to
.
C
o
n
s
id
er
(
)
to
b
e
t
h
e
s
et
o
f
k
n
ea
r
est
n
ei
g
h
b
o
r
s
o
f
in
th
e
tr
ain
in
g
s
et,
th
en
t
h
e
p
r
ed
icted
class
lab
el
is
f
o
r
is
g
iv
en
as p
er
(
5
):
̂
=
∑
є
(
)
(
=
)
(
5
)
Her
e
I
(
.
)
is
th
e
in
d
icato
r
f
u
n
ct
io
n
an
d
is
th
e
clas
s
lab
el
o
f
t
h
e
i
-
th
n
ea
r
est
n
eig
h
b
o
r
an
d
̂
is
th
e
p
r
ed
icted
class
lab
el
f
o
r
.
LR
is
a
lin
e
ar
class
if
icatio
n
m
o
d
el
th
at
p
r
ed
icts
th
e
p
r
o
b
ab
ilit
y
o
f
a
b
in
ar
y
o
u
tco
m
e
.
I
t m
o
d
els th
e
p
r
o
b
a
b
ilit
y
(
=1
∣
)
u
s
in
g
th
e
lo
g
is
tic
f
u
n
ctio
n
g
iv
en
as
(
6
)
.
(
=
1
)
=
1
1
+
(
6
)
Her
e,
x
is
th
e
in
p
u
t
v
ec
to
r
,
w
is
th
e
weig
h
t
v
ec
to
r
,
an
d
e
i
s
th
e
b
ase
o
f
th
e
n
atu
r
al
alg
o
r
ith
m
.
RF
c
las
s
if
ier
cr
ea
tes
m
u
ltip
le
d
ec
is
io
n
tr
ee
s
d
u
r
in
g
th
e
tr
ain
in
g
p
h
ase.
L
et
u
s
co
n
s
id
er
(
)
to
b
e
th
e
p
r
ed
icti
o
n
o
f
th
e
i
-
t
h
d
ec
is
io
n
tr
ee
,
th
en
th
e
p
r
ed
ict
ed
class
lab
el
f
o
r
i
n
p
u
t
x
is
g
iv
en
as
p
er
(
7
)
.
Her
e
n
i
n
d
ica
tes
th
e
n
u
m
b
er
o
f
d
ec
is
io
n
tr
ee
s
.
̂
=
{
1
(
)
,
2
(
)
,
…
…
…
,
(
)
}
(
7
)
GB
is
also
an
en
s
em
b
le
m
eth
o
d
lik
e
RF
c
lass
if
ier
.
I
t
b
u
ild
s
a
s
tr
o
n
g
m
o
d
el
b
y
s
eq
u
en
tially
ac
cu
m
u
latin
g
wea
k
lear
n
e
r
s
(
ty
p
ically
d
ec
is
io
n
tr
ee
s
)
an
d
also
am
en
d
in
g
th
e
er
r
o
r
s
m
ad
e
b
y
p
r
e
v
io
u
s
lear
n
er
s
.
T
h
e
p
r
e
d
ictio
n
o
f
t
h
e
en
s
em
b
le
is
a
weig
h
ted
s
u
m
o
f
th
e
p
r
ed
ictio
n
s
o
f
all
th
e
wea
k
lear
n
er
s
.
C
o
n
s
id
er
(
)
as th
e
p
r
ed
ictio
n
o
f
th
e
m
-
th
wea
k
lea
r
n
er
.
T
h
e
f
in
al
p
r
ed
ictio
n
is
co
m
p
u
ted
as p
er
(
8
)
.
̂
=
∑
=
1
(
)
(
8
)
Her
e,
M
is
th
e
n
u
m
b
er
o
f
wea
k
lear
n
er
s
an
d
ar
e
th
e
co
r
r
esp
o
n
d
in
g
weig
h
ts
.
3.
6
.
Descript
io
n o
f
d
a
t
a
s
et
s
T
h
e
s
o
f
twar
e
f
au
lt p
r
ed
ictio
n
d
atasets
r
ef
er
en
ce
d
,
n
am
ely
J
M1
,
AR
1
,
C
M1
,
KC
2
,
MW1
,
PC
1
,
MC2
,
an
d
KC
1
,
a
r
e
well
-
k
n
o
wn
d
at
asets
co
m
m
o
n
ly
u
s
ed
in
r
esea
r
ch
f
o
r
e
v
alu
atin
g
ML
m
o
d
el
s
in
th
e
c
o
n
tex
t
o
f
s
o
f
twar
e
d
ef
ec
t
p
r
e
d
ictio
n
.
T
h
ese
d
atasets
co
n
tain
a
wid
e
r
an
g
e
o
f
s
tatic
co
d
e
m
et
r
ics
an
d
attr
ib
u
tes
ass
o
ciate
d
with
s
o
f
twar
e
m
o
d
u
les,
wh
ich
s
er
v
e
as
f
ea
tu
r
es
f
o
r
tr
ain
i
n
g
p
r
ed
ictiv
e
m
o
d
els.
T
h
e
J
MI
d
ataset
is
o
f
ten
u
s
ed
f
o
r
ev
alu
atin
g
ML
m
o
d
els
in
s
o
f
twar
e
d
ef
ec
t
p
r
ed
ictio
n
task
s
.
I
t
co
n
tain
s
s
tatic
co
d
e
m
etr
ics
an
d
attr
ib
u
tes
ex
t
r
ac
ted
f
r
o
m
J
av
a
p
r
o
jects,
in
clu
d
in
g
m
ea
s
u
r
es
r
elate
d
to
co
d
e
co
m
p
le
x
ity
,
s
ize,
an
d
o
b
ject
-
o
r
ien
ted
d
esig
n
p
r
o
p
er
ties
.
Attr
ib
u
tes
m
ay
in
clu
d
e
lin
es
o
f
co
d
e,
Mc
C
ab
e'
s
cy
clo
m
atic
co
m
p
lex
ity
,
Halstead
'
s
m
etr
ics,
an
d
v
ar
io
u
s
o
th
er
s
o
f
twar
e
m
etr
ics
[
2
8
]
,
[
2
9
]
.
T
h
e
r
esear
ch
in
[
3
0
]
–
[
3
2
]
it
is
d
em
o
n
s
tr
ate
s
h
o
w
th
ese
m
etr
ics
p
lay
a
r
o
le
in
s
o
f
twar
e
f
au
lt
p
r
ed
ictio
n
.
T
h
e
d
ataset
p
r
o
v
id
es
lab
eled
in
s
tan
ce
s
in
d
icatin
g
wh
eth
er
a
s
o
f
twar
e
m
o
d
u
le
co
n
tain
s
d
ef
ec
ts
o
r
n
o
t.
T
h
ey
co
n
s
is
t
o
f
attr
ib
u
tes
lik
e
s
tatic
co
d
e
m
etr
ics
an
d
attr
ib
u
tes
co
llected
f
r
o
m
s
o
f
twar
e
r
ep
o
s
ito
r
ies,
f
o
cu
s
in
g
o
n
c
o
d
e
c
h
u
r
n
an
d
ch
a
n
g
e
-
r
elate
d
m
etr
ics.
A
t
t
r
i
b
u
t
es
m
a
y
i
n
cl
u
d
e
l
i
n
es
ad
d
e
d
,
l
i
n
e
s
d
e
le
t
e
d
,
n
u
m
b
e
r
o
f
c
o
d
e
m
o
d
i
f
i
c
a
t
i
o
n
s
,
a
n
d
o
t
h
e
r
m
e
t
r
i
cs
r
e
l
a
t
e
d
t
o
c
o
d
e
e
v
o
l
u
ti
o
n
a
n
d
c
h
a
n
g
e
p
a
t
t
e
r
n
s
.
