T
E
L
K
O
M
NIKA
T
elec
o
mm
un
ica
t
io
n,
Co
m
pu
t
ing
,
E
lect
ro
nics
a
nd
Co
ntr
o
l
Vo
l.
19
,
No
.
4
,
A
u
g
u
s
t
2021
,
p
p
.
1
4
0
7
~
1
4
1
8
I
SS
N:
1
6
9
3
-
6
9
3
0
,
ac
cr
ed
ited
First Gr
ad
e
b
y
Ke
m
e
n
r
is
te
k
d
i
k
ti,
Dec
r
ee
No
: 2
1
/E/KPT
/2
0
1
8
DOI
:
1
0
.
1
2
9
2
8
/
T
E
L
KOM
NI
KA
.
v
1
9
i4
.
1
8
7
7
2
1407
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//jo
u
r
n
a
l.u
a
d
.
a
c.
id
/in
d
ex
.
p
h
p
/TELK
OM
N
I
K
A
Deep
lea
rni
ng
w
it
h f
o
ca
l los
s a
ppro
a
ch f
o
r attac
k
s
cl
a
ss
ificatio
n
Yesi No
v
a
ria
K
un
a
ng
1
,
Siti
Nur
m
a
ini
2
,
Der
i
s
Stia
w
a
n
3
,
B
ha
k
t
i Y
u
dh
o
Su
pra
pto
4
1
Do
c
to
ra
l
E
n
g
in
e
e
rin
g
De
p
a
rtem
e
n
t
,
F
a
c
u
lt
y
o
f
En
g
in
e
e
rin
g
,
Un
iv
e
rsitas
S
riw
ij
a
y
a
,
In
d
o
n
e
sia
1,
2
In
telli
g
e
n
t
S
y
ste
m
Re
se
a
r
c
h
G
r
o
u
p
,
F
a
c
u
l
t
y
o
f
Co
m
p
u
ter S
c
ien
c
e
,
Un
iv
e
rsitas
S
riw
ij
a
y
a
,
In
d
o
n
e
si
a
3
Co
m
p
u
ter Ne
tw
o
rk
in
g
&
In
f
o
r
m
a
ti
o
n
S
y
ste
m
s,
F
a
c
u
lt
y
o
f
Co
m
p
u
ter S
c
ien
c
e
,
Un
iv
e
rsitas
S
riw
ij
a
y
a
,
I
n
d
o
n
e
sia
1
Co
m
p
u
ter S
c
ien
c
e
De
p
a
rtem
e
n
t,
Un
iv
e
rsitas
Bin
a
Da
r
m
a
,
In
d
o
n
e
sia
4
El
e
c
tri
c
a
l
En
g
in
e
e
rin
g
De
p
a
rtem
e
n
t,
Un
iv
e
rsitas
S
rw
ij
a
y
a
,
In
d
o
n
e
sia
Art
icle
I
nfo
AB
ST
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
u
n
20
,
2
0
2
0
R
ev
i
s
ed
J
u
n
7
,
2021
A
cc
ep
ted
J
u
n
17
,
2021
T
h
e
ra
p
id
d
e
v
e
lo
p
m
e
n
t
o
f
d
e
e
p
lea
rn
in
g
im
p
ro
v
e
s
th
e
d
e
te
c
ti
o
n
a
n
d
c
las
si
f
ica
ti
o
n
o
f
a
tt
a
c
k
s
o
n
in
tru
sio
n
d
e
tec
ti
o
n
sy
ste
m
s.
Ho
w
e
v
e
r,
th
e
u
n
b
a
la
n
c
e
d
d
a
ta
issu
e
in
c
re
a
se
s
th
e
c
o
m
p
lex
it
y
o
f
th
e
a
rc
h
it
e
c
tu
re
m
o
d
e
l.
T
h
is
stu
d
y
p
ro
p
o
se
s
a
n
o
v
e
l
d
e
e
p
le
a
rn
in
g
m
o
d
e
l
to
o
v
e
rc
o
m
e
th
e
p
ro
b
lem
o
f
c
las
si
fy
in
g
m
u
lt
i
-
c
las
s
a
tt
a
c
k
s.
T
h
e
d
e
e
p
lea
rn
i
n
g
m
o
d
e
l
c
o
n
sists
o
f
t
w
o
sta
g
e
s.
T
h
e
p
re
-
tu
n
in
g
sta
g
e
u
se
s
a
u
to
m
a
ti
c
fe
a
tu
re
e
x
trac
ti
o
n
w
it
h
a
d
e
e
p
a
u
to
e
n
c
o
d
e
r.
T
h
e
se
c
o
n
d
sta
g
e
is
f
in
e
-
tu
n
i
n
g
u
si
n
g
d
e
e
p
n
e
u
ra
l
n
e
tw
o
rk
c
las
si
f
ier
s
w
it
h
f
u
ll
y
c
o
n
n
e
c
ted
l
a
y
e
rs.
T
o
re
d
u
c
e
im
b
a
lan
c
e
d
c
las
s
d
a
ta,
t
h
e
f
e
a
tu
re
e
x
trac
ti
o
n
w
a
s
im
p
le
m
e
n
ted
u
si
n
g
th
e
d
e
e
p
a
u
to
e
n
c
o
d
e
r
a
n
d
im
p
ro
v
e
d
f
o
c
a
l
lo
ss
f
u
n
c
ti
o
n
in
t
h
e
c
las
sifier.
T
h
e
m
o
d
e
l
w
a
s
e
v
a
lu
a
ted
u
sin
g
3
lo
ss
f
u
n
c
ti
o
n
s,
in
c
lu
d
i
n
g
c
ro
ss
-
e
n
tro
p
y
,
w
e
i
g
h
ted
c
ro
ss
-
e
n
tro
p
y
,
a
n
d
f
o
c
a
l
lo
ss
e
s.
T
h
e
re
su
lt
s
c
o
u
ld
c
o
rre
c
t
th
e
c
las
s
i
m
b
a
lan
c
e
in
d
e
e
p
lea
rn
in
g
-
b
a
se
d
c
las
si
f
ica
ti
o
n
s.
A
tt
a
c
k
c
las
sif
i
c
a
ti
o
n
w
a
s
a
c
h
iev
e
d
u
sin
g
a
u
t
o
m
a
ti
c
e
x
trac
ti
o
n
w
it
h
th
e
f
o
c
a
l
lo
ss
o
n
t
h
e
CS
E
-
CI
C
-
IDS2
0
1
8
d
a
tas
e
t
is
a
h
ig
h
-
q
u
a
li
t
y
c
la
ss
i
f
ier
w
it
h
9
8
.
3
8
%
p
re
c
isio
n
,
9
8
.
2
7
%
se
n
siti
v
it
y
,
a
n
d
9
9
.
8
2
%
sp
e
c
if
icit
y
.
K
ey
w
o
r
d
s
:
A
ttac
k
cla
s
s
i
f
icatio
n
Fo
ca
l
lo
s
s
I
m
b
alan
ce
d
d
ata
I
n
tr
u
s
io
n
d
etec
tio
n
s
y
s
te
m
Mu
lti
-
cla
s
s
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
:
Sit
i N
u
r
m
ai
n
i
I
n
telli
g
en
t S
y
s
te
m
R
e
s
ea
r
ch
G
r
o
u
p
,
Facu
lt
y
o
f
C
o
m
p
u
ter
Sci
en
ce
Dep
ar
t
m
e
n
t
Un
i
v
er
s
ita
s
Sri
w
ij
a
y
a
Fak
u
ltas
I
l
m
u
Ko
m
p
u
ter
,
Srij
a
y
a
Neg
ar
a
B
u
k
it B
esar
St.
P
ale
m
b
an
g
3
0
1
3
9
E
m
ail: s
iti
n
u
r
m
ai
n
i
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
in
cr
ea
s
in
g
u
s
e
o
f
th
e
i
n
ter
n
et
o
f
th
i
n
g
s
tech
n
o
lo
g
y
h
as si
g
n
i
f
ica
n
tl
y
in
cr
ea
s
ed
th
e
a
m
o
u
n
t o
f
d
at
a
g
en
er
ated
d
ail
y
.
A
p
ar
t
f
r
o
m
p
r
o
v
id
in
g
t
h
e
b
en
ef
i
ts
an
d
f
ac
il
ities
th
at
i
m
p
r
o
v
e
th
e
q
u
alit
y
o
f
lif
e,
b
ig
d
ata
an
d
in
ter
n
e
t
o
f
t
h
i
n
g
s
(
I
o
T
)
p
o
s
e
t
h
e
ch
alle
n
g
e
o
f
v
u
l
n
er
ab
ilit
y
a
n
d
s
ec
u
r
it
y
is
s
u
e
s
[
1
]
-
[
4
]
.
Vu
ln
er
ab
ilit
y
i
n
cr
ea
s
ed
d
u
e
to
th
e
lar
g
e
n
u
m
b
er
o
f
p
h
y
s
ical
i
n
f
r
as
tr
u
ct
u
r
es
co
n
n
ec
ted
to
th
e
n
et
w
o
r
k
[
5
]
,
[
6
]
.
T
o
ad
d
r
ess
th
i
s
is
s
u
e,
t
h
e
n
et
w
o
r
k
i
n
tr
u
s
io
n
d
etec
tio
n
s
y
s
te
m
(
NI
DS)
i
s
an
ap
p
r
o
p
r
iate
alter
n
ati
v
e
f
o
r
s
ec
u
r
i
n
g
m
o
d
e
r
n
n
et
w
o
r
k
s
[
7
]
.
T
h
e
s
ec
u
r
it
y
s
o
lu
tio
n
s
o
f
NI
D
S
-
b
ased
I
o
T
h
av
e
b
ee
n
p
r
o
p
o
s
ed
,
s
u
ch
as
m
ac
h
in
e
an
d
d
ee
p
le
ar
n
in
g
-
b
ased
m
et
h
o
d
s
[
7
]
-
[
1
0
]
.
d
ee
p
lear
n
in
g
(
DL
)
h
a
s
i
m
p
r
es
s
iv
el
y
p
er
f
o
r
m
ed
in
b
ig
d
ata
d
o
m
a
in
s
[
1
1
]
,
in
clu
d
in
g
co
m
p
u
ter
v
is
io
n
[
1
2
]
,
[
1
2
]
s
p
ee
ch
r
ec
o
g
n
itio
n
[
1
3
]
,
an
d
m
e
d
ical
i
m
a
g
i
n
g
[
1
4
]
-
[
1
6
]
.
A
l
s
o
,
s
o
m
e
s
t
u
d
ies
h
av
e
u
s
ed
th
e
DL
al
g
o
r
it
h
m
o
n
th
e
in
tr
u
s
io
n
d
etec
tio
n
s
y
s
te
m
,
s
u
ch
as
d
ee
p
B
o
ltzm
a
n
n
m
ac
h
i
n
es
[
1
7
]
,
d
ee
p
b
elief
n
et
w
o
r
k
s
[
1
8
]
,
an
d
d
ee
p
n
eu
r
al
n
et
w
o
r
k
s
[
1
9
]
,
[
2
0
]
.
T
h
e
DL
tec
h
n
iq
u
e
o
r
its
co
m
b
i
n
atio
n
i
m
p
r
o
v
es
th
e
ac
cu
r
ac
y
an
d
d
etec
tio
n
s
p
ee
d
[
2
1
]
.
Fu
r
th
er
m
o
r
e,
it
ef
f
icie
n
tl
y
d
etec
t
s
attac
k
v
ar
ian
ts
an
d
p
atter
n
s
[
2
2
]
.
T
h
er
e
ar
e
s
ev
er
al
ch
allen
g
es
to
NI
DS
in
ef
f
icien
tl
y
an
d
ef
f
ec
ti
v
el
y
d
etec
ti
n
g
t
h
e
n
et
w
o
r
k
’
s
ab
n
o
r
m
al
b
eh
a
v
io
r
.
I
o
T
ap
p
licatio
n
s
'
b
ig
d
ata
n
at
u
r
e
p
r
esen
t
s
d
if
f
ic
u
ltie
s
w
it
h
th
e
a
m
o
u
n
t
a
n
d
co
m
p
le
x
it
y
o
f
d
ata
[
1
]
.
Ma
n
y
ap
p
licatio
n
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1693
-
6930
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
,
Vo
l.
19
,
No
.
4
,
A
u
g
u
s
t 2
0
2
1
:
1
4
0
7
-
1
4
1
8
1408
en
cr
y
p
t
I
n
ter
n
et
tr
af
f
ic
u
s
i
n
g
s
ec
u
r
it
y
p
r
o
to
co
ls
.
T
h
e
en
cr
y
p
tio
n
en
co
u
r
a
g
es
m
o
r
e
d
iv
er
s
e
an
d
s
o
p
h
is
ticated
attac
k
s
t
h
at
n
ee
d
to
b
e
d
etec
te
d
.
Seco
n
d
,
in
r
ea
l
n
e
t
w
o
r
k
tr
a
f
f
ics,
d
ata
d
is
tr
ib
u
t
io
n
i
s
u
s
u
all
y
i
m
b
alan
ce
d
[
2
3
]
.
I
n
th
i
s
ca
s
e,
th
e
n
u
m
b
er
o
f
s
a
m
p
les
b
elo
n
g
i
n
g
to
t
h
e
r
eg
u
la
r
tr
af
f
ic
i
s
m
u
c
h
h
i
g
h
er
th
a
n
th
e
attac
k
s
a
m
p
le.
T
h
er
ef
o
r
e,
th
e
class
i
f
icatio
n
r
e
s
u
lt
s
lean
to
w
ar
d
s
t
h
e
b
en
i
g
n
d
ata
[
2
4
]
.
Mo
s
t o
f
t
h
e
d
atase
ts
u
s
ed
to
a
n
al
y
ze
NI
DS,
s
u
c
h
a
s
n
e
t
w
o
r
k
s
ec
u
r
it
y
la
y
er
-
k
n
o
w
l
ed
g
e
d
i
s
co
v
er
y
in
d
atab
ase
(
NSL
-
KDD
)
[
2
5
]
,
th
e
co
m
m
u
n
icatio
n
s
s
ec
u
r
it
y
e
s
tab
lis
h
m
e
n
t
a
n
d
th
e
C
a
n
a
d
ian
in
s
tit
u
te
f
o
r
c
y
b
er
s
ec
u
r
it
y
i
n
tr
u
s
io
n
d
etec
ti
o
n
s
y
s
te
m
2018
d
ataset
(
C
SE
-
C
I
C
-
I
DS2
0
1
8
)
[
2
6
]
,
h
av
e
a
s
e
v
er
e
clas
s
i
m
b
alan
ce
p
r
o
b
lem
.
Sev
er
al
ap
p
r
o
ac
h
es
h
av
e
b
ee
n
e
m
p
lo
y
ed
to
s
o
lv
e
t
h
i
s
p
r
o
b
lem
,
s
u
ch
a
s
o
v
er
-
s
a
m
p
l
i
n
g
[
2
3
]
,
[
2
7
]
,
[
2
8
]
,
u
n
d
er
-
s
a
m
p
lin
g
[
2
9
]
,
s
p
r
ea
d
s
u
b
-
s
a
m
p
le
[
3
0
]
,
an
d
class
b
ala
n
ce
r
[
2
3
]
.
Ho
w
e
v
er
,
th
e
f
u
n
d
am
en
tal
p
r
o
b
lem
o
f
class
i
m
b
ala
n
ce
i
n
attac
k
cla
s
s
es
r
e
m
ain
s
an
i
n
ter
esti
n
g
is
s
u
e
to
s
tu
d
y
.