S
o
m
e
a
t
t
r
i
b
u
t
e
s
l
i
k
e
c
o
m
b
i
n
a
t
i
o
n
o
f
s
i
z
e
,
c
o
m
p
l
e
x
i
t
y
,
a
n
d
d
e
s
i
g
n
-
r
e
l
a
t
e
d
m
e
t
r
i
cs
,
d
e
p
e
n
d
i
n
g
o
n
t
h
e
s
p
e
c
i
f
i
c
c
h
a
r
a
c
te
r
i
s
t
i
cs
o
f
t
h
e
s
o
f
t
w
a
r
e
p
r
o
j
e
c
ts
f
r
o
m
w
h
i
c
h
t
h
e
d
a
t
a
w
a
s
c
o
l
l
e
c
t
e
d
,
a
r
e
i
n
c
l
u
d
e
d
.
O
v
e
r
al
l
,
t
h
es
e
d
a
t
as
e
ts
o
f
f
e
r
a
r
ic
h
s
o
u
r
c
e
o
f
i
n
f
o
r
m
a
t
i
o
n
f
o
r
t
r
a
in
i
n
g
a
n
d
e
v
a
l
u
a
ti
n
g
m
a
c
h
i
n
e
-
l
e
a
r
n
i
n
g
m
o
d
e
l
s
f
o
r
s
o
f
t
w
a
r
e
f
a
u
l
t
p
r
e
d
i
c
t
i
o
n
.
B
y
l
ev
e
r
a
g
i
n
g
t
h
e
d
i
v
e
r
s
e
s
e
t
o
f
s
t
ati
c
c
o
d
e
m
e
t
r
i
c
s
a
n
d
a
t
t
r
i
b
u
t
es
p
r
o
v
i
d
e
d
i
n
t
h
e
s
e
d
a
t
a
s
et
s
,
r
es
e
a
r
c
h
e
r
s
c
a
n
d
e
v
e
l
o
p
r
o
b
u
s
t
p
r
e
d
i
c
t
i
v
e
m
o
d
e
ls
c
ap
a
b
l
e
o
f
i
d
e
n
t
i
f
y
i
n
g
p
o
t
e
n
t
i
a
l
d
e
f
e
ct
s
i
n
s
o
f
t
wa
r
e
m
o
d
u
l
e
s
,
t
h
e
r
e
b
y
a
i
d
i
n
g
i
n
t
h
e
im
p
r
o
v
e
m
e
n
t
o
f
s
o
f
t
w
a
r
e
q
u
a
l
ity
a
n
d
r
e
l
i
a
b
i
l
it
y
.
4.
RE
SU
L
T
S AN
A
L
YS
I
S AN
D
DIS
CU
SS
I
O
N
Per
f
o
r
m
an
ce
ev
al
u
atio
n
an
d
t
esti
n
g
ar
e
cr
itical
asp
ec
t
s
o
f
ass
es
s
in
g
th
e
ef
f
ec
tiv
en
ess
an
d
ef
f
icien
cy
o
f
th
e
im
p
lem
en
ted
s
o
f
twar
e
f
au
lt
p
r
e
d
ictio
n
s
y
s
tem
.
W
e
ex
p
lain
th
e
ev
al
u
atio
n
p
r
o
ce
s
s
in
to
two
co
m
p
o
n
en
ts
:
alg
o
r
ith
m
tim
e
c
o
m
p
lex
ity
an
d
test
in
g
m
eth
o
d
o
lo
g
ies.
W
h
en
e
v
alu
atin
g
th
e
tim
e
co
m
p
lex
ity
o
f
d
if
f
er
en
t
d
ata
b
alan
cin
g
tech
n
iq
u
es
an
d
ML
alg
o
r
ith
m
s
,
it'
s
ess
en
tial
to
co
n
s
id
er
b
o
th
th
e
tim
e
r
eq
u
ir
e
d
f
o
r
tr
ain
in
g
th
e
m
o
d
els an
d
th
e
ti
m
e
n
ee
d
ed
f
o
r
m
ak
in
g
p
r
e
d
ictio
n
s
o
n
n
ew
d
ata.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
14
,
No
.
6
,
Dec
em
b
er
2
0
2
5
:
4
7
8
7
-
4
8
0
1
4792
GAN
s
d
em
o
n
s
tr
ated
a
tim
e
c
o
m
p
lex
ity
o
f
ap
p
r
o
x
im
ately
2
5
m
in
u
tes
p
er
d
ataset.
GANs
ty
p
ically
r
eq
u
ir
e
m
o
r
e
tim
e
f
o
r
tr
ain
in
g
co
m
p
a
r
ed
to
o
th
er
b
alan
cin
g
tech
n
iq
u
es
d
u
e
t
o
th
eir
co
m
p
l
ex
ar
ch
itectu
r
e
a
n
d
iter
ativ
e
tr
ain
in
g
p
r
o
ce
s
s
.
Ho
wev
er
,
th
e
y
ca
n
g
e
n
er
at
e
m
o
r
e
ac
cu
r
ate
an
d
p
r
ec
is
e
s
y
n
th
etic
d
ata.
Oth
er
b
alan
ci
n
g
tech
n
iq
u
es
(
e.
g
.
,
SMOT
E
,
Nea
r
Miss
)
d
e
m
o
n
s
tr
ated
a
tim
e
c
o
m
p
lex
it
y
th
at
was
ty
p
ically
1
to
2
s
ec
o
n
d
s
.
T
r
ad
itio
n
al
d
ata
b
alan
cin
g
tech
n
i
q
u
es
lik
e
SMOT
E
an
d
Nea
r
Miss
ar
e
co
m
p
u
tatio
n
ally
less
ex
p
en
s
iv
e
co
m
p
ar
ed
to
GA
Ns.
T
h
ey
in
v
o
lv
e
s
im
p
ler
a
lg
o
r
ith
m
s
an
d
g
en
er
ate
b
ala
n
ce
d
d
atasets
m
o
r
e
q
u
ick
ly
.
W
h
en
e
v
alu
atin
g
th
e
tim
e
c
o
m
p
lex
ity
o
f
ML
alg
o
r
ith
m
s
,
it'
s
cr
u
cial
to
c
o
n
s
id
er
b
o
th
tr
ain
i
n
g
tim
e
an
d
p
r
e
d
ictio
n
tim
e.
Dec
is
io
n
t
r
ee
s
h
av
e
a
tim
e
co
m
p
le
x
i
ty
as
O(
n
*
m
lo
g
m
)
,
w
h
er
e
n
is
th
e
n
u
m
b
er
o
f
s
am
p
les
,
wh
ile
m
is
th
e
n
u
m
b
er
o
f
f
ea
tu
r
es.
T
h
e
tim
e
co
m
p
lex
ity
o
f
R
F
is
O(
n
*
m
lo
g
m
*
k
)
,
w
h
er
e
k
is
th
e
n
u
m
b
er
o
f
tr
ee
s
e
x
is
tin
g
in
th
e
f
o
r
est.
SVM
h
as
a
tim
e
co
m
p
lex
ity
o
f
O(
n
2
*
m
)
,
w
h
e
r
e
n
is
th
e
n
u
m
b
er
o
f
s
am
p
les
an
d
m
is
th
e
n
u
m
b
er
o
f
f
ea
tu
r
es.
SVMs
ca
n
b
e
co
m
p
u
tatio
n
ally
ex
p
en
s
iv
e
f
o
r
lar
g
e
d
atasets
h
en
ce
,
we
d
id
n
o
t
em
p
lo
y
it
h
er
e.
KNN
h
as
a
tim
e
co
m
p
lex
ity
o
f
O(
n
*
k
*
m
)
,
wh
er
e
n
is
th
e
n
u
m
b
er
o
f
s
am
p
les,
k
is
th
e
n
u
m
b
e
r
o
f
n
eig
h
b
o
r
s
,
an
d
m
is
th
e
n
u
m
b
er
o
f
f
ea
tu
r
es.
LR
h
as
a
tim
e
co
m
p
lex
ity
o
f
O(
n
*
m
)
,
wh
er
e
n
is
th
e
n
u
m
b
er
o
f
s
am
p
les
an
d
m
is
th
e
n
u
m
b
er
o
f
f
ea
tu
r
es.
K
-
Me
an
s
h
as
a
tim
e
c
o
m
p
lex
ity
o
f
O(
n
*
k
*
t*
m
)
,
wh
er
e
n
is
th
e
n
u
m
b
er
o
f
s
am
p
les,
k
is
th
e
n
u
m
b
er
o
f
clu
s
ter
s
,
t
is
th
e
n
u
m
b
er
o
f
iter
atio
n
s
,
an
d
m
is
th
e
n
u
m
b
er
o
f
f
ea
tu
r
e
s
.
An
aly
zin
g
m
o
d
el
p
er
f
o
r
m
an
ce
o
n
th
e
KC
2
d
ataset
ac
r
o
s
s
d
if
f
er
en
t
d
ata
b
alan
cin
g
tech
n
iq
u
es
p
r
o
v
id
es
v
alu
ab
le
in
s
ig
h
ts
in
to
th
eir
ef
f
ec
tiv
en
ess
.
GAN
b
alan
ce
d
d
ata
co
n
s
is
ten
tly
en
h
a
n
ce
s
ac
cu
r
ac
y
ac
r
o
s
s
m
o
d
els
co
m
p
ar
ed
to
th
e
u
n
b
alan
ce
d
s
ce
n
ar
io
,
with
RF
lead
in
g
in
p
er
f
o
r
m
an
ce
.
T
h
is
s
u
g
g
ests
th
e
ef
f
icac
y
o
f
GAN
b
alan
cin
g
in
im
p
r
o
v
in
g
m
o
d
el
ac
cu
r
ac
y
wit
h
o
u
t
s
ig
n
if
ica
n
t
d
ata
lo
s
s
.
Ho
wev
er
,
em
p
l
o
y
in
g
d
ata
b
alan
c
in
g
tech
n
i
q
u
es
s
u
ch
as
SMOT
E
an
d
Nea
r
Mi
s
s
f
u
r
th
er
im
p
r
o
v
es
ac
cu
r
ac
y
,
n
o
tab
ly
co
m
p
ar
e
d
to
th
e
u
n
b
ala
n
ce
d
d
ata.
SMOT
E
b
alan
ce
d
d
ata
lead
s
to
s
ig
n
if
i
ca
n
t
ac
cu
r
ac
y
im
p
r
o
v
e
m
en
ts
ac
r
o
s
s
m
o
d
els,
with
R
F
an
d
KNN
ac
h
iev
in
g
th
e
h
ig
h
est
ac
cu
r
ac
y
.
Nea
r
Miss
b
alan
ce
d
d
ata
also
en
h
a
n
ce
s
ac
cu
r
ac
y
,
b
u
t
t
o
a
less
er
e
x
ten
t
co
m
p
ar
ed
to
SMOT
E
b
alan
cin
g
.
Fig
u
r
e
2
d
ep
icts
th
e
g
e
n
er
ato
r
an
d
d
is
cr
im
in
ato
r
lo
s
s
o
f
KC
2
,
wh
ile
Fig
u
r
e
3
d
ep
icts
th
e
c
o
m
p
ar
is
o
n
o
f
ac
cu
r
ac
ies
f
o
r
ML
alg
o
r
i
th
m
s
ap
p
lied
to
KC
2
.
I
n
s
u
m
m
ar
y
,
wh
ile
GAN
b
alan
cin
g
s
h
o
ws
p
r
o
m
is
e,
SMOT
E
b
alan
cin
g
em
er
g
es
a
s
th
e
m
o
s
t
ef
f
ec
tiv
e
tech
n
iq
u
e
f
o
r
im
p
r
o
v
in
g
m
o
d
el
p
er
f
o
r
m
an
ce
o
n
th
e
KC
2
d
ataset,
f
o
llo
wed
b
y
Nea
r
M
is
s
b
alan
cin
g
.
T
h
ese
in
s
ig
h
t
s
aid
in
s
elec
tin
g
ap
p
r
o
p
r
ia
te
d
ata
b
alan
cin
g
tech
n
iq
u
es
f
o
r
class
if
icatio
n
task
s
o
n
th
e
KC
2
d
atase
t.
I
n
th
e
an
aly
s
is
f
o
r
th
e
KC
2
d
ata
s
et
,
th
e
g
en
er
ato
r
lo
s
s
r
ea
ch
ed
its
m
in
im
u
m
at
e
p
o
c
h
3
4
5
2
as
d
e
p
icted
i
n
Fig
u
r
e
2
,
in
d
icatin
g
o
p
tim
al
p
e
r
f
o
r
m
an
ce
f
o
r
s
y
n
th
etic
d
ata
g
en
e
r
atio
n
u
s
in
g
th
e
GAN
m
o
d
el.
Me
a
n
wh
ile,
th
e
d
is
cr
im
in
ato
r
lo
s
s
also
d
ec
r
ea
s
es,
in
d
icatin
g
im
p
r
o
v
e
d
d
is
cr
im
in
atio
n
b
etwe
en
r
ea
l
a
n
d
s
y
n
t
h
etic
d
ata.
T
h
is
m
iles
to
n
e
s
u
g
g
ests
th
at
th
e
g
en
er
a
to
r
h
as
c
o
n
v
e
r
g
ed
,
p
r
o
d
u
cin
g
th
e
h
ig
h
est q
u
ality
s
y
n
th
etic
d
ata
am
o
n
g
th
e
1
0
,
0
0
0
ep
o
ch
s
tr
ain
ed
.
Fig
u
r
e
2
.
Gen
e
r
ato
r
a
n
d
d
is
cr
i
m
in
ato
r
lo
s
s
o
f
KC
2
Fig
u
r
e
3
.
Acc
u
r
ac
y
c
o
m
p
ar
is
o
n
o
f
KC
2
An
aly
zin
g
m
o
d
el
p
er
f
o
r
m
an
c
e
o
n
th
e
J
M1
d
ataset
ac
r
o
s
s
v
ar
io
u
s
d
ata
b
alan
cin
g
tech
n
iq
u
es
r
ev
ea
ls
in
s
ig
h
tf
u
l
tr
en
d
s
.
I
n
th
e
an
al
y
s
is
f
o
r
th
e
J
M1
d
ataset
,
as
d
e
p
icted
in
Fig
u
r
e
4
,
th
e
g
en
er
a
to
r
lo
s
s
r
ea
ch
ed
its
m
in
im
u
m
at
ep
o
c
h
1
1
3
2
,
in
d
icatin
g
o
p
tim
al
p
er
f
o
r
m
a
n
ce
f
o
r
s
y
n
th
etic
d
ata
g
e
n
er
atio
n
u
s
in
g
th
e
GAN
m
o
d
el.