A
l
s
o
,
d
ata
q
u
alit
y
is
s
u
es
tr
ig
g
er
i
m
b
alan
ce
d
d
ata
p
r
o
b
lem
s
[
3
1
]
,
[
3
2
]
.
T
h
er
ef
o
r
e,
o
th
er
s
tr
ate
g
ies
ar
e
n
ee
d
ed
to
s
o
lv
e
th
is
p
r
o
b
lem
,
esp
ec
iall
y
f
o
r
m
u
lti
-
cla
s
s
ca
s
es.
T
h
er
e
h
a
v
e
b
ee
n
d
e
v
elo
p
m
e
n
ts
i
n
t
h
e
f
i
eld
s
o
f
f
o
ca
l
lo
s
s
in
i
m
a
g
e
r
ec
o
g
n
i
tio
n
,
b
io
m
ed
ical
s
cien
ce
s
,
an
d
s
tab
ilit
y
tr
ain
i
n
g
[
3
3
]
-
[
3
5
]
.
Usi
n
g
th
i
s
k
n
o
w
le
d
g
e,
th
e
i
m
p
r
o
v
ed
f
o
ca
l
lo
s
s
f
u
n
ct
io
n
f
o
r
a
m
u
lti
-
class
m
o
d
el
is
u
s
ed
to
p
r
ev
en
t
class
i
m
b
ala
n
ce
an
d
o
v
er
-
f
itti
n
g
a
ttack
c
lass
if
ica
tio
n
.
T
h
is
r
esear
ch
f
o
cu
s
e
s
o
n
ef
f
icien
t
tr
ain
i
n
g
o
n
all
d
ata
s
ets,
b
ased
o
n
th
e
ex
t
r
em
e
clas
s
i
m
b
alan
ce
.
Mo
r
eo
v
er
,
th
e
tr
ain
i
n
g
is
b
ased
o
n
m
u
lti
-
clas
s
attac
k
s
b
y
u
tili
zi
n
g
th
e
f
o
ca
l lo
s
s
f
u
n
c
tio
n
u
s
ed
in
th
e
d
ee
p
lear
n
in
g
m
o
d
els.
T
h
is
s
tu
d
y
p
r
o
p
o
s
es
a
m
u
lti
-
cl
ass
f
o
ca
l
lo
s
s
f
u
n
ctio
n
o
f
d
ee
p
lear
n
in
g
to
ad
d
r
ess
u
n
b
alan
ce
d
d
ata.
T
h
e
r
esu
l
t
is
co
m
p
ar
ed
w
it
h
th
e
cr
o
s
s
-
e
n
tr
o
p
y
(
C
E
)
lo
s
s
an
d
w
ei
g
h
ted
cr
o
s
s
-
en
tr
o
p
y
f
u
n
ctio
n
s
.
A
s
a
co
n
tr
ib
u
tio
n
,
th
is
s
t
u
d
y
p
r
o
p
o
s
es
th
e
d
ee
p
a
u
to
-
e
n
co
d
er
(
DA
E
)
,
co
m
b
i
n
ed
w
i
th
d
ee
p
n
e
u
r
al
n
et
w
o
r
k
(
D
NN)
m
o
d
el
u
s
in
g
a
m
u
lti
-
clas
s
f
o
ca
l
lo
s
s
f
u
n
cti
o
n
.
T
h
is
is
ai
m
ed
to
ad
d
r
es
s
th
e
d
i
f
f
er
e
n
t
clas
s
i
m
b
ala
n
ce
f
o
r
th
e
at
tac
k
class
i
f
icatio
n
.
E
x
p
er
i
m
e
n
tal
r
esu
lt
s
s
h
o
w
t
h
at
th
e
p
r
e
-
tr
ai
n
i
n
g
s
tag
e,
d
ee
p
au
to
-
en
co
d
er
,
h
as
ad
v
an
ta
g
es
i
n
m
o
r
e
co
m
p
lica
ted
f
ea
tu
r
e
s
lear
n
ed
f
r
o
m
t
h
e
o
r
ig
in
al
d
ata.
Fo
ca
l
lo
s
s
f
u
n
ctio
n
,
s
ca
led
f
r
o
m
cr
o
ss
-
e
n
tr
o
p
y
lo
s
s
,
is
a
m
o
r
e
ef
f
ec
t
iv
e
alter
n
a
tiv
e
to
p
r
ev
io
u
s
ap
p
r
o
ac
h
es
in
d
ea
lin
g
w
ith
t
h
e
class
i
m
b
ala
n
ce
in
m
u
lti
-
cla
s
s
attac
k
class
i
f
icatio
n
.
2.
RE
L
AT
E
D
WO
RK
Net
w
o
r
k
I
n
tr
u
s
io
n
Dete
c
tio
n
S
y
s
te
m
h
as
b
ee
n
s
t
u
d
ied
w
id
el
y
o
v
er
th
e
p
ast
s
ev
er
al
y
ea
r
s
.
T
h
is
s
ec
tio
n
b
r
ief
l
y
d
i
s
cu
s
s
e
s
s
o
m
e
p
u
b
li
s
h
ed
ap
p
r
o
ac
h
es
to
d
ee
p
lear
n
in
g
m
et
h
o
d
s
,
in
p
ar
tic
u
lar
to
i
m
b
alan
ce
d
d
ataset
s
.
I
n
2
0
1
9
,
L
in
et
a
l
.
[
2
6
]
u
s
ed
d
e
ep
lear
n
in
g
f
o
r
d
y
n
a
m
ic
n
et
w
o
r
k
an
o
m
al
y
d
etec
tio
n
.
T
h
e
s
y
n
t
h
etic
m
i
n
o
r
it
y
o
v
er
s
a
m
p
li
n
g
tech
n
iq
u
e
(
SM
OT
E
)
alg
o
r
ith
m
w
as
e
x
p
er
im
en
tall
y
ap
p
lied
to
h
an
d
le
th
e
i
m
b
ala
n
ce
d
class
p
r
o
b
lem
in
th
e
C
SE
-
C
I
C
I
DS2
0
1
8
d
ataset.
A
s
a
clas
s
if
ier
,
a
d
ee
p
n
eu
r
al
n
et
w
o
r
k
m
o
d
el
w
a
s
u
s
ed
w
ith
lo
n
g
s
h
o
r
t
-
ter
m
m
e
m
o
r
y
(
L
ST
M)
b
ased
,
co
m
b
in
ed
w
it
h
an
atte
n
tio
n
m
ec
h
a
n
i
s
m
(
A
M)
,
to
en
h
an
ce
p
er
f
o
r
m
a
n
ce
.
T
h
e
SMOT
E
alg
o
r
ith
m
ap
p
lie
d
to
p
r
o
m
o
te
th
e
p
r
o
p
o
r
tio
n
o
f
m
i
n
o
r
it
y
clas
s
o
p
ti
m
izes
t
h
e
d
ee
p
lear
n
in
g
m
o
d
el.
T
h
e
m
o
d
el
ac
h
iev
ed
t
h
e
b
est
r
e
s
u
lt
s
,
w
it
h
a
n
ac
cu
r
ac
y
o
f
9
6
.
2
%,
an
d
th
e
r
ec
al
l
r
ate
r
ea
ch
ed
9
8
%
f
o
r
6
ca
teg
o
r
ies
class
.
Mo
r
e
r
ec
en
tl
y
,
Z
h
a
n
g
et
a
l
.
[
2
7
]
in
tr
o
d
u
ce
d
a
h
y
b
r
id
SMOT
E
th
at
co
m
b
i
n
es
SM
OT
E
an
d
Gau
s
s
ia
n
m
i
x
tu
r
e
m
o
d
el
(
GM
M)
b
ased
clu
s
ter
i
n
g
to
i
m
p
r
o
v
e
th
e
m
i
n
o
r
ity
cla
s
s
'
s
d
etec
tio
n
r
ate.
T
h
e
s
y
n
th
et
ic
m
i
n
o
r
it
y
o
v
er
-
s
a
m
p
li
n
g
tec
h
n
iq
u
e
(
S
MO
T
E
)
an
d
g
au
s
s
ian
m
i
x
t
u
r
e
(
SGM
)
p
r
o
ce
s
s
in
g
w
a
s
in
teg
r
ated
w
it
h
a
co
n
v
o
lu
tio
n
al
n
e
u
r
al
n
e
t
w
o
r
k
(
C
NN)
f
o
r
b
i
n
ar
y
a
n
d
m
u
lt
i
-
class
cla
s
s
i
f
icatio
n
.
T
h
e
y
clai
m
ed
th
a
t
t
h
e
SG
M
m
o
d
el
i
n
cr
ea
s
e
s
d
etec
tio
n
an
d
r
ed
u
ce
s
th
e
ti
m
e
co
s
t.
T
h
e
p
r
o
p
o
s
ed
m
et
h
o
d
w
a
s
ev
al
u
ated
w
it
h
5
class
e
s
i
m
b
alan
ce
d
tech
n
iq
u
e
an
d
2
c
la
s
s
i
f
icatio
n
alg
o
r
ith
m
s
.
T
h
e
y
w
er
e
v
er
i
f
ied
u
s
i
n
g
t
h
e
Un
i
v
e
r
s
it
y
o
f
Ne
w
So
u
t
h
W
ales
-
NB
2
0
1
5
(
UNSW
-
NB
1
5
)
an
d
th
e
C
a
n
ad
ian
in
s
tit
u
te
f
o
r
c
y
b
er
s
ec
u
r
it
y
in
tr
u
s
io
n
d
ete
ctio
n
s
y
s
te
m
2
0
1
7
(
C
I
C
I
DS2
0
1
7
)
d
atasets
.
T
h
e
ev
alu
atio
n
o
f
t
h
e
C
I
C
I
DS2
0
1
7
d
ataset
s
h
o
w
s
th
at
th
e
m
et
h
o
d
ac
h
iev
es
a
n
ex
ce
lle
n
t
d
etec
tio
n
r
ate
o
f
9
9
.
8
5
%
in
t
h
e
1
5
-
cla
s
s
c
lass
if
ica
tio
n
.
Ho
w
ev
er
,
t
h
e
d
etec
tio
n
r
ates
f
o
r
w
eb
at
tack
b
r
u
te
f
o
r
ce
ar
e
s
till
le
s
s
th
a
n
5
0
%,
lo
w
er
t
h
an
r
an
d
o
m
o
v
er
s
a
m
p
li
n
g
(R
OS
)
an
d
S
MO
T
E
.
A
s
f
o
r
th
e
UNSW
-
NB
1
5
d
ataset,
th
e
d
et
ec
tio
n
r
ates f
o
r
b
in
ar
y
a
n
d
1
0
class
i
f
icatio
n
s
r
ea
ch
9
9
.
7
4
% a
n
d
9
6
.
5
4
%.
A
b
d
u
l
h
a
m
m
ed
et
a
l.
[
2
3
]
u
s
e
d
v
ar
io
u
s
tec
h
n
iq
u
e
s
,
s
u
ch
as
o
v
er
-
s
a
m
p
li
n
g
,
u
n
d
er
-
s
a
m
p
li
n
g
,
s
p
r
ea
d
s
u
b
s
a
m
p
le,
a
n
d
class
b
alan
ce
r
,
to
s
o
lv
e
i
m
b
ala
n
ce
d
d
ata
p
r
o
b
le
m
s
f
o
r
b
in
ar
y
clas
s
es.
Sev
e
r
al
class
i
f
ier
s
,
s
u
ch
as r
an
d
o
m
f
o
r
e
s
t (
R
F),
DNN,
v
o
tin
g
,
v
ar
iat
io
n
al
a
u
to
-
e
n
co
d
er
,
ar
e
u
s
ed
in
th
e
e
v
al
u
atio
n
.
T
h
e
ex
p
er
i
m
en
t
s
o
n
th
e
C
I
DD
S
-
0
0
1
d
ataset
s
h
o
wed
th
at
DNN
w
it
h
t
h
e
d
o
w
n
-
s
a
m
p
li
n
g
m
et
h
o
d
an
d
clas
s
b
alan
ce
r
is
t
h
e
m
o
s
t
ef
f
ec
tiv
e.
B
y
th
e
ex
p
er
i
m
en
ta
l
r
esu
lt
s
,
t
h
e
clas
s
d
is
tr
ib
u
t
io
n
h
as
a
l
ig
h
t
i
m
p
ac
t
o
n
t
h
e
cla
s
s
i
f
icatio
n
p
r
o
ce
s
s
.
Fu
r
t
h
er
m
o
r
e,
A
b
d
u
l
h
a
m
m
ed
et
a
l
.
[
3
6
]
p
r
o
p
o
s
ed
th
e
u
n
i
f
o
r
m
d
is
tr
ib
u
t
io
n
b
ased
b
ala
n
cin
g
(
UDB
B
)
f
o
r
i
m
b
alan
ce
d
cla
s
s
e
s
.
T
o
r
ed
u
ce
f
ea
t
u
r
es,
t
h
e
a
u
to
-
e
n
co
d
er
(
AE
)
an
d
p
r
in
cip
le
co
m
p
o
n
e
n
t
a
n
al
y
s
i
s
(P
C
A
)
w
er
e
u
s
ed
in
e
v
alu
a
tio
n
w
it
h
v
ar
io
u
s
clas
s
i
f
ier
m
et
h
o
d
s
.
T
h
e
s
im
u
latio
n
r
es
u
lt
s
o
n
th
e
o
r
ig
i
n
al
d
is
tr
ib
u
tio
n
o
f
t
h
e
C
I
C
I
DS2
0
1
7
d
ataset
s
h
o
w
e
d
th
at
P
C
A
p
r
o
d
u
ce
s
b
ette
r
ac
cu
r
ac
y
th
a
n
A
E
,
at
9
9
.
6
%.
Ho
w
e
v
er
,
b
y
i
m
p
le
m
en
t
in
g
UD
B
B
,
th
e
d
etec
tio
n
ac
cu
r
ac
y
w
as
r
ed
u
ce
d
to
9
8
.
9
%,
alth
o
u
g
h
it
b
etter
d
et
ec
ted
s
o
m
e
attac
k
s
.
I
n
an
o
th
er
ex
p
er
i
m
en
t,
Hu
a
[
2
9
]
u
s
ed
u
n
d
er
-
s
a
m
p
li
n
g
a
n
d
f
ea
t
u
r
e
s
elec
tio
n
i
n
p
r
e
-
p
r
o
ce
s
s
in
g
.
T
h
e
p
r
o
p
o
s
ed
tr
af
f
ic
class
i
f
icatio
n
u
s
in
g
L
i
g
h
t
GB
M,
b
ased
o
n
th
e
C
SE
-
C
I
C
-
I
DS2
0
1
8
d
ataset.
T
h
e
m
o
d
el
u
s
ed
o
n
l
y
1
0
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
Dee
p
lea
r
n
in
g
w
ith
fo
ca
l lo
s
s
a
p
p
r
o
a
c
h
fo
r
a
tta
ck
s
cla
s
s
ifica
tio
n
(
Yes
i No
va
r
ia
K
u
n
a
n
g
)
1409
f
ea
t
u
r
es
t
h
a
t
w
er
e
s
elec
ted
u
s
i
n
g
R
a
n
d
o
m
Fo
r
est.
T
h
e
y
co
m
p
ar
ed
th
eir
m
o
d
els
w
it
h
v
ar
io
u
s
m
ac
h
in
e
lear
n
i
n
g
alg
o
r
ith
m
s
,
a
n
d
C
NN
d
ee
p
l
ea
r
n
in
g
.
T
h
e
b
est
r
es
u
lts
f
o
r
th
e
o
v
er
all
ac
cu
r
ac
y
o
b
tain
e
d
r
ea
ch
ed
9
8
.