Me
an
wh
ile,
th
e
d
is
cr
im
in
ato
r
lo
s
s
also
d
ec
r
ea
s
es,
in
d
ica
tin
g
im
p
r
o
v
ed
d
is
cr
im
in
atio
n
b
etwe
en
r
ea
l
a
n
d
s
y
n
th
etic
d
ata.
T
h
is
m
iles
to
n
e
s
u
g
g
ests
th
at
th
e
g
en
er
at
o
r
h
as
co
n
v
er
g
e
d
,
p
r
o
d
u
cin
g
t
h
e
h
ig
h
est
q
u
ality
s
y
n
th
etic
d
ata
a
m
o
n
g
th
e
1
0
,
0
0
0
ep
o
ch
s
tr
ain
e
d
.
GAN
-
b
alan
ce
d
d
ata
c
o
n
s
is
ten
tly
im
p
r
o
v
es
ac
c
u
r
ac
y
co
m
p
ar
ed
t
o
th
e
u
n
b
alan
ce
d
s
ce
n
ar
io
ac
r
o
s
s
m
o
d
els,
with
R
F
s
h
o
wca
s
in
g
p
ar
ticu
lar
ly
h
ig
h
ac
cu
r
ac
y
.
T
h
is
u
n
d
er
s
co
r
es
GAN
b
alan
cin
g
'
s
ef
f
ec
tiv
en
ess
in
en
h
an
cin
g
m
o
d
el
p
e
r
f
o
r
m
an
ce
wh
il
e
m
ain
tain
in
g
d
ata
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
E
n
h
a
n
ci
n
g
s
o
ftw
a
r
e
fa
u
lt p
r
ed
ictio
n
th
r
o
u
g
h
d
a
ta
b
a
la
n
cin
g
tech
n
iq
u
es a
n
d
ma
ch
i
n
e
lea
r
n
in
g
(
A
ksh
a
t R
a
j)
4793
in
teg
r
ity
.
Ho
we
v
er
,
e
m
p
lo
y
i
n
g
tech
n
i
q
u
es
lik
e
SMOT
E
a
n
d
Nea
r
Miss
f
u
r
th
er
en
h
an
ce
s
ac
cu
r
ac
y
,
n
o
tab
ly
co
m
p
ar
ed
to
t
h
e
u
n
b
ala
n
ce
d
d
ata.
SMOT
E
b
alan
ce
d
d
ata
n
o
tab
ly
im
p
r
o
v
es
ac
cu
r
ac
y
ac
r
o
s
s
m
o
d
els,
th
o
u
g
h
LR
ex
h
ib
its
lo
wer
ac
cu
r
ac
y
with
th
is
tech
n
iq
u
e
,
as
ca
n
b
e
in
f
er
r
ed
f
r
o
m
Fig
u
r
e
5
.
Nea
r
Miss
b
alan
ce
d
d
ata
also
en
h
an
ce
s
ac
cu
r
ac
y
,
alb
ei
t
to
a
les
s
er
ex
ten
t
co
m
p
ar
ed
to
SMOT
E
b
alan
cin
g
.
I
n
s
u
m
m
ar
y
,
wh
ile
GAN
b
alan
cin
g
s
h
o
ws
p
r
o
m
is
e,
SMOT
E
b
alan
cin
g
em
er
g
es
as
th
e
m
o
s
t
ef
f
ec
tiv
e
tech
n
iq
u
e
f
o
r
im
p
r
o
v
in
g
m
o
d
el
p
er
f
o
r
m
an
ce
o
n
th
e
J
M1
d
ata
s
et,
f
o
llo
wed
cl
o
s
ely
b
y
Nea
r
Miss
b
alan
cin
g
.
T
h
ese
in
s
ig
h
t
s
ass
is
t
in
s
elec
tin
g
ap
p
r
o
p
r
iate
d
ata
b
ala
n
cin
g
tec
h
n
iq
u
es f
o
r
class
if
icatio
n
task
s
o
n
th
e
J
M1
d
ataset.
Fig
u
r
e
4
.
Gen
e
r
ato
r
a
n
d
d
is
cr
i
m
in
ato
r
lo
s
s
o
f
J
M1
Fig
u
r
e
5
.
Acc
u
r
ac
y
c
o
m
p
ar
is
o
n
o
f
J
M1
I
n
t
h
e
a
n
a
l
y
s
is
f
o
r
t
h
e
KC
1
d
ata
s
e
t
,
t
h
e
g
e
n
e
r
at
o
r
l
o
s
s
r
e
a
c
h
ed
i
t
s
m
i
n
i
m
u
m
a
t
e
p
o
c
h
4
1
2
8
a
s
d
e
p
i
ct
e
d
i
n
F
i
g
u
r
e
6
.
I
t
i
n
d
i
c
a
t
e
s
t
h
e
o
p
t
i
m
a
l
p
e
r
f
o
r
m
a
n
c
e
f
o
r
s
y
n
t
h
e
t
i
c
d
a
t
a
g
e
n
e
r
a
t
i
o
n
u
s
i
n
g
t
h
e
G
A
N
m
o
d
e
l
.
M
e
a
n
w
h
i
l
e
,
t
h
e
d
is
c
r
i
m
i
n
at
o
r
lo
s
s
al
s
o
d
e
c
r
e
as
es
,
i
n
d
i
c
at
i
n
g
im
p
r
o
v
e
d
d
i
s
c
r
i
m
i
n
a
ti
o
n
b
e
t
w
ee
n
r
e
a
l
a
n
d
s
y
n
t
h
e
ti
c
d
a
t
a
.
T
h
is
m
il
es
t
o
n
e
s
u
g
g
e
s
ts
t
h
a
t
t
h
e
g
e
n
e
r
a
t
o
r
h
as
c
o
n
v
e
r
g
e
d
,
p
r
o
d
u
c
i
n
g
t
h
e
h
i
g
h
e
s
t
q
u
a
l
i
t
y
s
y
n
t
h
e
ti
c
d
a
ta
a
m
o
n
g
t
h
e
1
0
,
0
0
0
e
p
o
c
h
s
t
r
a
i
n
e
d
.
T
h
e
K
C
1
d
a
ta
s
et
a
n
a
l
y
s
is
r
e
v
e
a
ls
t
h
a
t
G
A
N
b
al
a
n
c
ed
d
a
t
a
c
o
n
s
is
t
e
n
tl
y
i
m
p
r
o
v
e
s
a
c
c
u
r
a
c
y
a
c
r
o
s
s
m
o
d
e
l
s
,
i
n
d
i
c
a
ti
n
g
i
ts
e
f
f
e
c
t
i
v
e
n
ess
w
i
t
h
o
u
t
d
a
t
a
l
o
s
s
.
U
n
b
al
a
n
c
e
d
d
a
t
a
l
e
a
d
s
t
o
l
o
w
e
r
a
c
c
u
r
a
c
y
,
b
u
t
S
MO
T
E
a
n
d
N
e
a
r
M
is
s
t
e
c
h
n
i
q
u
e
s
n
o
t
a
b
l
y
e
n
h
a
n
c
e
i
t
.
SM
O
T
E
b
al
a
n
ce
d
d
a
t
a
v
a
r
i
es
i
n
e
f
f
e
c
t
i
v
e
n
es
s
a
m
o
n
g
m
o
d
e
ls
,
w
i
t
h
R
F
a
n
d
K
N
N
b
e
n
e
f
i
t
i
n
g
t
h
e
m
o
s
t
a
s
ca
n
b
e
s
e
e
n
f
r
o
m
F
i
g
u
r
e
7
.