3
7
%.
Ho
w
e
v
er
,
th
e
i
n
f
l
u
e
n
ce
o
f
t
h
e
m
o
d
el
o
n
t
h
e
m
i
n
o
r
it
y
clas
s
wa
s
n
o
t d
is
c
u
s
s
ed
.
Yan
g
et
a
l.
[
3
7
]
ap
p
lied
an
i
m
p
r
o
v
ed
co
n
d
itio
n
al
v
ar
iatio
n
al
au
to
en
co
d
er
w
it
h
a
d
ee
p
n
eu
r
al
n
et
w
o
r
k
in
NI
DS.
A
n
i
m
p
r
o
v
ed
co
n
d
itio
n
al
v
ar
iat
io
n
al
Au
to
E
n
co
d
er
(
I
C
VA
E
)
tr
ain
i
n
g
e
x
p
lo
r
es
th
e
r
elatio
n
s
h
ip
b
et
w
ee
n
d
ata
f
ea
t
u
r
es
a
n
d
att
ac
k
clas
s
es.
T
h
is
ai
m
s
at
b
al
an
cin
g
tr
ai
n
i
n
g
d
ata
s
e
ts
a
n
d
i
m
p
r
o
v
e
d
etec
tio
n
p
er
f
o
r
m
a
n
ce
i
n
m
i
n
o
r
it
y
attac
k
.
T
h
ey
u
s
ed
cr
o
s
s
-
en
tr
o
p
y
as
th
e
f
u
n
ctio
n
o
f
r
ec
o
n
s
tr
u
ct
io
n
lo
s
s
o
f
th
e
d
ec
o
d
er
.
T
h
e
r
esu
lts
o
f
t
h
i
s
ch
a
llen
g
e
s
h
o
w
ed
t
h
at
t
h
e
b
est
i
n
d
i
v
id
u
al
d
etec
tio
n
s
y
s
te
m
o
b
tain
s
u
p
to
8
9
.
0
8
%
an
d
8
5
.
9
7
%
o
f
th
e
m
u
l
ti
-
c
lass
clas
s
i
f
icatio
n
i
n
t
h
e
UN
SW
B
1
5
an
d
NS
L
-
KDD
d
atasets
,
r
e
s
p
ec
tiv
el
y
.
T
h
ey
c
lai
m
ed
th
at
I
C
V
A
E
-
DNN
in
cr
ea
s
es
d
etec
t
io
n
r
ates
o
f
m
i
n
o
r
it
y
a
n
d
u
n
k
n
o
w
n
attac
k
s
.
A
l
s
o
,
an
u
n
s
u
p
er
v
is
ed
au
to
-
en
co
d
er
w
a
s
u
s
ed
b
y
L
i
et
a
l
.
[
3
8
]
to
o
v
er
co
m
e
i
m
b
ala
n
ce
p
r
o
b
lem
s
in
NI
DS.
T
h
e
y
u
s
ed
th
e
r
an
d
o
m
f
o
r
est
to
s
elec
t
s
ig
n
i
f
ica
n
t
f
ea
t
u
r
es
in
t
h
e
C
SE
-
C
I
C
-
I
DS2
0
1
8
d
ataset,
an
d
p
er
f
o
r
m
ed
an
o
m
al
y
d
etec
ti
o
n
f
o
r
ea
ch
attac
k
.
Ho
w
e
v
er
,
th
e
r
esu
lts
o
f
A
E
-
I
DS
f
o
r
attac
k
s
,
in
w
eb
atta
ck
s
(
SQ
L
in
j
ec
tio
n
,
b
r
u
te
f
o
r
ce
w
eb
,
an
d
b
r
u
te
f
o
r
ce
-
XS
S),
ar
e
s
till
lo
w
an
d
o
p
tim
iz
ed
.
Si
m
i
lar
l
y
,
th
e
u
n
s
u
p
er
v
is
ed
au
to
-
en
co
d
er
m
o
d
el
w
a
s
u
s
ed
b
y
Z
h
ao
et
a
l.
[
3
9
]
T
h
ey
in
tr
o
d
u
ce
d
th
e
s
e
m
i
-
s
u
p
er
v
is
ed
d
is
cr
i
m
i
n
an
t
au
to
-
e
n
co
d
er
(
SS
DA
)
to
o
v
er
co
m
e
n
e
w
attac
k
s
.
I
n
s
p
ir
ed
b
y
ex
i
s
ti
n
g
r
e
s
ea
r
ch
,
th
i
s
s
t
u
d
y
u
s
es
D
A
E
to
ex
tr
ac
t
attac
k
d
ata.
Fu
r
th
er
m
o
r
e,
th
e
f
o
ca
l
lo
s
s
is
u
s
ed
to
in
cr
ea
s
e
th
e
d
etec
tio
n
r
ate
o
f
m
i
n
o
r
it
y
a
ttack
s
.
T
h
e
C
SE
-
C
I
C
-
I
DS2
0
1
8
d
ataset
is
u
s
ed
to
test
th
e
m
o
d
el
in
m
u
lt
i
-
c
lass
cla
s
s
i
f
icat
io
n
an
d
co
m
p
ar
e
th
e
i
m
p
ac
t o
f
t
h
e
th
r
ee
-
lo
s
s
f
u
n
ctio
n
s
o
n
u
n
b
alan
c
ed
p
r
o
ce
s
s
es.
3.
RE
S
E
ARCH
M
E
T
H
O
D
T
h
is
s
tu
d
y
i
m
p
r
o
v
es
i
n
tr
u
s
io
n
d
etec
tio
n
s
y
s
te
m
s
'
ab
ilit
y
to
d
etec
t
m
in
o
r
it
y
a
ttack
s
clas
s
u
s
i
n
g
d
ee
p
lear
n
in
g
m
o
d
els
w
i
th
d
ee
p
au
to
-
e
n
co
d
er
(
DA
E
)
p
r
e
-
tu
n
i
n
g
p
r
o
ce
s
s
es
a
n
d
f
i
n
e
-
tu
n
i
n
g
u
s
i
n
g
DN
N.
T
h
e
class
i
f
icatio
n
p
r
o
ce
s
s
u
s
ed
3
s
ce
n
ar
io
s
,
i
n
clu
d
i
n
g
ca
teg
o
r
ic
al
cr
o
s
s
-
e
n
tr
o
p
y
(
C
E
)
lo
s
s
,
f
o
ca
l
-
lo
s
s
(
FL
)
,
a
n
d
w
ei
g
h
ted
ca
teg
o
r
ical
cr
o
s
s
-
e
n
tr
o
p
y
(
W
C
E
)
,
as
ill
u
s
tr
ated
in
F
ig
u
r
e
1.
T
h
e
m
o
d
el
was
ev
al
u
ated
u
s
i
n
g
C
SE
-
C
I
C
-
I
D
S2
0
1
8
,
w
h
ic
h
r
ep
r
esen
ts
a
r
ec
en
t a
ttack
d
atase
t
[
4
0
]
.
Fig
u
r
e
1
.
Dee
p
lear
n
in
g
ar
c
h
it
ec
tu
r
e
-
b
ased
attac
k
s
cla
s
s
i
f
ica
tio
n
3
.
1
.
Da
t
a
s
et
T
h
e
C
SE
-
C
I
C
-
I
DS2
0
1
8
d
ataset
co
n
s
i
s
ts
o
f
8
0
f
ea
t
u
r
es,
in
c
l
u
d
in
g
lab
els.
T
h
e
f
ea
tu
r
es
o
f
t
h
e
d
ataset
ar
e
g
en
er
ated
an
d
ex
tr
ac
ted
w
it
h
C
I
C
F
lo
w
Me
ter
[
4
0
]
,
[
4
1
]
.
T
h
e
d
esig
n
ed
s
ce
n
ar
io
co
n
s
i
s
ts
o
f
6
attac
k
s
,
in
cl
u
d
in
g
d
en
ia
l
o
f
s
er
v
ices
(
Do
S),
d
is
tr
ib
u
ted
d
en
ial
o
f
s
er
v
i
ce
s
(
DDo
S),
b
o
tn
et
,
b
r
u
te
-
f
o
r
c
e
,
w
eb
attac
k
s
,
a
n
d
in
f
iltra
tio
n
,
as
p
r
esen
ted
in
T
ab
le
1
.
T
h
e
y
ar
e
g
r
o
u
p
ed
in
to
1
4
attac
k
s
u
b
-
cla
s
s
e
s
.
T
h
e
t
o
tal
d
ata
am
o
u
n
t
i
s
1
6
,
2
3
2
,
9
4
3
r
ec
o
r
d
s
,
d
o
m
i
n
ate
d
b
y
8
3
.
0
7
%
b
en
ig
n
tr
a
f
f
ic.
T
h
is
s
t
u
d
y
u
s
ed
1
0
%
d
ata
f
o
r
tr
ain
i
n
g
an
d
2
.
5
%
f
o
r
test
i
n
g
.
A
b
o
u
t
5
1
%
o
f
t
h
e
d
ata
w
as u
s
ed
f
o
r
b
en
ig
n
co
m
p
o
s
it
io
n
.
T
h
e
s
tr
u
ctu
r
e
o
f
m
alicio
u
s
d
ata
is
u
s
ed
b
ased
o
n
th
e
a
m
o
u
n
t
o
f
d
ata.
T
ab
le
1
clar
if
ies
t
h
e
co
m
p
o
s
itio
n
o
f
th
e
tr
ain
in
g
a
n
d
test
i
n
g
d
ata
u
s
ed
.
T
h
e
in
f
iltra
tio
n
a
ttack
in
cl
u
d
es
a
s
tealt
h
y
att
ac
k
th
at
u
tili
ze
s
a
n
in
ter
n
al
n
et
w
o
r
k
f
o
r
ille
g
al
ac
ce
s
s
.
T
h
e
ch
ar
ac
ter
is
tic
s
o
f
in
f
iltra
tio
n
tr
a
f
f
ic
an
d
b
en
ig
n
ar
e
v
er
y
clo
s
e,
w
h
ich
i
m
p
lie
s
a
d
if
f
ic
u
lt
y
in
d
etec
ti
n
g
t
h
e
n
e
t
w
o
r
k
I
DS
[
4
2
]
.
A
s
a
r
esu
lt,
th
e
i
n
f
iltra
tio
n
attac
k
w
as
eli
m
i
n
ated
in
t
h
e
ex
p
er
i
m
en
t
b
ec
au
s
e
t
h
is
s
tu
d
y
d
is
cu
s
s
ed
th
e
f
o
cu
s
,
e
m
p
h
a
s
izi
n
g
t
h
e
co
n
s
id
er
atio
n
o
f
i
m
b
ala
n
ce
d
class
f
ac
to
r
s
in
ac
cu
r
ac
y
d
etec
tio
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1693
-
6930
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
,
Vo
l.
19
,
No
.
4
,
A
u
g
u
s
t 2
0
2
1
:
1
4
0
7
-
1
4
1
8
1410
T
ab
le
1
.
C
o
m
p
o
s
itio
n
o
f
th
e
C
SE
-
C
I
C
-
I
DS2
0
1
8
d
ataset
u
s
ed
C
a
t
e
g
o
r
y
S
i
z
e
T
r
a
i
n
T
e
st
B
e
n
i
g
n
1
3
,
4
8
4
,
7
0
8
8
3
.
0
7
%
8
0
3
,
0
2
5
5
1
.
1
2
%
2
0
1
,
2
3
8
5
1
.
2
1
%
B
o
t
2
8
6
,
1
9
1
1
.
7
6
%
8
5
,
8
4
2
5
.
4
6
%
2
1
,
4
7
9
5
.
4
7
%
B
r
u
t
e
F
o
r
c
e
3
8
0
,
9
4
9
2
.
3
5
%
1
1
4
,
3
8
7
7
.
2
8
%
2
8
,
4
6
5
7
.
2
4
%
D
D
o
S
6
5
4
,
3
0
0
4
.
0
3
%
3
5
4
,
9
0
7
2
2
.
5
9
%
8
8
,
5
9
3
2
2
.
5
5
%
D
o
S
1
,
2
6
3
,
9
3
3
7
.
7
9
%
2
1
2
,
0
5
3
1
3
.
5
0
%
5
2
,
9
9
5
1
3
.
4
9
%
W
e
b
A
t
t
a
c
k
s
9
2
8
0
.
0
1
%
7
5
4
0
.
0
5
%
1
7
4
0
.
0
4
%
I
n
f
i
l
t
e
r
a
t
i
o
n
1
6
1
,
9
3
4
1
.
0
0
%
-
-
-
-
T
o
t
a
l
1
6
,
2
3
2
,
9
4
3
1
0
0
.
0
0
%
1
,
5
7
0
,
9
6
8
1
0
0
.
0
0
%
3
9
2
,
9
4
4
1
0
0
.
0
0
%
3
.
2
.
P
re
-
pro
ce
s
s
ing
Fro
m
8
0
f
ea
tu
r
es
o
f
t
h
e
C
SE
-
C
I
C
-
I
DS2
0
1
8
d
ataset
[
4
0
]
,
th
e
ti
m
esta
m
p
f
ea
t
u
r
e
w
as
eli
m
i
n
ated
,
an
d
o
n
l
y
7
9
w
er
e
u
s
ed
.
T
h
e
ti
m
e
s
t
a
m
p
is
e
n
co
d
ed
in
f
o
r
m
atio
n
t
h
at
ex
p
lai
n
s
t
h
e
o
cc
u
r
r
e
n
ce
o
f
a
n
attac
k
.
I
t is
q
u
ite
ess
e
n
tial
f
o
r
p
r
ed
ictio
n
in
ti
m
e
s
er
ies.
Ho
w
e
v
er
,
it
is
n
o
t
e
s
s
e
n
tial
f
o
r
class
if
ica
tio
n
w
h
e
r
e
th
e
m
o
d
el
m
u
s
t
r
ec
o
g
n
ize
t
h
e
attac
k
b
ased
o
n
i
ts
ch
ar
ac
ter
is
t
ics.
Fe
at
u
r
e
f
lo
w
d
u
r
atio
n
h
as m
o
r
e
i
m
p
ac
t
o
n
i
d
en
tify
i
n
g
attac
k
s
,
s
u
c
h
as
DDo
S a
n
d
Do
S,
d
u
e
t
o
its
r
ap
id
n
at
u
r
e.
I
n
t
h
e
d
etec
t
io
n
an
d
cla
s
s
i
f
icat
io
n
m
o
d
el
o
f
attac
k
,
t
h
e
ti
m
e
o
f
o
cc
u
r
r
en
ce
is
n
o
t
n
ec
e
s
s
ar
y
.
T
h
is
is
b
ec
au
s
e,
in
its
i
m
p
le
m
e
n
tatio
n
,
t
h
e
attac
k
h
ap
p
en
s
at
a
n
y
ti
m
e.
T
h
er
ef
o
r
e,
th
e
f
ea
tu
r
e
is
el
i
m
i
n
ated
as i
n
p
r
ev
io
u
s
s
t
u
d
ies
[
2
6
]
,
[
2
7
]
,
[
3
8
]
.
T
h
e
f
ir
s
t
s
ta
g
e
o
f
d
ataset
p
r
e
-
p
r
o
ce
s
s
i
n
g
i
s
f
ea
tu
r
e
e
n
co
d
in
g
,
w
h
ic
h
tr
an
s
f
o
r
m
s
t
h
e
d
ata
f
r
o
m
ca
teg
o
r
ical
in
to
n
u
m
er
ical.