N
e
a
r
M
is
s
b
a
l
a
n
c
e
d
d
a
t
a
m
a
i
n
t
a
i
n
s
h
i
g
h
a
c
c
u
r
a
c
y
,
t
h
o
u
g
h
n
o
t
s
i
g
n
i
f
i
c
a
n
t
l
y
b
e
t
t
e
r
t
h
a
n
S
M
O
T
E
b
a
la
n
c
ed
d
a
t
a
.
I
n
s
u
m
m
a
r
y
,
G
A
N
b
al
a
n
c
i
n
g
i
s
r
o
b
u
s
t
f
o
r
K
C
1
,
b
u
t
t
h
e
e
f
f
e
c
ti
v
e
n
e
s
s
o
f
S
MO
T
E
a
n
d
N
e
a
r
Mi
s
s
t
e
c
h
n
i
q
u
e
s
v
a
r
ie
s
a
c
r
o
s
s
m
o
d
e
l
s
,
a
i
d
i
n
g
i
n
s
e
l
ec
t
i
n
g
s
u
i
ta
b
l
e
b
a
l
a
n
c
i
n
g
t
ec
h
n
i
q
u
e
s
f
o
r
cl
a
s
s
i
f
i
ca
t
i
o
n
t
as
k
s
o
n
t
h
e
K
C
1
d
a
t
a
s
et
.
Fig
u
r
e
6
.
Gen
e
r
ato
r
a
n
d
d
is
cr
i
m
in
ato
r
lo
s
s
o
f
KC
1
Fig
u
r
e
7
.
Acc
u
r
ac
y
c
o
m
p
ar
is
o
n
o
f
KC
1
Fig
u
r
e
8
d
e
p
i
c
t
s
th
e
g
en
er
ato
r
an
d
d
is
cr
im
in
ato
r
lo
s
s
o
f
PC
1
d
ataset.
I
n
th
e
an
aly
s
is
f
o
r
th
e
PC
1
d
ataset,
th
e
g
en
er
ato
r
lo
s
s
r
ea
ch
ed
its
m
in
im
u
m
at
e
p
o
ch
1
7
6
7
,
in
d
icatin
g
o
p
tim
al
p
er
f
o
r
m
an
ce
f
o
r
s
y
n
th
etic
d
ata
g
en
e
r
atio
n
u
s
in
g
th
e
GAN
m
o
d
el.
Me
a
n
wh
ile,
th
e
d
is
cr
im
in
ato
r
lo
s
s
also
d
ec
r
ea
s
es,
in
d
icatin
g
im
p
r
o
v
e
d
d
is
cr
im
in
atio
n
b
etwe
en
r
ea
l
a
n
d
s
y
n
t
h
etic
d
ata.
T
h
is
m
iles
to
n
e
s
u
g
g
ests
th
at
th
e
g
en
er
a
to
r
h
as
c
o
n
v
e
r
g
ed
,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
14
,
No
.
6
,
Dec
em
b
er
2
0
2
5
:
4
7
8
7
-
4
8
0
1
4794
p
r
o
d
u
cin
g
th
e
h
ig
h
est
q
u
ality
s
y
n
th
etic
d
ata
am
o
n
g
th
e
1
0
,
0
0
0
ep
o
c
h
s
tr
ain
ed
.
An
al
y
zin
g
th
e
PC
1
d
ataset's
d
ata
b
alan
cin
g
tech
n
iq
u
es
r
ev
ea
ls
d
is
tin
ct
p
atter
n
s
in
m
o
d
e
l
p
er
f
o
r
m
an
ce
.
GAN
b
ala
n
ce
d
d
ata
co
n
s
is
ten
tly
b
o
o
s
ts
ac
cu
r
ac
y
ac
r
o
s
s
m
o
d
els
,
with
R
F a
ch
iev
in
g
th
e
h
ig
h
est ac
cu
r
ac
y
.
C
o
n
v
e
r
s
ely
,
u
n
b
al
an
ce
d
d
ata
lead
s
to
d
ec
r
ea
s
ed
ac
cu
r
ac
y
,
u
n
d
er
lin
in
g
th
e
ch
allen
g
es
p
o
s
ed
b
y
im
b
alan
ce
d
class
d
is
tr
ib
u
tio
n
s
.
Ho
wev
er
,
b
o
t
h
SMOT
E
an
d
Nea
r
Miss
tech
n
i
q
u
es
n
o
tab
ly
en
h
a
n
ce
ac
c
u
r
ac
y
co
m
p
ar
ed
to
th
e
u
n
b
alan
ce
d
s
ce
n
ar
io
.
SMOT
E
b
alan
ce
d
d
ata
v
a
r
ies
in
ef
f
ec
tiv
en
ess
am
o
n
g
m
o
d
els,
p
ar
ticu
lar
ly
b
en
ef
itin
g
R
F
an
d
GB
.
Nea
r
Miss
b
alan
ce
d
d
ata
m
ain
tain
s
co
n
s
is
ten
tly
h
ig
h
ac
cu
r
ac
y
,
ac
h
iev
in
g
p
er
f
e
ct
s
co
r
es
f
o
r
R
F
an
d
GB
.
Fig
u
r
e
9
d
e
p
icts
th
e
ac
cu
r
ac
y
o
f
PC
1
d
atasets
ac
r
o
s
s
all
th
e
m
o
d
els
d
ev
elo
p
ed
.
Ho
wev
er
,
LR
an
d
KNN
s
h
o
w
s
lig
h
tly
lo
wer
ac
cu
r
ac
y
co
m
p
a
r
ed
to
o
th
er
tech
n
iq
u
es.
I
n
s
u
m
m
ar
y
,
GA
N
b
alan
cin
g
em
er
g
es
as
a
r
o
b
u
s
t
s
tr
ateg
y
f
o
r
im
p
r
o
v
in
g
m
o
d
el
p
er
f
o
r
m
an
ce
o
n
th
e
PC
1
d
ataset,
wh
ile
th
e
ef
f
ec
tiv
en
ess
o
f
SMOT
E
an
d
Nea
r
Miss
tech
n
iq
u
es
v
ar
ies
ac
r
o
s
s
m
o
d
els.
T
h
ese
in
s
ig
h
ts
o
f
f
e
r
g
u
id
an
ce
in
s
elec
tin
g
ap
p
r
o
p
r
iate
d
ata
b
alan
cin
g
tech
n
iq
u
es
f
o
r
class
if
icatio
n
task
s
o
n
th
e
PC
1
d
ata
s
et.
I
n
th
e
an
aly
s
is
f
o
r
th
e
C
M1
d
atase
t,
th
e
g
en
er
ato
r
lo
s
s
r
ea
ch
ed
its
m
in
im
u
m
at
e
p
o
c
h
6
1
8
9
,
i
n
d
icatin
g
o
p
tim
al
p
e
r
f
o
r
m
a
n
ce
f
o
r
s
y
n
th
etic
d
ata
g
en
er
atio
n
u
s
in
g
t
h
e
GAN
m
o
d
el.
Me
an
wh
ile,
th
e
d
is
cr
im
in
ato
r
lo
s
s
also
d
ec
r
ea
s
es,
in
d
icatin
g
im
p
r
o
v
ed
d
is
cr
im
in
atio
n
b
etwe
en
r
ea
l
an
d
s
y
n
t
h
etic
d
ata.
T
h
is
m
iles
to
n
e
s
u
g
g
ests
th
at
th
e
g
en
er
ato
r
h
as
co
n
v
er
g
e
d
,
p
r
o
d
u
cin
g
t
h
e
h
ig
h
est
q
u
ality
s
y
n
th
etic
d
ata
a
m
o
n
g
t
h
e
1
0
,
0
0
0
e
p
o
ch
s
tr
ai
n
ed
.
Fig
u
r
e
8
.
Gen
e
r
ato
r
a
n
d
d
is
cr
i
m
in
ato
r
lo
s
s
o
f
PC
1
Fig
u
r
e
9
.
Acc
u
r
ac
y
c
o
m
p
ar
is
o
n
o
f
PC
1
Fig
u
r
e
1
0
d
em
o
n
s
tr
ates
th
e
g
en
er
ato
r
an
d
d
is
cr
im
in
ato
r
l
o
s
s
f
o
r
C
M1
d
ataset
,
wh
ile
Fig
u
r
e
1
1
d
ep
icts
th
e
ac
cu
r
ac
y
o
b
tai
n
ed
ac
r
o
s
s
ML
m
o
d
els
d
ev
el
o
p
ed
in
th
e
s
tu
d
y
o
f
C
M1
d
ataset.
An
aly
zin
g
th
e
C
M1
d
ataset's
d
ata
b
alan
cin
g
tec
h
n
iq
u
es
s
h
ed
s
lig
h
t
o
n
m
o
d
e
l
p
er
f
o
r
m
an
ce
ac
r
o
s
s
v
ar
io
u
s
s
ce
n
ar
io
s
.
GAN
b
alan
ce
d
d
ata
co
n
s
is
ten
tly
en
h
an
ce
s
ac
cu
r
ac
y
ac
r
o
s
s
m
o
d
els,
with
R
F
lead
in
g
i
n
p
er
f
o
r
m
an
ce
.
T
h
is
u
n
d
er
s
co
r
es
th
e
ef
f
ec
tiv
e
n
ess
o
f
GAN
b
alan
cin
g
in
im
p
r
o
v
in
g
m
o
d
el
ac
cu
r
ac
y
with
o
u
t
s
ig
n
if
ican
t
d
ata
lo
s
s
.
C
o
n
v
er
s
ely
,
u
n
b
alan
ce
d
da
ta
lead
s
to
r
ed
u
ce
d
ac
c
u
r
ac
y
,
h
i
g
h
lig
h
tin
g
th
e
c
h
allen
g
es
p
o
s
ed
b
y
im
b
alan
ce
d
class
d
is
tr
ib
u
tio
n
s
.
F
i
g
u
r
e
1
0
.
G
e
n
e
r
a
t
o
r
a
n
d
d
i
s
c
r
im
i
n
a
t
o
r
l
o
s
s
o
f
C
M
1
Fig
u
r
e
1
1
.
Acc
u
r
ac
y
co
m
p
a
r
is
o
n
o
f
C
M1
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
E
n
h
a
n
ci
n
g
s
o
ftw
a
r
e
fa
u
lt p
r
ed
ictio
n
th
r
o
u
g
h
d
a
ta
b
a
la
n
cin
g
tech
n
iq
u
es a
n
d
ma
ch
i
n
e
lea
r
n
in
g
(
A
ksh
a
t R
a
j)
4795
SMOT
E
b
alan
ce
d
d
ata
d
em
o
n
s
tr
ates
v
ar
ied
ef
f
ec
tiv
en
ess
am
o
n
g
m
o
d
els,
p
ar
ticu
lar
ly
b
en
ef
itin
g
R
F
an
d
GB
.
T
h
is
tech
n
iq
u
e
s
u
b
s
tan
tially
en
h
an
ce
s
ac
cu
r
ac
y
ac
r
o
s
s
m
o
d
els
co
m
p
ar
ed
to
th
e
u
n
b
alan
ce
d
s
ce
n
ar
io
.
Nea
r
Miss
b
alan
ce
d
d
ata
m
ain
tain
s
h
ig
h
ac
cu
r
ac
y
lev
els,
p
a
r
ticu
lar
ly
ac
h
iev
in
g
p
er
f
e
ct
s
co
r
es
f
o
r
LR
.
Ho
wev
er
,
KNN
ex
h
ib
its
s
lig
h
tly
lo
wer
ac
cu
r
ac
y
co
m
p
ar
ed
to
o
th
er
tec
h
n
iq
u
es.
I
n
s
u
m
m
ar
y
,
GAN
b
alan
cin
g
em
er
g
es
as
a
r
elia
b
le
s
tr
ateg
y
f
o
r
en
h
an
ci
n
g
m
o
d
el
p
er
f
o
r
m
an
ce
o
n
th
e
C
M1
d
ataset,
wh
il
e
SMOT
E
an
d
Nea
r
Miss
tech
n
i
q
u
es
o
f
f
er
a
d
d
itio
n
al
im
p
r
o
v
e
m
en
ts
with
v
ar
y
in
g
ef
f
ec
tiv
e
n
ess
am
o
n
g
m
o
d
els.
T
h
ese
in
s
ig
h
ts
p
r
o
v
id
e
g
u
id
a
n
ce
f
o
r
s
elec
tin
g
ap
p
r
o
p
r
iate
d
ata
b
alan
ci
n
g
tec
h
n
iq
u
es
f
o
r
class
if
icatio
n
task
s
o
n
th
e
C
M1
d
ataset.
F
i
g
u
r
e
1
2
d
ep
i
c
t
s
th
e
g
en
e
r
a
t
o
r
a
n
d
d
i
s
c
r
im
i
n
a
to
r
l
o
s
s
o
f
M
C
2
,
w
h
i
l
e
F
i
g
u
r
e
1
3
d
e
p
i
c
t
s
th
e
a
c
c
u
r
a
c
y
o
f
M
C
2
f
o
r
t
h
e
ML
m
o
d
e
l
s
c
o
n
s
i
d
er
e
d
.
I
n
t
h
e
a
n
a
l
y
s
i
s
f
o
r
th
e
M
C
2
d
a
t
a
s
e
t,
t
h
e
g
e
n
e
r
a
to
r
l
o
s
s
r
e
a
c
h
e
d
i
t
s
m
in
i
m
u
m
a
t
e
p
o
ch
8
9
9
8
,
i
n
d
i
c
a
t
in
g
o
p
t
im
a
l
p
e
r
f
o
r
m
an
c
e
f
o
r
s
y
n
t
h
e
t
i
c
d
a
ta
g
e
n
e
r
a
t
io
n
u
s
i
n
g
t
h
e
G
A
N
m
o
d
e
l
.
M
ea
n
wh
i
l
e,
t
h
e
d
i
s
cr
i
m
in
a
t
o
r
l
o
s
s
a
l
s
o
d
e
c
r
ea
s
e
s
,
i
n
d
ic
a
t
i
n
g
i
m
p
r
o
v
e
d
d
i
s
c
r
i
m
in
a
t
i
o
n
b
e
t
w
e
en
r
ea
l
a
n
d
s
y
n
t
h
e
t
i
c
d
a
t
a
.
T
h
i
s
m
i
l
e
s
t
o
n
e
s
u
g
g
e
s
t
s
t
h
a
t
t
h
e
g
e
n
er
a
t
o
r
h
a
s
c
o
n
v
er
g
e
d
,
p
r
o
d
u
c
in
g
t
h
e
h
i
g
h
e
s
t
q
u
a
l
i
ty
s
y
n
t
h
e
t
i
c
d
a
t
a
a
m
o
n
g
th
e
1
0
,
0
0
0
e
p
o
ch
s
t
r
a
i
n
e
d
.