T
h
e
f
ea
tu
r
e
e
n
co
d
in
g
p
r
o
ce
s
s
,
u
s
in
g
o
n
e
-
h
o
t
e
n
co
d
in
g
,
c
h
an
g
e
d
th
e
p
r
o
to
co
l
an
d
lab
el
f
ea
tu
r
es
i
n
to
n
u
m
er
ic
d
ata.
T
h
e
p
r
o
ce
s
s
m
ap
p
ed
th
e
p
r
o
to
co
l
f
ea
tu
r
e
to
3
in
s
tan
ce
s
,
in
cl
u
d
in
g
t
r
an
s
m
is
s
io
n
c
o
n
tr
o
l
p
r
o
to
co
l
(
T
C
P
)
,
u
s
er
d
atag
r
a
m
p
r
o
to
co
l
(
UDP
)
,
a
n
d
Ho
p
-
by
-
Ho
p
I
P
v
6
(
HOP
OPT
)
.
A
ls
o
,
t
h
e
lab
el
f
ea
t
u
r
es
b
ec
a
m
e
6
f
ea
tu
r
e
atta
ck
ca
teg
o
r
ies
b
y
eli
m
i
n
ati
n
g
t
h
e
in
f
i
ltra
tio
n
cla
s
s
.
Fi
n
all
y
,
8
0
f
ea
tu
r
es
f
o
r
d
ata
an
d
6
f
ea
tu
r
e
lab
els
w
er
e
o
b
tain
ed
.
T
h
e
n
ex
t
s
tag
e
w
as
t
h
e
f
ea
tu
r
e
s
ca
li
n
g
p
r
o
ce
s
s
to
tu
r
n
all
d
ata
v
alu
es
i
n
to
th
e
s
p
ec
i
f
ied
r
an
g
e.
T
h
is
p
r
o
ce
s
s
is
n
ec
es
s
ar
y
f
o
r
f
ea
tu
r
e
s
w
it
h
a
h
i
g
h
v
al
u
e
an
d
n
o
t
d
o
m
i
n
at
i
n
g
o
t
h
er
s
.
Feat
u
r
e
s
ca
lin
g
u
s
es
th
e
s
a
m
e
ap
p
r
o
a
ch
o
f
Min
-
m
ax
s
ca
li
n
g
w
ith
r
an
g
e
[
0
,
1
]
as
in
a
p
r
ev
io
u
s
s
tu
d
y
[
4
3
]
.
A
f
ter
p
r
e
-
p
r
o
ce
s
s
in
g
,
th
e
d
ata
is
r
ea
d
y
f
o
r
th
e
tr
ai
n
in
g
a
n
d
test
i
n
g
p
r
o
ce
s
s
.
3
.
3.
Dee
p lea
rning
a
rc
hite
ct
ure
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
o
f
t
h
e
i
n
tr
u
s
io
n
d
etec
tio
n
s
y
s
te
m
is
d
es
ig
n
ed
u
s
i
n
g
th
e
p
r
e
-
t
u
n
i
n
g
a
n
d
f
i
n
e
-
tu
n
i
n
g
p
r
o
ce
s
s
.
T
h
e
d
ee
p
lear
n
in
g
ar
ch
itect
u
r
e
u
s
ed
au
to
m
a
tic
f
ea
t
u
r
e
ex
tr
ac
tio
n
w
i
th
d
ee
p
au
to
-
en
co
d
er
(
DA
E
)
in
th
e
p
r
e
-
tu
n
in
g
s
ta
g
e
an
d
th
e
d
ee
p
n
eu
r
al
n
et
w
o
r
k
(
DNN)
ar
ch
itectu
r
e
in
th
e
f
i
n
e
-
tu
n
i
n
g
s
tag
e.
DA
E
p
er
f
o
r
m
s
t
h
e
p
r
o
ce
s
s
o
f
f
ea
tu
r
e
ex
tr
ac
tio
n
w
it
h
th
e
e
n
co
d
in
g
a
n
d
d
ec
o
d
i
n
g
p
h
a
s
e.
Au
to
-
e
n
co
d
er
g
en
er
ates
o
u
tp
u
t
̂
,
w
h
ic
h
is
r
ec
o
n
s
tr
u
cted
f
r
o
m
i
n
p
u
t
.
I
n
s
i
n
g
le
-
la
y
er
Au
to
-
en
co
d
er
,
w
h
e
n
th
e
in
p
u
t v
ec
to
r
∈
ℝ
,
th
e
v
e
cto
r
en
co
d
in
g
f
u
n
ctio
n
ℎ
,
in
f
o
r
w
ar
d
p
r
o
p
ag
atio
n
f
o
r
h
id
d
en
la
y
er
l
(
=
1
)
,
is
n
o
tate
d
as
s
h
o
w
n
in
(
1
)
.
T
h
e
d
ec
o
d
in
g
f
u
n
ctio
n
is
n
o
tated
as
s
h
o
w
n
i
n
(
2
)
;
ℎ
=
(
)
=
(
(
1
)
+
(
1
)
)
(
1
)
̂
=
(
ℎ
)
=
(
(
2
)
ℎ
+
(
2
)
)
(
2
)
ar
e
w
e
ig
h
t
m
atr
ice
s
,
ar
e
b
ias
v
ec
to
r
s
,
is
a
n
i
n
p
u
t
v
ec
to
r
f
o
r
m
d
ata
s
et,
a
n
d
f
(
.
)
a
n
d
g
(
.
)
ar
e
ac
tiv
atio
n
f
u
n
ctio
n
s
u
s
ed
o
n
a
h
id
d
en
la
y
er
.
T
h
e
ex
p
er
im
e
n
t
u
s
ed
s
e
v
er
al
v
ar
ian
ts
o
f
R
e
LU
ac
ti
v
atio
n
f
u
n
ctio
n
s
,
s
u
c
h
as
Se
LU
, P
R
e
LU
,
E
L
U,
an
d
L
ea
k
y
R
e
L
U
i
n
th
e
h
id
d
en
la
y
er
a
n
d
s
ig
m
o
id
in
t
h
e
las
t la
y
er
o
f
au
to
-
e
n
co
d
er
.
Dee
p
au
to
-
e
n
co
d
er
(
DA
E
)
is
th
e
d
ev
elo
p
m
e
n
t
o
f
a
s
i
n
g
le
A
E
w
i
th
a
h
i
g
h
er
n
u
m
b
er
o
f
la
y
er
s
.
T
h
e
f
u
n
ctio
n
co
m
p
o
s
itio
n
s
in
t
h
e
en
co
d
er
an
d
d
ec
o
d
er
ar
e
an
d
,
r
esp
ec
tiv
el
y
t
h
e
p
r
o
p
o
s
ed
ar
c
h
itect
u
r
e
u
s
e
s
7
h
id
d
en
la
y
er
s
,
w
it
h
th
e
o
u
tp
u
t
r
ec
o
n
s
tr
u
ct
io
n
̂
=
1
(
2
(
3
(
3
(
2
(
1
(
)
)
)
)
)
)
.
Fo
r
th
e
r
ec
o
n
s
tr
u
ct
io
n
p
r
o
ce
s
s
in
p
u
t
in
to
t
h
e
o
u
tp
u
t,
̂
u
s
e
s
th
e
M
SE
lo
s
s
f
u
n
ct
io
n
w
i
th
(
3
)
.
T
h
e
b
ac
k
p
r
o
p
ag
atio
n
p
r
o
ce
s
s
p
r
o
d
u
ce
s
a
lo
s
s
v
al
u
e
clo
s
e
to
ze
r
o
(
,
,
,
̂
)
=
1
2
‖
−
̂
‖
2
(
3
)
A
b
o
ttle
n
ec
k
la
y
er
(
m
id
d
le
la
y
er
)
w
it
h
a
s
m
a
ller
d
i
m
e
n
s
io
n
o
f
th
e
in
p
u
t
d
ataset
r
ep
r
esen
t
s
th
e
ex
tr
ac
ted
f
ea
t
u
r
es
=
3
=
(
(
3
)
.
2
+
(
3
)
)
.
T
h
e
r
esu
lt
o
f
th
e
f
ea
t
u
r
e
ex
tr
ac
tio
n
i
n
th
e
f
o
r
m
o
f
an
en
co
d
in
g
s
tr
u
ct
u
r
e
an
d
th
e
w
ei
g
h
t
an
d
b
ias
v
a
lu
e
s
ar
e
tr
an
s
f
er
r
ed
to
th
e
d
ee
p
n
eu
r
al
n
et
w
o
r
k
(
DNN)
s
tr
u
ct
u
r
e
f
o
r
th
e
f
in
e
-
tu
n
i
n
g
tr
ain
i
n
g
p
r
o
ce
s
s
.
T
h
e
e
n
co
d
in
g
r
esu
lt
o
f
th
i
s
s
tr
u
ct
u
r
e
b
ec
o
m
e
s
a
n
i
n
p
u
t
v
ec
to
r
i
n
t
h
e
DNN
clas
s
if
ier
ar
ch
itect
u
r
e.
T
h
e
DNN
m
o
d
el
u
s
e
s
th
e
w
ei
g
h
t,
b
ias
an
d
v
al
u
e
o
f
t
h
e
A
E
p
r
e
-
tr
ain
i
n
g
,
to
p
r
o
d
u
ce
th
e
o
u
tp
u
t
class
p
r
ed
ictio
n
̂
,
w
r
it
ten
a
s
;
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
Dee
p
lea
r
n
in
g
w
ith
fo
ca
l lo
s
s
a
p
p
r
o
a
c
h
fo
r
a
tta
ck
s
cla
s
s
ifica
tio
n
(
Yes
i No
va
r
ia
K
u
n
a
n
g
)
1411
̂
=
(
(
)
.
(
)
+
(
)
)
(
4
)
Ou
tp
u
t
̂
(
p
r
ed
ictio
n
tar
g
et
lab
e
l)
w
ill
clo
s
e
to
y
(
v
ec
to
r
tar
g
et
lab
el)
b
y
u
s
in
g
th
e
ac
tiv
a
tio
n
f
u
n
ctio
n
s
o
f
t
m
ax
(
.
)
in
t
h
e
last
la
y
er
.
Fo
r
all
th
e
tr
ai
n
in
g
d
ataset
s
(
,
)
,
th
e
f
u
n
ct
io
n
o
f
lo
s
s
ca
n
b
e
s
o
l
v
ed
b
y
:
(
,
)
=
1
∑
ℒ
(
̂
,
)
=
1
=
1
∑
(
,
;
,
)
=
1
(
5
)
w
h
er
e
ℒ
is
th
e
lo
s
s
f
u
n
ctio
n
,
a
n
d
is
th
e
n
u
m
b
er
o
f
d
ataset
s
.
O
n
e
f
o
c
u
s
o
f
th
is
s
tu
d
y
is
to
co
m
p
ar
e
t
h
e
ℒ
lo
s
s
f
u
n
ctio
n
u
s
in
g
cr
o
s
s
-
e
n
tr
o
p
y
,
w
ei
g
h
ted
cr
o
s
s
-
e
n
tr
o
p
y
,
an
d
f
o
ca
l lo
s
s
.
T
h
e
DNN
m
o
d
el
ca
r
r
ied
o
u
t
t
h
e
tr
ain
i
n
g
p
r
o
ce
s
s
f
o
r
clas
s
i
f
y
i
n
g
m
u
lti
-
cla
s
s
at
tack
s
.
A
h
y
p
e
r
p
ar
am
eter
tu
n
in
g
p
r
o
ce
s
s
is
p
er
f
o
r
m
ed
to
g
et
th
e
b
est
d
ee
p
lear
n
in
g
m
o
d
el
b
y
lo
o
k
in
g
at
th
e
r
ate
d
etec
tio
n
r
esu
lt
o
f
th
e
attac
k
cla
s
s
i
f
icatio
n
.
T
h
is
t
u
n
i
n
g
p
r
o
ce
s
s
tr
ied
v
ar
io
u
s
m
o
d
e
l
v
ar
ian
ts
b
ased
o
n
t
h
e
n
u
m
b
e
r
o
f
h
id
d
en
la
y
er
s
,
th
e
n
u
m
b
er
o
f
n
o
d
es,
lear
n
in
g
r
ate
v
alu
e,
b
atch
-
s
ize,
ac
ti
v
ati
o
n
f
u
n
ctio
n
,
an
d
k
er
n
e
l
in
it
iali
za
tio
n
to
g
et
th
e
b
est
m
o
d
el.
3
.
4.
L
o
s
s
f
un
ct
io
n
I
n
th
e
ca
s
e
o
f
a
m
u
lti
-
clas
s
w
it
h
th
e
n
u
m
b
er
o
f
clas
s
es
(
>
2
)
,
th
e
eq
u
atio
n
o
f
th
e
lo
s
s
f
u
n
ct
io
n
f
o
r
th
e
ca
teg
o
r
ical
cr
o
s
s
-
e
n
tr
o
p
y
(
C
E
)
i
s
:
ℒ
(
̂
,
)
=
=
−
∑
=
1
l
og
(
̂
)
(
6
)
is
th
e
n
u
m
b
er
o
f
clas
s
es,
is
t
h
e
g
r
o
u
n
d
tr
u
t
h
clas
s
,
an
d
̂
∈
[
0
,
1
]
is
th
e
m
o
d
el
'
s
p
r
ed
icted
p
r
o
b
ab
ili
t
y
f
o
r
th
e
clas
s
.
W
h
er
e
=
1
b
elo
n
g
s
to
t
h
e
ac
tu
al
lab
el
o
f
; o
th
er
w
i
s
e,
it
eq
u
als 0
.
Fo
r
i
m
b
ala
n
ce
d
class
ca
s
e
s
,
t
h
e
C
E
lo
s
s
f
u
n
ctio
n
is
m
o
d
i
f
i
ed
b
y
ad
d
in
g
a
w
ei
g
h
ti
n
g
f
ac
to
r
[
4
4
]
to
o
b
tain
th
e
C
E
a
s
s
h
o
w
n
i
n
(
7
)
.
W
e
ighte
d
=
−
∑
∝
=
1
l
og
(
̂
)
(
7
)
w
h
er
e
is
t
h
e
w
eig
h
t
f
ac
to
r
f
o
r
class
.
T
h
e
d
ef
icien
c
y
o
f
C
E
lo
s
s
is
t
h
at
m
a
n
y
s
a
m
p
les
co
n
tr
ib
u
te
t
o
a
s
ig
n
i
f
ica
n
t
ac
c
u
m
u
latio
n
o
f
t
h
e
lo
s
s
v
alu
e
ab
o
v
e
t
h
e
r
ar
e
class
[
3
3
]
,
[
3
4
]
.
T
h
er
ef
o
r
e,
w
h
en
t
h
e
e
x
tr
e
m
e
b
alan
ce
i
s
s
u
e
i
n
th
e
ca
s
e
o
f
m
u
lt
i
-
c
lass
at
tac
k
class
i
f
i
ca
tio
n
is
r
eso
lv
ed
,
th
e
s
ce
n
ar
io
tak
e
s
ad
v
an
ta
g
e
o
f
t
h
e
f
o
ca
l
lo
s
s
f
u
n
c
tio
n
p
r
o
p
o
s
ed
b
y
L
in
et
a
l
.
[
3
3
]
.
T
h
e
f
o
ca
l
lo
s
s
f
u
n
ctio
n
d
o
es
n
o
t
p
r
o
v
id
e
th
e
s
a
m
e
w
eig
h
ted
v
alu
e
o
n
all
tr
ai
n
i
n
g
d
ata.