C
o
m
p
ar
i
n
g
d
a
t
a
b
a
la
n
c
i
n
g
t
e
ch
n
iq
u
e
s
o
n
t
h
e
M
C
2
d
a
t
a
s
e
t
r
ev
e
a
l
s
d
i
s
t
i
n
c
t
p
e
r
f
o
r
m
an
c
e
p
a
t
te
r
n
s
.
G
A
N
b
a
l
an
c
ed
d
a
t
a
g
e
n
er
a
l
ly
en
h
an
c
e
s
m
o
d
e
l
a
c
c
u
r
a
c
y
b
y
g
e
n
e
r
a
t
in
g
b
a
l
an
c
e
d
d
a
t
a
s
e
t
s
w
i
t
h
o
u
t
s
ig
n
if
ic
a
n
t
i
n
f
o
r
m
a
t
io
n
lo
s
s
.
U
n
b
a
la
n
c
ed
d
a
t
a
le
a
d
s
to
l
o
w
e
r
a
cc
u
r
a
c
y
a
s
m
o
d
e
l
s
s
t
r
u
g
g
l
e
w
i
th
m
in
o
r
i
ty
c
l
a
s
s
r
e
p
r
e
s
en
t
a
t
i
o
n
.
S
M
O
T
E
b
a
la
n
c
e
d
d
a
t
a
v
ar
i
e
s
i
n
e
f
f
e
c
t
iv
en
e
s
s
a
m
o
n
g
m
o
d
e
l
s
;
LR
n
o
t
a
b
ly
s
t
r
u
g
g
le
s
.
N
e
ar
M
i
s
s
b
a
l
a
n
ce
d
d
a
t
a
s
l
i
g
h
tl
y
d
e
c
r
ea
s
e
s
K
N
N
's
a
c
c
u
r
a
c
y
c
o
m
p
ar
e
d
t
o
o
t
h
er
t
e
ch
n
iq
u
e
s
.
I
n
s
u
m
m
a
r
y
,
w
h
i
l
e
G
A
N
b
a
l
an
c
i
n
g
co
n
s
i
s
t
e
n
t
l
y
im
p
r
o
v
e
s
a
c
c
u
r
a
c
y
,
th
e
ef
f
e
c
t
iv
e
n
e
s
s
o
f
S
M
O
T
E
a
n
d
Ne
a
r
M
i
s
s
t
e
c
h
n
i
q
u
e
s
v
a
r
ie
s
a
cr
o
s
s
m
o
d
e
l
s
.
T
h
e
s
e
i
n
s
i
g
h
t
s
a
id
i
n
s
e
l
e
c
t
i
n
g
t
h
e
m
o
s
t
s
u
i
t
ab
l
e
b
a
l
a
n
c
in
g
te
c
h
n
i
q
u
e
f
o
r
c
l
a
s
s
i
f
i
c
a
t
i
o
n
t
a
s
k
s
o
n
t
h
e
M
C
2
d
a
t
a
s
e
t
,
c
o
n
s
i
d
er
i
n
g
s
p
e
c
i
f
i
c
m
o
d
e
l
r
e
q
u
i
r
e
m
en
t
s
a
n
d
d
a
t
a
s
e
t
c
h
ar
a
c
t
er
i
s
t
i
c
s
.
Fig
u
r
e
1
2
.
Gen
er
ato
r
an
d
d
is
cr
im
in
ato
r
lo
s
s
o
f
MC2
Fig
u
r
e
1
3
.
Acc
u
r
ac
y
co
m
p
a
r
is
o
n
o
f
MC2
Fig
u
r
e
1
4
d
e
p
icts
th
e
g
e
n
er
ato
r
an
d
d
is
cr
im
in
ato
r
lo
s
s
o
f
M
C
2
wh
ile
Fig
u
r
e
1
5
d
ep
icts
th
e
ac
cu
r
ac
y
o
f
MC2
f
o
r
ML
m
o
d
els
co
n
s
id
er
ed
.
An
aly
zin
g
m
o
d
el
p
er
f
o
r
m
an
ce
o
n
th
e
MW1
d
ataset
ac
r
o
s
s
v
ar
io
u
s
d
ata
b
alan
cin
g
tech
n
iq
u
es
r
ev
ea
ls
n
o
tab
le
tr
en
d
s
.
GAN
b
alan
ce
d
d
ata
co
n
s
is
ten
tly
en
h
an
ce
s
ac
cu
r
ac
y
co
m
p
ar
ed
to
th
e
u
n
b
alan
ce
d
s
ce
n
ar
io
,
wit
h
R
F
ac
h
iev
in
g
th
e
h
ig
h
est
ac
cu
r
ac
y
am
o
n
g
m
o
d
els.
T
h
i
s
u
n
d
er
s
co
r
es
th
e
ef
f
ec
tiv
en
ess
o
f
GAN
b
alan
ci
n
g
in
im
p
r
o
v
in
g
m
o
d
el
p
er
f
o
r
m
an
ce
with
o
u
t
s
ig
n
if
ica
n
t
d
ata
lo
s
s
.
Ho
wev
er
,
em
p
lo
y
in
g
tech
n
iq
u
es
lik
e
SMOT
E
an
d
Nea
r
Miss
f
u
r
t
h
er
b
o
o
s
ts
ac
cu
r
ac
y
n
o
tab
ly
co
m
p
a
r
ed
to
th
e
u
n
b
alan
ce
d
d
ata.
SMOT
E
b
al
an
ce
d
d
ata
n
o
tab
l
y
im
p
r
o
v
es
ac
cu
r
ac
y
ac
r
o
s
s
m
o
d
els,
with
LR
ac
h
iev
in
g
t
h
e
h
ig
h
est
ac
cu
r
ac
y
.
Nea
r
Miss
b
alan
ce
d
d
ata
also
en
h
an
ce
s
ac
cu
r
ac
y
b
u
t
to
a
less
er
ex
ten
t
co
m
p
ar
ed
to
SMOT
E
b
alan
cin
g
.
I
n
s
u
m
m
a
r
y
,
wh
ile
GAN
b
alan
cin
g
s
h
o
ws
p
r
o
m
is
e,
SMOT
E
b
alan
cin
g
em
er
g
es
as
th
e
m
o
s
t
ef
f
ec
tiv
e
tech
n
iq
u
e
f
o
r
en
h
a
n
c
in
g
m
o
d
el
p
er
f
o
r
m
an
ce
o
n
t
h
e
MW1
d
ataset,
f
o
llo
wed
b
y
N
ea
r
Miss
b
alan
cin
g
.
T
h
ese
in
s
ig
h
ts
g
u
id
e
in
s
elec
t
in
g
ap
p
r
o
p
r
iate
d
ata
b
alan
cin
g
tech
n
iq
u
es
f
o
r
class
if
icatio
n
task
s
o
n
th
e
MW1
d
ataset.
I
n
th
e
an
al
y
s
is
f
o
r
th
e
MW1
d
ataset,
th
e
g
en
er
a
to
r
lo
s
s
r
ea
ch
ed
its
m
in
im
u
m
at
ep
o
ch
3
6
4
9
,
in
d
icatin
g
o
p
tim
al
p
er
f
o
r
m
a
n
ce
f
o
r
s
y
n
th
etic
d
ata
g
e
n
e
r
atio
n
u
s
in
g
th
e
GAN
m
o
d
el.
Me
an
wh
ile,
th
e
d
is
cr
im
in
ato
r
lo
s
s
also
d
ec
r
ea
s
es,
in
d
icatin
g
im
p
r
o
v
ed
d
is
cr
im
in
atio
n
b
etwe
en
r
ea
l
an
d
s
y
n
th
etic
d
ata.