I
n
c
o
n
tr
ast,
f
o
ca
l
lo
s
s
r
ed
u
ce
s
th
e
w
ei
g
h
t
o
f
w
el
l
-
c
la
s
s
i
f
ied
d
ata.
I
ts
i
m
p
ac
t o
n
f
o
ca
l lo
s
s
e
m
p
h
a
s
izes
tr
ain
i
n
g
o
n
d
ata
th
at
i
s
d
if
f
ic
u
lt
to
class
i
f
y
w
i
th
as
s
h
o
w
n
i
n
(
8
)
:
ℒ
(
̂
,
)
=
−
∑
(
1
−
̂
)
.
=
1
l
og
(
̂
)
(
8
)
w
it
h
as
a
m
o
d
u
lar
it
y
f
ac
to
r
to
r
ed
u
ce
th
e
w
eig
h
t
o
f
w
ell
-
c
lass
i
f
ied
class
e
s
.
W
h
en
=
0
,
th
e
lo
s
s
eq
u
als
to
cr
o
s
s
-
e
n
tr
o
p
y
.
T
h
er
ef
o
r
e,
>
=
0
is
s
et
to
ev
al
u
ate
t
h
e
e
f
f
ec
t
o
f
s
a
m
p
les
cla
s
s
i
f
ied
w
it
h
a
lo
s
s
f
ac
to
r
.
T
h
e
p
ar
am
eter
is
th
e
w
ei
g
h
t to
b
alan
ce
f
o
ca
l lo
s
s
,
an
d
it in
cr
ea
s
e
s
th
e
ac
c
u
r
ac
y
v
a
lu
e
f
o
r
th
e
i
m
b
alan
ce
class
.
3
.
5
.
E
x
peri
m
ent
a
l
s
et
up
a
n
d
perf
o
r
m
a
nce
m
et
ric
s
T
h
e
ex
p
er
im
e
n
t
w
as
r
u
n
o
n
th
e
clo
u
d
m
ac
h
in
e
i
n
th
e
Go
o
g
l
e
C
o
lab
o
r
ato
r
y
p
latf
o
r
m
.
T
h
e
m
o
d
el
w
as
d
ev
elo
p
ed
u
s
i
n
g
th
e
P
y
th
o
n
p
r
o
g
r
a
m
m
i
n
g
lan
g
u
ag
e
w
it
h
co
m
p
u
tat
io
n
u
tili
z
in
g
a
T
en
s
o
r
Fl
o
w
-
GP
U
lib
r
ar
y
o
f
Ker
as
[
4
5
]
,
a
d
ee
p
lear
n
in
g
f
r
a
m
e
w
o
r
k
.
T
h
e
h
y
p
er
p
ar
a
m
et
er
tu
n
in
g
p
r
o
ce
s
s
u
s
ed
T
alo
s
L
ib
r
ar
y
[
4
6
]
.
T
h
is
o
b
s
er
v
atio
n
u
s
ed
ac
cu
r
ac
y
,
s
en
s
iti
v
it
y
,
an
d
s
p
ec
if
ic
it
y
to
m
ea
s
u
r
e
th
e
p
er
f
o
r
m
a
n
ce
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el.
T
h
e
ev
al
u
ate
d
p
er
f
o
r
m
a
n
ce
u
s
ed
th
e
ac
c
u
r
a
c
y
f
u
n
ctio
n
to
ass
e
s
s
t
h
e
m
o
d
e
l's
ab
ilit
y
to
class
if
y
attac
k
s
co
r
r
ec
tl
y
.
I
n
th
e
ca
s
e
o
f
i
m
b
ala
n
ce
d
d
atas
ets,
th
e
p
r
ed
icted
r
esu
lt
w
as
d
o
m
i
n
ated
b
y
lar
g
e
n
u
m
b
er
s
o
f
class
es.
T
h
er
ef
o
r
e,
it
i
s
n
ec
es
s
ar
y
to
e
x
a
m
in
e
th
e
m
o
d
el
'
s
s
p
ec
if
icit
y
an
d
s
en
s
iti
v
it
y
f
o
r
i
m
b
a
lan
ce
d
d
ata
s
e
t
ca
s
e
[
4
7
]
.
T
h
e
s
en
s
iti
v
it
y
r
e
s
u
lts
s
h
o
w
ed
h
o
w
p
r
ec
is
e
l
y
t
h
e
m
o
d
el
d
etec
ted
an
attac
k
.
T
h
e
s
p
ec
if
icit
y
s
h
o
w
ed
th
e
p
r
o
b
ab
ilit
y
t
h
at
t
h
e
m
o
d
el
d
o
es n
o
t
m
ak
e
m
i
s
tak
e
s
i
n
r
ec
o
g
n
izi
n
g
an
attac
k
.
4.
RE
SU
L
T
S
A
ND
AN
AL
Y
SI
S
4
.
1
.
H
y
per
-
pa
ra
m
et
er
t
un
in
g
T
h
e
h
y
p
er
-
p
ar
a
m
eter
t
u
n
i
n
g
p
r
o
ce
s
s
is
cr
u
cial
i
n
o
b
tain
i
n
g
a
n
et
w
o
r
k
ar
c
h
itect
u
r
e
(
n
u
m
b
er
o
f
n
e
u
r
o
n
s
an
d
la
y
er
s
)
.
Mo
r
eo
v
er
,
t
h
e
p
r
o
ce
s
s
w
a
s
u
s
ed
to
o
b
tain
t
h
e
m
o
s
t
ap
p
r
o
p
r
iate
h
y
p
er
-
pa
r
a
m
ete
r
v
alu
e
s
i
n
t
h
e
d
ee
p
lear
n
in
g
m
o
d
el.
I
n
th
e
i
n
it
ial
p
h
ase,
s
e
v
er
al
h
y
p
er
-
p
ar
a
m
eter
p
r
o
ce
s
s
es
w
er
e
p
er
f
o
r
m
ed
o
n
s
ev
er
al
h
id
d
en
la
y
er
s
an
d
n
o
d
es,
b
atc
h
s
ize,
lear
n
in
g
r
ate,
ac
ti
v
atio
n
f
u
n
ctio
n
,
an
d
k
er
n
el
i
n
itial
f
o
r
d
ee
p
lear
n
in
g
m
o
d
el
(
DA
E
-
DNN)
.
T
h
e
ex
p
er
i
m
en
ts
u
s
ed
ca
teg
o
r
ical
cr
o
s
s
-
e
n
tr
o
p
y
as
th
e
lo
s
t
f
u
n
ctio
n
.
T
h
e
b
est
ar
ch
itectu
r
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1693
-
6930
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
,
Vo
l.
19
,
No
.
4
,
A
u
g
u
s
t 2
0
2
1
:
1
4
0
7
-
1
4
1
8
1412
o
b
tain
ed
f
o
r
th
e
C
E
lo
s
s
o
n
th
e
C
SE
-
C
I
C
-
I
DS2
0
1
8
d
ataset
is
th
e
D
A
E
s
tr
u
ctu
r
e,
w
it
h
7
h
i
d
d
en
lay
er
s
(
8
0
-
70
-
40
-
30
-
25
-
30
-
40
-
70
-
8
0
)
.
I
t
w
a
s
ex
tr
ac
ted
a
n
d
tr
an
s
f
er
r
ed
to
t
h
e
m
o
d
el
DNN
to
b
ec
o
m
e
8
0
-
70
-
40
-
30
-
25
-
6
.
T
h
is
ar
ch
itect
u
r
e
w
a
s
o
b
tain
ed
b
y
t
r
y
in
g
d
if
f
er
en
t
v
ar
iatio
n
s
in
t
h
e
n
u
m
b
er
o
f
h
id
d
en
la
y
er
s
D
A
E
[
1
,
3,
5
an
d
7
]
.
T
h
e
b
est
lear
n
in
g
r
ate
v
alu
e
u
s
ed
w
a
s
0
.
0
0
1
,
w
ith
t
h
e
v
al
u
es
o
f
[
0
.
0
0
0
0
1
,
0
.
0
0
0
1
,
0
.
0
0
1
,
0
.
0
1
,
0
.
1
]
.
T
h
e
ex
p
er
i
m
e
n
t
u
s
ed
th
e
i
n
itial
lec
u
n
_
u
n
if
o
r
m
k
er
n
e
l,
as
w
ell
as
th
e
L
ea
k
y
R
e
L
U
ac
ti
v
atio
n
f
u
n
ct
io
n
.
T
h
e
L
ea
k
y
R
eL
U
w
a
s
p
r
ev
io
u
s
l
y
s
elec
ted
th
r
o
u
g
h
t
h
e
tu
n
in
g
p
r
o
ce
s
s
o
f
v
ar
io
u
s
ac
ti
v
atio
n
f
u
n
ctio
n
s
.
T
h
e
b
atch
s
ize
v
a
lu
e
f
o
r
th
e
b
est
m
o
d
el
u
s
ed
is
2
5
6
.
A
ls
o
,
th
is
w
as
o
b
tain
ed
th
r
o
u
g
h
th
e
t
u
n
i
n
g
p
r
o
ce
s
s
w
it
h
b
atch
s
ize
v
ar
iatio
n
s
3
2
,
6
4
,
an
d
2
5
6
.
Af
ter
g
etti
n
g
th
e
b
est m
o
d
el
w
ith
th
e
C
E
lo
s
s
f
u
n
ctio
n
as
th
e
b
asis
o
f
co
m
p
ar
is
o
n
,
tu
n
i
n
g
f
o
r
th
e
f
o
ca
l
lo
s
s
p
ar
a
m
eter
w
as p
er
f
o
r
m
ed
.
T
w
o
p
ar
a
m
eter
s
w
er
e
tu
n
ed
i
n
a
m
u
lt
i
-
c
lass
f
o
ca
l lo
s
s
,
in
wh
ich
th
e
s
etti
n
g
s
o
f
γ
r
ed
u
ce
d
th
e
ef
f
ec
t o
f
th
e
m
o
d
u
latio
n
f
ac
to
r
.
T
h
e
α
p
ar
am
et
er
is
th
e
w
ei
g
h
t f
ac
to
r
f
o
r
th
e
class
.
T
h
e
f
o
ca
l lo
s
s
p
ar
am
eter
s
tu
n
ed
ar
e
r
an
g
e
γ
∈
[
0
,
5
]
an
d
α
∈
[
0
,
1
]
,
as r
ec
o
m
m
en
d
ed
in
[
3
3
]
.
4
.
2
.
Resul
t
o
f
v
a
rio
us
o
f
f
o
c
a
l lo
s
s
T
h
e
ef
f
ec
ti
v
en
e
s
s
o
f
th
e
f
o
ca
l
lo
s
s
f
u
n
c
tio
n
i
n
attac
k
cl
a
s
s
i
f
icatio
n
w
as
m
ea
s
u
r
ed
b
y
ta
k
i
n
g
t
h
e
b
est
tu
n
in
g
r
es
u
lt
i
n
f
o
ca
l
lo
s
s
p
a
r
a
m
eter
s
.
T
h
e
tr
ain
in
g
p
r
o
ce
s
s
w
a
s
p
er
f
o
r
m
ed
w
it
h
th
e
n
u
m
b
er
o
f
ep
o
ch
=3
0
.
T
ab
le
2
s
u
m
m
ar
izes t
h
e
o
v
er
a
ll h
y
p
er
-
p
ar
a
m
eter
t
u
n
i
n
g
r
e
s
u
lts
f
o
r
th
e
f
o
ca
l lo
s
s
f
u
n
ctio
n
,
w
it
h
v
ar
io
u
s
v
a
lu
e
s
o
f
an
d
.
A
ls
o
,
t
h
e
w
ei
g
h
t a
s
s
i
g
n
ed
to
th
e
r
ar
e
class
h
a
s
a
s
tab
le
r
an
g
e.
Ho
w
ev
er
,
it
in
ter
ac
t
s
w
ith
γ
,
m
a
k
i
n
g
it
n
ec
es
s
ar
y
to
s
elec
t
t
h
e
t
w
o
p
ar
am
eter
s
to
g
et
h
er
,
as
s
h
o
w
n
in
T
ab
les
2
(
a
)
an
d
2
(
b
)
.
I
n
g
en
er
al,
in
cr
ea
s
ed
s
lig
h
tl
y
a
s
f
lu
ct
u
at
ed
.
I
n
t
h
i
s
ca
s
e,
=
0
.
5
w
o
r
k
s
b
est
w
h
en
=
1
.
T
h
e
b
est r
esu
lts
ar
e
t
h
e
ac
c
u
r
ac
y
v
al
u
e
o
f
9
8
.
2
2
3
%,
th
e
s
en
s
iti
v
it
y
o
f
9
8
.
2
2
3
%,
an
d
s
p
ec
if
icit
y
o
f
9
9
.
8
1
4
% f
o
r
th
e
en
t
ir
e
attac
k
clas
s
e
s
.
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
r
ea
ch
ed
t
h
e
h
ig
h
es
t
ac
cu
r
ac
y
m
etr
ic
at
=1
.
I
t
is
r
ea
s
o
n
ab
le
b
ec
au
s
e
m
in
i
m
ize
s
th
e
lo
s
s
co
n
tr
ib
u
tio
n
o
f
th
e
d
o
m
in
a
n
t
clas
s
s
a
m
p
le
th
at
is
e
asil
y
clas
s
i
f
ied
.
W
h
en
p
ar
am
e
ter
in
cr
ea
s
es,
th
e
p
r
o
b
a
b
ilit
y
o
f
co
r
r
ec
t
class
if
ic
atio
n
(
1
−
̂
)
d
ec
r
ea
s
es.
T
h
is
p
r
o
b
ab
il
it
y
in
cr
ea
s
e
s
th
e
w
ei
g
h
t
o
f
m
in
o
r
ity
clas
s
s
a
m
p
les
t
h
at
ar
e
d
if
f
ic
u
lt
to
b
e
class
i
f
ied
.
As
a
r
esu
lt,
t
h
e
m
o
d
el
f
o
cu
s
es
o
n
t
h
e
d
if
f
ic
u
lt
y
c
lass
o
f
class
if
ied
s
a
m
p
les t
h
at
lo
w
er
s
cla
s
s
i
f
icat
io
n
ac
cu
r
ac
y
.
T
ab
le
2
.
E
x
p
er
im
e
n
tal
r
esu
lts
w
it
h
a
v
ar
ie
t
y
o
f
v
al
u
es
α
an
d
γ
f
o
r
class
i
f
y
i
n
g
attac
k
s
w
i
t
h
3
0
ep
o
ch
tr
ain
in
g
p
r
o
ce
s
s
es.
(
a)
C
SE
-
C
I
C
-
I
S2
0
1
8
w
it
h
α
-
b
alan
ce
d
C
E
ac
h
ie
v
e
s
at
m
o
s
t 9
8
.
2
1
% a
cc
u
r
ac
y
.
(
b
)
I
n
co
n
tr
ast,
u
s
in
g
FL
w
i
th
t
h
e
s
a
m
e
n
e
t
w
o
r
k
w
it
h
v
ar
y
in
g
γ
/
α
g
i
v
es a
cc
u
r
ac
y
a
t 9
8
.
2
2
3
% a
t γ
=1
an
d
α
=0
.
5
s
e
ttin
g
s
(
a)
Var
y
i
n
g
α
f
o
r
C
E
lo
s
s
(
γ
=0
)
Α
A
c
c
u
r
a
c
y
(
%)
S
e
n
si
t
i
v
ity
(
%)
S
p
e
c
i
f
i
ci
ty
(
%)
0
.
1
9
8
.
1
7
9
8
.
1
7
9
9
.
7
9
0
.
2
5
9
8
.
14
9
8
.
1
4
9
9
.
7
0
0
.
5
9
8
.
1
9
9
8
.
1
9
9
9
.
7
9
0
.
7
5
9
8
.
2
1
9
8
.
2
1
9
9
.
7
9
1
9
8
.
16
9
8
.
1
6
9
9
.
7
7
(
b
)
Var
y
in
g
γ
f
o
r
F
L
(
w
.
o
p
ti
m
al
α
)
γ
A
c
c
u
r
a
c
y
(
%)
S
e
n
si
t
i
v
ity
(
%)
S
p
e
c
i
f
i
ci
ty
(
%)
0
0
.
7
5
9
8
.
2
10
9
8
.
2
1
0
9
9
.
7
9
2
0
.
1
0.
1
9
8
.
1
8
5
9
8
.
1
8
5
9
9
.
7
8
8
0
.
2
0.
7
5
9
8
.
1
5
2
9
8
.
1
5
2
9
9
.
8
0
7
0
.
5
1
9
8
.
2
20
9
8
.
2
2
0
9
9
.
8
1
3
1
0
.
5
9
8
.
2
2
3
9
8
.
2
2
3
9
9
.
8
1
4
2
0.
5
9
8
.
119
9
8
.
1
1
9
9
9
.
7
7
8
5
0
.
7
5
9
8
.
062
9
8
.
0
6
2
9
9
.
7
1
9
4
.
3
.
P
er
f
o
rm
a
nce
a
n
d c
o
m
p
a
riso
n
T
h
e
r
esu
lts
o
f
co
n
f
ig
u
r
i
n
g
NI
DS
w
it
h
f
o
ca
l
lo
s
s
(
NI
DS
-
F
L
)
w
er
e
ev
a
lu
ated
b
y
co
m
p
ar
i
n
g
th
e
m
w
it
h
cr
o
s
s
-
e
n
tr
o
p
y
lo
s
s
(
NI
DS
-
C
E
)
,
an
d
w
eig
h
ted
cr
o
s
s
-
e
n
tr
o
p
y
lo
s
s
(
NI
DS
-
W
C
E
)
ac
co
r
d
in
g
l
y
.
E
q
u
al
v
al
u
es
o
f
n
et
w
o
r
k
ar
ch
itect
u
r
e
(
n
u
m
b
er
o
f
h
id
d
en
la
y
er
s
,
n
u
m
b
er
o
f
n
o
d
es),
an
d
h
y
p
er
-
p
ar
a
m
eter
v
a
lu
e
w
er
e
u
s
ed
.
T
h
e
NI
DS
-
C
E
an
d
NI
DS
-
W
C
E
co
n
f
ig
u
r
atio
n
d
o
n
o
t
u
s
e
th
e
γ
an
d
α
.
T
h
e
w
ei
g
h
ted
cr
o
s
s
-
en
tr
o
p
y
u
s
ed
a
b
alan
ce
d
m
o
d
e.
I
t
m
ea
n
s
t
h
at
t
h
is
f
u
n
ct
io
n
r
ep
licates
th
e
s
m
aller
cla
s
s
u
n
ti
l
t
h
e
n
u
m
b
er
o
f
s
a
m
p
les
in
t
h
e
m
i
n
o
r
it
y
an
d
lar
g
er
class
e
s
is
eq
u
al.
I
n
th
e
f
ir
s
t
s
ta
g
e,
tr
ain
i
n
g
w
as
co
n
d
u
cted
u
s
in
g
ep
o
ch
=3
0
.
Af
ter
th
e
tr
ain
i
n
g
p
r
o
ce
s
s
,
an
ev
alu
a
tio
n
w
a
s
p
er
f
o
r
m
ed
u
s
i
n
g
te
s
tin
g
d
ata.
Fig
u
r
es
2
s
h
o
w
a
ll
t
h
e
m
e
tr
ic
co
m
p
ar
is
o
n
s
w
it
h
v
ar
io
u
s
v
ar
ian
t
s
o
f
th
e
lo
s
s
f
u
n
ctio
n
.
I
t
s
h
o
w
s
th
a
t
f
o
r
ep
o
ch
=3
0
,
alm
o
s
t
th
e
m
o
d
els
'
o
v
e
r
all
p
er
f
o
r
m
a
n
ce
u
s
in
g
th
e
f
o
ca
l
lo
s
s
f
u
n
ctio
n
w
a
s
b
etter
th
an
C
E
a
n
d
W
C
E
.
R
e
s
p
ec
tiv
el
y
,
th
e
ac
c
u
r
ac
y
v
al
u
e
i
s
9
8
.
2
3
%,
p
r
ec
is
io
n
to
9
8
.
3
4
%,
r
ec
all
(
s
en
s
iti
v
it
y
)
to
9
8
.
2
3
%,
an
d
s
p
ec
if
icit
y
to
9
8
.
2
5
%,
as
s
h
o
w
n
in
Fig
u
r
e
2
(
a
)
.
T
h
is
r
esear
ch
u
s
ed
a
m
u
lti
-
clas
s
class
i
f
icat
io
n
f
o
r
B
o
T
,
B
r
u
te
Fo
r
ce
,
DDOS,
Do
S,
an
d
w
eb
attac
k
s
.
T
h
e
r
es
u
lts
s
h
o
w
ed
t
h
at
NI
DS
's p
er
f
o
r
m
an
ce
u
s
i
n
g
f
o
ca
l
lo
s
s
w
as
h
i
g
h
er
t
h
a
n
cr
o
s
s
-
en
tr
o
p
y
a
n
d
w
e
ig
h
ted
cr
o
s
s
-
en
tr
o
p
y
,
as
p
r
esen
ted
in
F
ig
u
r
e
2
(
a
)
.
T
h
e
d
etailed
r
esu
lt
f
o
r
th
e
en
tire
class
m
a
y
b
e
o
b
s
er
v
ed
in
T
ab
le
3
.
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
ex
ce
llen
tl
y
d
etec
t
ed
B
o
T
an
d
DDo
S
attac
k
s
,
w
it
h
an
o
v
er
all
p
er
f
o
r
m
an
ce
ab
o
v
e
9
9
.
9
%.
T
h
e
o
v
er
all
p
er
f
o
r
m
an
ce
f
o
r
Do
S
atta
ck
s
w
as
h
i
g
h
er
th
an
9
0
%,
w
h
ile
f
o
r
B
r
u
te
Fo
r
ce
,
r
ec
all
p
er
f
o
r
m
an
ce
at
tain
ed
9
4
%,
w
ith
a
p
r
ec
is
io
n
o
f
o
n
l
y
8
4
%.
T
h
e
f
ile
t
r
an
s
f
e
r
p
r
o
to
co
l
(
F
T
P
)
-
b
r
u
te
attac
k
h
as
a
c
h
ar
ac
ter
is
tic
t
h
at
r
ese
m
b
les
th
e
s
lo
w
-
h
y
p
er
tex
t
t
r
an
s
f
er
p
r
o
to
co
l
(
H
T
T
P
)
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
Dee
p
lea
r
n
in
g
w
ith
fo
ca
l lo
s
s
a
p
p
r
o
a
c
h
fo
r
a
tta
ck
s
cla
s
s
ifica
tio
n
(
Yes
i No
va
r
ia
K
u
n
a
n
g
)
1413
Do
s
.
B
o
th
o
f
th
e
m
ar
e
o
f
ten
m
is
c
lass
if
ied
as
a
r
esu
lt,
th
is
r
ed
u
ce
d
th
e
r
ec
o
g
n
itio
n
p
er
f
o
r
m
an
ce
o
f
b
o
th
t
y
p
es
o
f
attac
k
s
.
T
ab
le
3
.
P
er
f
o
r
m
a
n
ce
o
f
ea
ch
attac
k
clas
s
f
o
r
ep
o
ch
=3
0
B
e
n
i
g
n
B
o
T
B
r
u
t
e
F
o
r
c
e
D
D
o
S
D
o
S
W
e
b
A
t
t
a
c
k
P
r
e
c
i
si
o
n
(
%)
N
I
D
S
-
CE
9
9
.
9
6
9
9
.
9
6
8
4
.
2
9
9
.
9
6
9
6
.
0
8
8
1
.
9
4
N
I
D
S
-
W
C
E
9
9
.
9
4
9
9
.
9
8
8
3
.
7
7
9
9
.
8
2
9
6
.
5
4
9
7
.
3
N
I
D
S
-
FL
9
9
.
9
7
9
9
.
9
8
8
3
.
9
1
9
9
.
9
8
9
6
.
5
9
6
.
2
7
R
e
c
a
l
l
(
%)
N
I
D
S
-
CE
9
9
.
9
7
9
9
.
9
7
9
3
.
1
6
9
9
.
9
8
9
0
.
6
6
7
.
8
2
N
I
D
S
-
W
C
E
9
9
.
9
2
9
9
.
9
5
9
4
.
0
1
9
9
.
9
8
9
0
.
2
1
4
1
.
3
8
N
I
D
S
-
FL
9
9
.
9
7
9
9
.
9
7
9
3
.
9
7
9
9
.
9
9
9
0
.
3
2
7
4
.
1
4
F1
-
S
c
o
r
e
(
%)
N
I
D
S
-
CE
9
9
.
9
7
9
9
.
9
6
8
8
.
4
5
9
9
.
9
7
9
3
.
2
6
7
4
.
2
1
N
I
D
S
-
W
C
E
9
9
.
9
3
9
9
.
9
6
8
8
.
6
9
9
.
9
9
3
.
2
7
5
8
.
0
7
N
I
D
S
-
FL
9
9
.
9
7
9
9
.
9
7
8
8
.
6
5
9
9
.
9
8
9
3
.
3
1
8
3
.
7
7
T
o
in
v
es
tig
a
te
th
e
p
er
f
o
r
m
a
n
ce
o
f
t
h
e
lo
s
s
f
u
n
ct
io
n
ag
a
in
s
t
t
h
e
i
m
b
alan
ce
d
d
ataset,
th
is
s
t
u
d
y
ex
a
m
in
ed
t
h
e
m
o
d
el’
s
e
f
f
icie
n
c
y
i
n
cla
s
s
i
f
y
in
g
t
y
p
es
o
f
at
tac
k
s
,
e
s
p
ec
iall
y
in
m
in
o
r
it
y
cla
s
s
es.
T
h
e
w
eb
attac
k
s
ar
e
a
m
i
n
o
r
it
y
cla
s
s
t
h
at
o
n
l
y
a
m
o
u
n
ts
to
0
.
0
5
%
o
f
th
e
to
tal
d
ata
tr
ain
ed
in
T
ab
le
1
.
A
cc
o
r
d
in
g
to
T
ab
le
3
an
d
Fig
u
r
e
2
(
b
)
,
th
e
NI
DS
-
F
L
o
u
t
p
er
f
o
r
m
s
th
e
o
th
er
m
et
h
o
d
s
to
class
i
f
y
w
eb
attac
k
s
.
T
h
er
e
is
a
s
ig
n
i
f
ica
n
t
in
cr
ea
s
e
in
t
h
e
v
al
u
e
o
f
p
r
ec
is
io
n
,
r
ec
all,
an
d
f
1
-
s
co
r
e
co
m
p
ar
ed
to
m
o
d
el
s
t
h
at
u
s
e
C
E
a
n
d
W
C
E
lo
s
s
es.
T
h
e
r
ec
all
(
s
en
s
i
tiv
it
y
)
r
ea
c
h
es
7
4
.
1
4
%,
im
p
l
y
i
n
g
a
n
ap
p
r
o
x
i
m
ate
in
cr
e
ase
o
f
7
%
f
r
o
m
C
E
as
a
p
r
i
m
ar
y
lo
s
s
f
u
n
ctio
n
.
T
h
e
m
o
d
el
t
h
at
u
s
e
s
W
C
E
in
m
i
n
o
r
it
y
clas
s
es
w
it
h
ep
o
ch
=3
0
d
o
es
n
o
t
h
a
v
e
e
x
ce
lle
n
t
s
e
n
s
it
iv
i
t
y
,
alth
o
u
g
h
it
h
as
a
g
o
o
d
p
r
ec
is
i
o
n
v
alu
e.
T
h
e
lo
s
s
v
al
u
e
o
f
t
h
e
3
m
o
d
els
in
F
ig
u
r
es
2
(
c
)
an
d
,
t
h
e
F
L
f
u
n
ctio
n
,
i
s
s
m
al
ler
th
a
n
t
h
e
C
E
an
d
W
C
E
.
A
l
s
o
,
th
e
n
u
m
b
er
o
f
m
i
s
cla
s
s
i
f
icatio
n
attac
k
s
w
er
e
co
m
p
ar
ed
.
T
h
e
er
r
o
r
co
u
n
t
v
al
u
e
o
f
NI
DS
-
F
L
i
n
Fig
u
r
e
2
(
d
)
is
lo
w
er
th
a
n
o
th
er
m
o
d
els,
w
h
ich
ar
e
o
n
l
y
6
9
6
5
.
T
h
e
s
u
p
er
io
r
ity
o
f
t
h
e
FL
s
h
o
w
s
t
h
e
ef
f
ec
t
o
f
m
o
d
u
lar
it
y
a
n
d
w
ei
g
h
t
f
ac
to
r
s
o
n
f
o
ca
l
lo
s
s
.
B
y
s
elec
ti
n
g
t
h
e
m
o
s
t
ap
p
r
o
p
r
iate
m
o
d
u
lar
it
y
an
d
w
ei
g
h
t
f
o
r
th
e
i
m
b
alan
ce
d
cla
s
s
,
th
e
lo
s
s
an
d
m
is
c
lass
if
ica
tio
n
v
a
lu
e
s
ca
n
b
e
m
in
i
m
ized
,
esp
ec
iall
y
f
o
r
th
e
m
in
o
r
it
y
clas
s
.
(
a)
(
b
)
(
c)
(
d
)
Fig
u
r
e
2
.
Gr
ap
h
s
o
f
co
m
p
ar
is
o
n
o
f
v
ar
io
u
s
te
s
tin
g
r
es
u
lt
m
etr
ics f
o
r
all
m
o
d
els (
NI
DS
-
C
E
,
NI
DS
-
W
C
E
,
a
n
d
NI
DS
-
FL
)
f
o
r
th
e
p
r
o
ce
s
s
ep
o
ch
=
3
0
;
(
a)
Ov
er
all
p
er
f
o
r
m
a
n
ce
o
f
attac
k
s
cla
s
s
i
f
icatio
n
;
(
b
)
P
e
r
f
o
r
m
a
n
ce
s
o
f
m
i
n
o
r
it
y
clas
s
-
W
eb
A
ttac
k
cla
s
s
;
(
c)
L
o
s
s
v
al
u
e
o
f
all
m
o
d
el
;
(
d
)
E
r
r
o
r
c
o
u
n
t o
f
all
m
o
d
el
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1693
-
6930
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
,
Vo
l.
19
,
No
.
4
,
A
u
g
u
s
t 2
0
2
1
:
1
4
0
7
-
1
4
1
8
1414
I
n
th
e
n
e
x
t
p
h
a
s
e,
th
e
m
o
d
el
w
a
s
ev
al
u
ated
b
y
in
cr
ea
s
i
n
g
t
h
e
n
u
m
b
er
o
f
ep
o
ch
s
to
2
0
0
.
As
s
h
o
w
n
b
y
th
e
lo
s
s
ac
h
ie
v
ed
in
Fi
g
u
r
e
3
(
a)
,
th
e
p
r
o
p
o
s
ed
d
ee
p
lear
n
in
g
m
o
d
el
u
s
i
n
g
F
L
w
it
h
th
e
p
r
ev
io
u
s
l
y
s
elec
ted
h
y
p
er
-
p
ar
a
m
eter
s
co
n
v
er
g
e
f
a
s
ter
th
a
n
C
E
an
d
W
C
E
.