T
h
is
m
iles
to
n
e
s
u
g
g
ests
th
at
t
h
e
g
en
er
ato
r
h
as
co
n
v
er
g
ed
,
p
r
o
d
u
cin
g
th
e
h
ig
h
est
q
u
ality
s
y
n
th
etic
d
ata
am
o
n
g
th
e
1
0
,
0
0
0
e
p
o
ch
s
tr
ain
e
d
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
14
,
No
.
6
,
Dec
em
b
er
2
0
2
5
:
4
7
8
7
-
4
8
0
1
4796
Fig
u
r
e
1
4
.
Gen
er
ato
r
an
d
d
is
cr
im
in
ato
r
lo
s
s
o
f
MC2
Fig
u
r
e
1
5
.
Acc
u
r
ac
y
co
m
p
a
r
is
o
n
o
f
MW1
Fig
u
r
e
1
6
d
e
p
icts
th
e
g
en
er
at
o
r
an
d
d
is
cr
im
in
ato
r
l
o
s
s
o
f
A
R
1
wh
ile
Fig
u
r
e
1
7
d
ep
icts
th
e
ac
cu
r
ac
y
o
f
AR
1
f
o
r
ML
m
o
d
els
co
n
s
i
d
er
ed
.
E
x
am
in
in
g
m
o
d
el
p
er
f
o
r
m
an
ce
o
n
th
e
AR
1
d
ataset
ac
r
o
s
s
d
iv
er
s
e
d
ata
b
alan
cin
g
s
tr
ateg
ies
p
r
o
v
id
es
in
s
ig
h
tf
u
l
o
b
s
er
v
atio
n
s
.
GA
N
b
alan
ce
d
d
ata
g
en
er
ally
e
n
h
an
ce
s
ac
cu
r
ac
y
co
m
p
ar
ed
to
t
h
e
u
n
b
ala
n
ce
d
s
ettin
g
,
esp
ec
ially
b
en
ef
itin
g
R
F
an
d
KNN
m
o
d
els.
T
h
is
u
n
d
er
s
co
r
es
th
e
ef
f
ec
tiv
en
ess
o
f
GAN
b
ala
n
cin
g
in
im
p
r
o
v
in
g
m
o
d
el
p
er
f
o
r
m
an
ce
wh
ile
m
ain
tain
in
g
d
ata
in
teg
r
ity
.
I
n
ter
esti
n
g
ly
,
th
e
Un
b
ala
n
ce
d
d
ata
s
ce
n
ar
io
alr
ea
d
y
y
iel
d
s
h
ig
h
ac
cu
r
ac
y
ac
r
o
s
s
all
m
o
d
els,
s
u
g
g
esti
n
g
m
in
im
al
im
p
ac
t o
n
m
o
d
el
p
er
f
o
r
m
an
ce
in
th
is
s
p
ec
if
ic
s
ce
n
ar
io
.
Fig
u
r
e
1
6
.
Gen
er
ato
r
an
d
d
is
cr
im
in
ato
r
lo
s
s
o
f
AR
1
Fig
u
r
e
1
7
.
Acc
u
r
ac
y
co
m
p
a
r
is
o
n
o
f
AR
1
Ho
wev
er
,
th
e
o
u
tc
o
m
es
ar
e
m
o
r
e
v
a
r
ied
with
SMOT
E
b
ala
n
ce
d
d
ata
,
with
R
F
an
d
KNN
ac
h
iev
in
g
h
ig
h
er
ac
c
u
r
ac
y
,
wh
ile
LR
an
d
GB
ex
h
ib
it
lo
wer
ac
c
u
r
ac
y
co
m
p
a
r
ed
to
o
th
er
tec
h
n
iq
u
es.
Mo
r
e
o
v
er
,
Nea
r
Miss
b
alan
ce
d
d
ata
s
h
o
ws
b
o
th
p
o
s
itiv
e
an
d
n
eg
ativ
e
im
p
ac
ts
,
en
h
an
cin
g
ac
c
u
r
ac
y
f
o
r
s
o
m
e
m
o
d
els
wh
ile
s
ig
n
if
ican
tly
r
ed
u
cin
g
it
f
o
r
o
th
e
r
s
.
I
n
s
u
m
m
ar
y
,
w
h
ile
GAN
b
alan
cin
g
c
o
n
s
is
ten
tly
b
o
o
s
ts
m
o
d
el
ac
cu
r
ac
y
f
o
r
th
e
AR
1
d
ataset,
th
e
ef
f
ec
tiv
en
ess
o
f
SMOT
E
an
d
Nea
r
Miss
b
alan
cin
g
tech
n
iq
u
es
v
ar
ies
ac
r
o
s
s
m
o
d
els.
T
h
ese
f
in
d
in
g
s
o
f
f
er
v
alu
ab
le
in
s
ig
h
ts
f
o
r
s
elec
tin
g
s
u
itab
le
d
ata
b
alan
cin
g
m
eth
o
d
s
f
o
r
class
if
icatio
n
task
s
o
n
th
e
AR
1
d
ataset.
I
n
th
e
a
n
aly
s
is
f
o
r
th
e
AR
1
d
at
aset,
th
e
g
e
n
er
ato
r
lo
s
s
r
ea
ch
ed
its
m
in
im
u
m
at
ep
o
ch
4
0
7
8
,
in
d
icatin
g
o
p
tim
a
l
p
er
f
o
r
m
an
ce
f
o
r
s
y
n
th
etic
d
a
ta
g
en
er
atio
n
u
s
in
g
th
e
GAN
m
o
d
el.
Me
an
wh
ile,
th
e
d
is
cr
im
in
ato
r
lo
s
s
also
d
e
cr
ea
s
es,
in
d
icatin
g
im
p
r
o
v
ed
d
is
cr
im
in
atio
n
b
etwe
en
r
ea
l
a
n
d
s
y
n
th
etic
d
ata.
T
h
is
m
iles
to
n
e
s
u
g
g
ests
th
at
t
h
e
g
en
er
ato
r
h
as
co
n
v
er
g
ed
,
p
r
o
d
u
cin
g
th
e
h
ig
h
est
q
u
ality
s
y
n
th
etic
d
ata
am
o
n
g
th
e
1
0
,
0
0
0
e
p
o
ch
s
tr
ain
e
d
.
T
ab
le
1
p
r
esen
ts
a
s
u
m
m
ar
y
o
f
th
e
o
v
e
r
all
r
esu
lts
o
b
tain
ed
b
y
d
e
v
elo
p
in
g
1
2
8
m
o
d
els (
8
d
atasets
*
4
d
ata
b
alan
cin
g
ap
p
r
o
ac
h
es*
4
ML
m
o
d
e
ls
)
.
T
h
e
r
esear
ch
f
in
d
in
g
s
o
n
s
y
n
th
etic
d
ata
g
en
er
atio
n
an
d
d
e
f
ec
t
p
r
ed
ictio
n
in
s
o
f
twar
e
en
g
in
ee
r
in
g
h
av
e
s
ev
er
al
p
o
ten
tial
ap
p
li
ca
tio
n
s
in
r
ea
l
-
wo
r
ld
s
ce
n
ar
io
s
lik
e
s
o
f
twar
e
q
u
ality
ass
u
r
an
ce
,
r
eso
u
r
ce
allo
ca
tio
n
,
r
is
k
m
an
ag
em
en
t,
au
to
m
ated
b
u
g
d
etec
tio
n
,
s
o
f
twar
e
m
ain
ten
an
ce
an
d
ev
o
lu
tio
n
,
co
n
tin
u
o
u
s
Evaluation Warning : The document was created with Spire.PDF for Python.