Fig
u
r
e
3
(
b
)
s
h
o
w
s
t
h
e
n
e
t
w
o
r
k
u
s
in
g
F
L
s
tab
ilize
s
af
ter
ar
o
u
n
d
3
0
ep
o
ch
s
,
w
h
ic
h
is
in
co
n
tr
ast
to
1
0
0
e
p
o
ch
s
an
d
8
0
ep
o
ch
s
w
it
h
C
E
lo
s
s
an
d
W
C
E
lo
s
s
,
r
esp
ec
tiv
el
y
.
Ho
w
e
v
er
,
w
it
h
th
e
i
n
cr
ea
s
i
n
g
n
u
m
b
er
o
f
ep
o
ch
s
,
th
e
m
o
d
els
u
s
in
g
th
e
cr
o
s
s
-
e
n
tr
o
p
y
f
u
n
cti
o
n
ar
e
m
o
r
e
s
tab
le.
A
l
s
o
,
th
e
y
ten
d
to
k
ee
p
in
cr
ea
s
in
g
co
m
p
ar
ed
to
m
o
d
els
th
at
u
s
e
f
o
ca
l
lo
s
s
a
n
d
w
eig
h
ted
cr
o
s
s
-
en
tr
o
p
y
f
u
n
c
tio
n
s
.
T
h
e
im
p
r
o
v
e
m
en
t is r
ea
s
o
n
ab
l
e
b
ec
au
s
e
th
e
h
y
p
er
-
p
ar
a
m
eter
tu
n
i
n
g
p
r
o
ce
s
s
w
a
s
p
er
f
o
r
m
ed
o
n
a
m
o
d
el
w
it
h
a
cr
o
s
s
-
e
n
tr
o
p
y
lo
s
s
f
u
n
ct
io
n
.
I
t
h
as
p
r
o
d
u
ce
d
a
m
o
d
el
w
it
h
th
e
m
o
s
t
ap
p
r
o
p
r
iate
h
y
p
er
-
p
ar
am
eter
s
f
o
r
d
ee
p
lear
n
in
g
.
T
h
e
r
esu
lti
n
g
f
o
ca
l
lo
s
s
cu
r
v
e
te
n
d
s
to
f
lu
ct
u
ate
d
u
e
to
f
ac
to
r
γ
.
T
h
er
ef
o
r
e,
w
i
th
a
n
o
s
cillati
n
g
c
u
r
v
e,
th
e
m
o
d
el
h
a
s
a
s
li
g
h
tl
y
h
i
g
h
e
r
v
alid
atio
n
v
a
lu
e
w
h
e
n
th
e
c
u
r
v
e
r
ea
ch
es it
s
p
ea
k
.
T
ab
le
4
s
h
o
w
s
th
e
b
est
co
m
p
ar
is
o
n
o
f
r
esu
lts
f
o
r
th
e
3
m
o
d
el
s
af
ter
2
0
0
ep
o
ch
s
.
T
h
e
o
v
er
all
p
er
f
o
r
m
a
n
ce
r
es
u
lts
ar
e
al
m
o
s
t
th
e
s
a
m
e
b
ased
o
n
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
F1
-
s
co
r
e,
a
n
d
s
p
ec
if
icit
y
.
Fo
r
in
s
ta
n
ce
,
t
h
e
r
ec
all
v
a
lu
e
in
d
ic
ates
a
t
in
y
d
i
f
f
er
en
ce
o
f
<0
.
0
1
%.
I
n
t
h
e
w
eb
attac
k
as
a
m
i
n
o
r
it
y
cla
s
s
,
t
h
e
o
v
er
all
p
er
f
o
r
m
a
n
ce
s
a
f
ter
2
0
0
ep
o
ch
s
f
o
r
m
o
d
els
t
h
at
u
s
e
f
o
ca
l
lo
s
s
f
u
n
c
tio
n
f
o
r
th
e
p
r
ec
is
io
n
,
r
e
ca
ll,
ac
cu
r
ac
y
,
a
n
d
F1
-
s
co
r
e
v
al
u
es,
r
esp
ec
ti
v
el
y
,
a
m
o
u
n
ted
to
9
7
.
7
6
%,
7
5
.
2
9
%,
an
d
8
5
.
0
7
%.
(
a)
(
b
)
Fig
u
r
e
3
.
NI
DS v
alid
atio
n
ac
c
u
r
ac
y
d
u
r
i
n
g
tr
ai
n
i
n
g
w
it
h
C
E
lo
s
s
,
w
eig
h
ted
W
C
E
,
v
s
.
f
o
ca
l
lo
s
s
;
(
a)
co
m
p
ar
is
o
n
o
f
lo
s
s
v
a
lu
e;
(
b
)
co
m
p
ar
is
o
n
o
f
ac
c
u
r
ac
y
T
ab
le
4
.
C
o
m
p
ar
is
o
n
o
f
th
e
Hi
g
h
e
s
t p
er
f
o
r
m
a
n
ce
s
o
f
t
h
e
NI
DS
m
o
d
els p
r
esen
ted
in
t
h
is
s
t
u
d
y
T
r
a
i
n
i
n
g
S
e
t
P
e
r
f
o
r
man
c
e
s i
n
%
(
e
p
o
c
h
=
3
0
)
T
e
st
i
n
g
S
e
t
P
e
r
f
o
r
man
c
e
s i
n
%
(
e
p
o
c
h
=
3
0
)
A
c
c
.
(
o
v
e
r
a
l
l
)
P
r
e
c
.
R
e
c
a
l
l
/
S
e
n
s.
F1
-
sco
r
e
S
p
e
c
i
f
.
L
o
ss
Er
r
o
r
C
o
u
n
t
A
c
c
.
(
o
v
e
r
a
l
l
)
P
r
e
c
.
R
e
c
a
l
l
/
S
e
n
s.
F1
-
sco
r
e
S
p
e
c
i
f
.
L
o
ss
Er
r
o
r
C
o
u
n
t
N
I
D
S
-
CE
98
.
2
0
9
8
.
2
9
9
8
.
2
0
9
8
.
2
2
9
9
.
8
0
0
.
0
3
4
0
4
2
8
3
3
0
9
8
.
2
0
9
8
.
2
9
9
8
.
2
0
9
8
.
2
2
9
9
.
8
0
0
.
0
3
3
5
8
7
0
7
6
N
I
D
S
-
W
C
E
9
8
.
1
6
9
8
.
2
8
9
8
.
1
6
9
8
.
1
8
9
9
.
7
9
0
.
0
3
4
7
2
2
8
8
9
2
9
8
.
1
7
9
8
.
2
8
9
8
.
1
7
9
8
.
1
9
9
9
.
7
8
0
.
0
3
4
4
1
7
1
9
1
N
I
D
S
-
FL
9
8
.
2
2
9
8
.
3
3
9
8
.
2
2
9
8
.
2
4
9
9
.
8
1
0
.
0
0
4
0
0
2
7
9
6
2
9
8
.
2
3
9
8
.
3
4
9
8
.
2
3
9
8
.
2
5
9
9
.
8
1
0
.
0
0
3
9
0
6
9
6
5
T
r
a
i
n
i
n
g
S
e
t
P
e
r
f
o
r
man
c
e
s i
n
%
(
e
p
o
c
h
=
2
0
0
)
T
e
st
i
n
g
S
e
t
P
e
r
f
o
r
man
c
e
s i
n
%
(
e
p
o
c
h
=
2
0
0
)
N
I
D
S
-
CE
9
8
.
2
6
9
8
.
3
8
9
8
.
2
6
9
8
.
2
8
9
9
.
8
2
0
.
0
3
2
6
1
2
7
3
0
3
9
8
.
2
7
9
8
.
3
8
9
8
.
2
7
9
8
.
2
9
9
9
.
8
1
0
.
0
3
2
8
2
6
8
1
4
N
I
D
S
-
W
C
E
9
8
.
2
6
9
8
.
3
7
9
8
.
2
6
9
8
.
2
8
9
9
.
8
2
0
.
0
3
2
8
4
2
7
3
8
0
9
8
.
2
6
9
8
.
3
6
9
8
.
2
6
9
8
.
2
8
9
9
.
8
1
0
.
0
3
2
7
2
6
8
5
2
N
I
D
S
-
FL
9
8
.
2
7
9
8
.
3
8
9
8
.
2
7
9
8
.
2
9
9
9
.
8
2
0
.
0
0
1
4
0
2
7
2
1
8
9
8
.
2
7
9
8
.
3
8
9
8
.
2
7
9
8
.
2
9
9
9
.
8
2
0
.
0
0
1
3
8
6
7
8
8
Mo
d
els
w
it
h
d
ee
p
au
to
-
en
co
d
er
p
r
e
-
tr
ain
in
g
p
r
o
ce
s
s
h
av
e
a
d
v
an
ta
g
es
i
n
o
v
er
co
m
i
n
g
t
h
e
i
m
b
alan
ce
p
r
o
b
lem
.
A
n
i
n
cr
ea
s
e
i
n
th
e
l
a
y
er
s
o
f
d
ee
p
au
to
-
e
n
co
d
er
n
et
w
o
r
k
r
ai
s
es
t
h
e
n
u
m
b
er
o
f
co
m
p
licated
f
ea
t
u
r
es
lear
n
ed
f
r
o
m
t
h
e
o
r
ig
in
al
d
ata.
T
r
an
s
f
er
-
la
y
er
w
it
h
4
d
ee
p
en
co
d
in
g
la
y
er
s
i
n
D
A
E
(
7
h
id
d
e
n
la
y
er
s
)
i
m
p
r
o
v
e
s
th
e
DNN
f
i
n
e
-
tu
n
i
n
g
p
r
o
ce
s
s
's
class
i
f
icatio
n
r
esu
lt
s
,
d
esp
ite
ex
tr
e
m
e
class
i
m
b
ala
n
ce
p
r
o
b
le
m
s
.
Ou
t
o
f
th
e
3
m
o
d
el
s
,
alth
o
u
g
h
i
n
t
h
e
w
eb
-
a
ttack
clas
s
w
h
er
e
th
e
tr
ai
n
i
n
g
d
ata
o
n
l
y
0
.
0
5
%
o
f
th
e
to
tal
d
ata,
th
e
s
e
n
s
i
tiv
it
y
v
alu
e
r
ea
ch
e
s
7
4
.
1
4
%,
esp
ec
i
all
y
in
th
o
s
e
t
h
at
u
s
e
th
e
f
o
ca
l
lo
s
s
f
u
n
ctio
n
.
Ho
w
e
v
er
,
in
t
h
e
w
eb
-
attac
k
class
,
th
e
tr
ain
i
n
g
d
ata
o
n
l
y
0
.
0
5
%
o
f
th
e
to
tal
d
ata.
I
n
th
e
m
o
d
el
th
at
u
s
e
s
cr
o
s
s
-
e
n
tr
o
p
y
,
th
e
r
esu
lts
ar
e
ac
ce
p
tab
le,
w
it
h
a
s
e
n
s
it
iv
i
t
y
r
ea
ch
i
n
g
al
m
o
s
t
6
8
%.
O
n
t
h
e
w
e
ig
h
ted
cr
o
s
s
-
e
n
tr
o
p
y
m
o
d
el,
th
e
s
e
n
s
it
i
v
it
y
is
o
n
l
y
4
1
.
3
8
%.
T
h
e
w
ei
g
h
ts
a
n
d
b
ias
i
n
t
h
e
D
NN
n
et
w
o
r
k
ar
e
i
n
itialized
ac
c
o
r
d
in
g
to
tr
an
s
f
er
r
ed
v
alu
e
f
r
o
m
t
h
e
e
n
co
d
in
g
la
y
er
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
Dee
p
lea
r
n
in
g
w
ith
fo
ca
l lo
s
s
a
p
p
r
o
a
c
h
fo
r
a
tta
ck
s
cla
s
s
ifica
tio
n
(
Yes
i No
va
r
ia
K
u
n
a
n
g
)
1415
o
n
DA
E
.
T
h
e
w
ei
g
h
ted
cr
o
s
s
-
en
tr
o
p
y
,
w
h
ich
u
s
es
th
e
w
ei
g
h
t
f
ac
to
r
p
r
o
p
o
r
tio
n
al
to
th
e
clas
s
f
r
eq
u
en
c
y
,
af
f
ec
t
s
th
e
tr
ain
in
g
t
h
at
is
u
n
s
u
itab
le
f
o
r
th
e
p
o
s
itiv
e
s
a
m
p
les at
t
h
e
b
eg
in
n
i
n
g
o
f
th
e
lear
n
in
g
p
r
o
ce
s
s
.
T
h
e
f
in
al
lo
s
s
la
y
er
o
f
th
e
d
ee
p
n
eu
r
al
n
et
w
o
r
k
i
s
th
e
s
o
f
t
m
a
x
lo
s
s
f
u
n
ctio
n
.
I
n
D
A
E
o
p
ti
m
izatio
n
,
th
e
w
ei
g
h
ts
ar
e
in
i
tialized
u
s
i
n
g
l
ec
u
n
_
u
n
i
f
o
r
m
.
A
s
a
r
e
s
u
l
t,
th
e
o
u
tp
u
t
f
r
o
m
ea
ch
la
y
er
i
s
u
n
if
o
r
m
l
y
d
is
tr
ib
u
ted
.
T
h
e
v
alu
e
o
f
th
e
o
u
tp
u
t
la
y
e
r
in
th
e
d
ee
p
n
eu
r
al
n
et
w
o
r
k
f
o
r
s
o
f
t
m
a
x
f
u
n
ctio
n
s
is
u
n
i
f
o
r
m
l
y
d
i
s
tr
ib
u
ted
.
A
p
p
r
o
p
r
iately
,
th
e
m
in
o
r
p
o
s
itiv
e
s
a
m
p
le
s
ar
e
le
s
s
cr
itica
l
in
t
h
e
i
n
it
ial
tr
ai
n
i
n
g
s
ta
g
e.
T
h
er
ef
o
r
e,
in
d
ee
p
lear
n
in
g
u
s
in
g
f
o
ca
l
lo
s
s
,
th
e
last
la
y
er
'
s
b
ias
ter
m
is
in
i
tial
ized
to
s
o
m
e
n
o
n
-
ze
r
o
v
al
u
e
[
3
3
]
.
Mo
r
e
f
o
cu
s
is
d
ir
ec
ted
at
th
e
p
o
s
iti
v
e
e
x
a
m
p
les
in
th
e
ea
r
l
y
tr
ain
i
n
g
s
tag
e,
an
d
th
e
w
h
o
le
tr
ain
in
g
p
r
o
ce
s
s
i
s
li
k
el
y
to
b
e
ef
f
ec
tiv
e.
I
n
d
ee
p
lear
n
i
n
g
,
cr
o
s
s
-
e
n
tr
o
p
y
lo
s
s
,
a
n
d
w
ei
g
h
ted
cr
o
s
s
-
en
tr
o
p
y
lo
s
s
w
a
s
b
ased
o
n
w
eig
h
t
in
itial
izatio
n
.
Di
f
f
er
en
t
lo
s
s
f
u
n
ctio
n
s
p
r
o
d
u
ce
d
v
ar
ied
p
r
ed
i
ctio
n
s
o
n
v
ar
io
u
s
m
o
d
el
s
.
I
n
3
0
ep
o
ch
s
,
th
e
m
o
d
el
u
s
ed
F
L
a
n
d
ac
h
ie
v
ed
s
li
g
h
t
l
y
b
etter
ac
cu
r
ac
y
th
a
n
C
E
lo
s
s
.
An
o
th
er
ad
v
a
n
ta
g
e
o
f
u
s
i
n
g
th
e
FL
f
u
n
ct
io
n
is
t
h
a
t
th
e
v
al
u
e
o
f
th
e
co
s
t
lo
s
s
is
n
ea
r
ze
r
o
.
T
h
e
lo
s
s
o
f
th
e
co
s
t
f
u
n
ctio
n
r
ep
r
esen
t
s
h
o
w
well
th
e
m
o
d
el
lear
n
s
co
n
ce
r
n
i
n
g
t
h
e
tr
ain
in
g
ex
a
m
p
les.
I
n
g
e
n
er
al,
w
it
h
o
u
t
m
o
d
i
f
y
in
g
th
e
d
i
s
tr
ib
u
tio
n
o
f
th
e
C
SE
-
C
I
C
-
I
DS2
0
1
8
d
ataset,
th
e
m
o
d
el
r
esu
lt
s
w
er
e
s
ati
s
f
ac
to
r
y
.
T
h
ese
r
esu
lts
w
er
e
b
etter
th
an
th
e
p
r
ev
io
u
s
s
t
u
d
y
u
s
i
n
g
d
ee
p
lear
n
i
n
g
a
n
d
r
esa
m
p
lin
g
tech
n
iq
u
es
i
n
th
e
s
a
m
e
d
ataset
,
as
s
h
o
w
n
in
T
ab
le
5
.
T
h
e
o
v
er
all
ac
cu
r
ac
y
o
b
tain
ed
at
9
8
.
2
7
%
w
as
b
etter
th
an
th
e
d
ee
p
lear
n
i
n
g
m
o
d
el
d
ev
el
o
p
ed
b
y
L
in
et
a
l
.
[
2
6
]
.
A
n
ac
cu
r
ac
y
o
f
9
6
.
2
%
w
a
s
ac
h
ie
v
e
d
b
y
L
i
n
et
a
l
.
[
2
6
]
,
u
s
i
n
g
SMOT
E
alg
o
r
ith
m
f
o
r
i
m
b
alan
ce
cla
s
s
,
a
n
d
DNN
m
o
d
el
b
ased
o
n
lo
n
g
s
h
o
r
t
ter
m
m
e
m
o
r
y
(
L
ST
M)
b
ased
an
d
,
c
o
m
b
i
n
ed
w
it
h
atte
n
tio
n
m
ec
h
a
n
is
m
(
A
M)
.
A
l
s
o
,
th
eir
s
tu
d
y
co
m
p
ar
ed
th
e
v
ar
io
u
s
m
ac
h
i
n
e
lear
n
i
n
g
tech
n
iq
u
es
w
it
h
d
ataset
s
t
h
at
h
av
e
b
ee
n
co
n
d
u
cted
o
v
er
-
s
a
m
p
lin
g
.
A
ll
o
f
th
e
m
u
ltip
le
m
eth
o
d
s
p
er
f
o
r
m
a
n
ce
s
ar
e
n
o
t
s
ig
n
i
f
ica
n
t
co
m
p
ar
ed
to
th
e
p
r
o
p
o
s
ed
m
o
d
el
in
t
h
i
s
s
t
u
d
y
.
Fo
r
in
s
ta
n
ce
,
f
o
r
w
e
b
attac
k
clas
s
es
as
m
i
n
o
r
it
y
s
a
m
p
le
s
,
th
e
y
clai
m
e
d
th
e
m
o
d
els
d
ev
e
lo
p
ed
r
ea
ch
ed
9
8
%
f
o
r
a
b
etter
r
ec
all
v
alu
e
th
an
th
e
m
o
d
el
in
th
is
s
tu
d
y
,
w
h
ic
h
is
o
n
l
y
7
5
.
2
9
%.
Ho
w
ev
er
,
th
e
p
r
ec
is
io
n
v
alu
e
an
d
F1
-
s
co
r
e
f
o
r
th
e
w
eb
attac
k
class
i
s
o
n
l
y
3
0
%.
I
n
co
n
tr
ast,
th
e
m
o
d
el
p
r
ec
is
io
n
a
n
d
F1
-
Sco
r
e
f
o
r
w
eb
a
ttack
cla
s
s
es
w
it
h
f
o
ca
l
lo
s
s
i
n
t
h
is
s
tu
d
y
is
9
7
.
7
6
%
an
d
8
5
.
0
7
%,
r
esp
ec
tiv
ely
,
a
f
te
r
2
0
0
ep
o
ch
tr
ain
in
g
.
T
h
is
s
t
u
d
y
is
s
u
p
er
io
r
to
t
h
e
p
r
ec
is
io
n
v
al
u
e
a
n
d
F1
-
s
co
r
e
c
o
m
p
ar
ed
to
t
h
e
p
er
f
o
r
m
an
ce
s
o
f
p
r
ev
io
u
s
r
esear
ch
,
as
d
etailed
in
T
ab
le
5
.
Hu
a
[
2
9
]
u
s
ed
v
ar
io
u
s
m
ac
h
in
e
an
d
d
ee
p
lear
n
in
g
,
as
w
el
l
as
u
n
d
er
-
s
a
m
p
lin
g
an
d
f
ea
t
u
r
e
s
elec
tio
n
tec
h
n
iq
u
es
f
o
r
p
r
e
-
p
r
o
ce
s
s
in
g
.
T
h
e
r
ec
all
v
alu
e
o
f
th
e
m
o
d
el
w
it
h
L
i
g
h
t
GB
M
w
a
s
ar
o
u
n
d
0
.
1
%,
s
lig
h
tl
y
h
i
g
h
er
t
h
a
n
th
e
m
o
d
el
p
r
o
p
o
s
ed
in
th
i
s
s
t
u
d
y
.
Ho
w
e
v
er
,
th
e
p
r
ec
is
io
n
v
al
u
e
o
f
t
h
is
m
o
d
el
i
s
0
.
2
4
%
s
u
p
er
io
r
to
th
e
m
o
d
el
th
e
y
p
r
o
p
o
s
ed
.
Un
f
o
r
tu
n
atel
y
,
t
h
eir
r
esear
ch
d
id
n
o
t
ex
p
lain
th
e
ef
f
ec
t
o
f
th
e
m
o
d
el
o
n
w
eb
attac
k
s
a
s
m
i
n
o
r
it
y
c
lass
es.
Z
h
ao
et
a
l
.,
[
3
9
]
w
it
h
th
e
s
e
m
i
-
s
u
p
er
v
is
ed
d
is
cr
i
m
i
n
an
t
au
to
-
e
n
co
d
er
(
SS
D
A
)
,
an
d
Fer
r
ag
et
a
l
.,
[
4
8
]
w
ith
Dee
p
A
u
to
-
e
n
co
d
er
,
u
tili
ze
d
u
n
s
u
p
er
v
is
ed
lear
n
i
n
g
w
it
h
o
u
t
m
o
d
i
f
y
in
g
th
e
d
ata
d
is
tr
ib
u
tio
n
o
n
t
h
e
d
atase
t.
Ho
w
e
v
er
,
w
it
h
t
h
e
h
y
p
er
-
p
a
r
a
m
eter
p
r
o
ce
s
s
p
er
f
o
r
m
ed
i
n
th
is
s
t
u
d
y
,
t
h
e
d
ee
p
lear
n
in
g
m
o
d
e
l
p
r
o
p
o
s
ed
r
esu
lted
in
b
etter
d
etec
tio
n
.
T
h
is
m
o
d
el
w
o
r
k
s
b
etter
w
i
th
les
s
d
ata
p
r
o
p
o
r
tio
n
an
d
d
ee
p
lear
n
in
g
u
s
i
n
g
f
o
ca
l lo
s
s
.
A
s
a
r
es
u
lt,
it
s
o
lv
es
m
u
ch
o
f
th
e
i
m
b
ala
n
ce
-
clas
s
p
r
o
b
lem
.
T
h
is
s
t
u
d
y
h
as
s
u
cc
ess
f
u
ll
y
d
e
m
o
n
s
tr
ated
th
e
s
i
g
n
if
ican
ce
o
f
d
ee
p
lea
r
n
i
n
g
.
T
h
is
h
a
s
b
ee
n
ac
h
iev
ed
u
s
i
n
g
d
ee
p
au
to
-
e
n
co
d
er
as
a
f
ea
tu
r
e
r
ed
u
ctio
n
tec
h
n
iq
u
e
w
i
t
h
f
o
ca
l
lo
s
s
f
u
n
ctio
n
s
.
I
t
h
a
s
p
r
o
v
id
ed
b
etter
r
esu
lts
in
ter
m
s
o
f
s
e
v
er
al
p
er
f
o
r
m
a
n
ce
m
etr
ics
f
o
r
I
DS,
esp
ec
iall
y
in
i
m
b
ala
n
ce
clas
s
es.
Ho
w
e
v
er
,
th
er
e
h
av
e
b
ee
n
ce
r
t
ain
li
m
ita
tio
n
s
an
d
co
n
s
tr
ain
ts
in
t
h
is
s
t
u
d
y
.
I
n
th
e
ev
a
l
u
atio
n
p
r
o
ce
s
s
,
th
e
in
f
il
tr
atio
n
attac
k
c
lass
in
t
h
e
in
itial
te
s
t w
as
eli
m
i
n
ated
w
i
th
o
u
t m
o
d
if
y
i
n
g
t
h
e
d
ataset.
T
h
is
is
b
ec
au
s
e
it
o
f
te
n
ca
u
s
ed
m
i
s
class
if
ica
tio
n
w
it
h
b
en
ig
n
class
.
Ho
w
ev
er
,
f
u
t
u
r
e
s
tu
d
ies
s
h
o
u
ld
d
ev
elo
p
a
d
ee
p
lear
n
in
g
m
o
d
el
w
it
h
a
t
w
o
-
s
tag
e
clas
s
i
f
icatio
n
th
at
d
etec
ts
in
f
il
tr
atio
n
attac
k
s
.
A
ls
o
,
it
is
b
eliev
ed
th
at
t
h
e
f
o
ca
l
lo
s
s
f
u
n
ctio
n
o
p
ti
m
izes
t
h
e
i
m
b
ala
n
ce
d
class
p
r
o
b
lem
u
s
i
n
g
o
th
er
d
ee
p
lear
n
in
g
a
lg
o
r
it
h
m
s
.
S
u
b
s
eq
u
e
n
tl
y
,
f
u
tu
r
e
s
t
u
d
ies
s
h
o
u
ld
tr
y
to
e
v
alu
ate
t
h
e
i
n
f
l
u
en
ce
o
f
th
e
f
o
ca
l
lo
s
s
f
u
n
ctio
n
w
it
h
d
if
f
er
en
t
d
ee
p
lear
n
in
g
alg
o
r
ith
m
s
.
T
h
is
s
tu
d
y
u
s
ed
th
e
C
SE
-
C
I
C
-
I
D
S2
0
1
8
d
ataset
in
th
e
tr
ai
n
i
n
g
a
n
d
test
in
g
p
r
o
ce
s
s
e
s
.
Fu
t
u
r
e
r
esear
ch
s
h
o
u
ld
co
v
er
an
an
o
m
al
y
-
b
a
s
ed
o
n
lin
e
in
tr
u
s
io
n
d
etec
tio
n
s
y
s
te
m
.
T
ab
le
5
.
C
o
m
p
ar
is
o
n
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
an
d
p
r
ev
io
u
s
s
t
u
d
ies i
n
C
SE
-
C
I
C
-
I
DS2
0
1
8
d
ataset
R
e
f
e
r
e
n
c
e
H
a
n
d
l
i
n
g
I
mb
a
l
a
n
c
e
d
C
l
a
ss
A
p
p
r
o
a
c
h
C
l
a
ssi
f
i
e
r
A
c
c
u
r
a
c
y
(
%)
P
r
e
c
i
si
o
n
(
%)
R
e
c
a
l
l
(
%)
F1
-
S
c
o
r
e
(
%)
[
2
6
]
S
M
O
T
E
M
L
P
L
S
T
M
L
S
T
M
+
A
M
90
93
9
6
.
2
91
91
96
89
93
96
-
-
-
[
2
9
]
U
n
d
e
r
-
samp
l
i
n
g
L
i
g
h
t
G
B
M
M
L
P
C
N
N
9
8
.
3
7
9
7
.
5
8
9
8
.
0
6
9
8
.
1
4
9
7
.
2
3
9
7
.
6
0
9
8
.
3
7
9
7
.
5
8
9
8
.
0
6
9
8
.
2
1
9
7
.
3
7
9
7
.
6
7
[
3
9
]
O
r
i
g
i
n
a
l
D
i
st
r
i
b
u
t
i
o
n
S
S
D
A
9
7
.
0
9
-
9
8
.
0
0
-
[
4
8
]
O
r
i
g
i
n
a
l
D
i
st
r
i
b
u
t
i
o
n
C
N
N
DA
9
7
.
3
8
9
7
.
3
7
-
-
9
7
.
2
8
9
8
.
1
8
-
-
O
u
r
M
e
t
h
o
d
O
r
i
g
i
n
a
l
D
i
st
r
i
b
u
t
i
o
n
D
A
E+
D
N
N
(
f
o
c
a
l
l
o
ss)
9
8
.
2
7
9
8
.
3
8
9
8
.
2
7
9
8
.
2
9
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1693
-
6930
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
,
Vo
l.
19
,
No
.
4
,
A
u
g
u
s
t 2
0
2
1
:
1
4
0
7
-
1
4
1
8
1416
5.
CO
NCLU
SI
O
N
T
h
is
s
tu
d
y
p
r
esen
ted
a
n
e
w
d
ee
p
lear
n
in
g
m
o
d
el
to
ad
d
r
es
s
th
e
p
r
o
b
lem
o
f
clas
s
if
y
i
n
g
m
u
lti
-
clas
s
attac
k
s
.
T
h
e
n
et
w
o
r
k
ar
ch
itec
t
u
r
e
w
as
p
ar
titi
o
n
ed
i
n
to
au
to
m
atic
f
ea
t
u
r
e
ex
tr
ac
tio
n
w
it
h
d
ee
p
au
to
en
co
d
er
u
s
i
n
g
7
h
id
d
en
la
y
er
s
,
a
n
d
a
class
i
f
ier
w
it
h
a
f
u
l
l
y
co
n
n
ec
ted
d
ee
p
n
eu
r
al
n
e
t
w
o
r
k
.
T
h
e
f
o
ca
l
lo
s
s
f
u
n
ctio
n
w
a
s
ad
j
u
s
ted
to
th
e
p
r
o
p
o
s
ed
m
o
d
el
in
an
i
m
b
ala
n
ce
d
d
ataset.
T
h
e
p
r
o
p
o
s
ed
d
ee
p
lear
n
in
g
m
o
d
el
u
s
ed
t
h
e
f
o
ca
l
lo
s
s
to
o
b
tain
a
f
aster
co
n
v
er
g
en
ce
th
a
n
cr
o
s
s
-
e
n
tr
o
p
y
lo
s
s
a
n
d
w
ei
g
h
ted
lo
s
s
.
C
o
n
ce
r
n
in
g
w
eb
attac
k
cla
s
s
e
s
as m
i
n
o
r
it
y
s
a
m
p
le
s
,
th
e
e
v
al
u
atio
n
r
esu
lts
o
f
th
e
C
SE
-
C
I
C
-
I
DS2
0
1
8
s
h
o
w
t
h
at
t
h
e
d
ee
p
lear
n
in
g
m
eth
o
d
w
it
h
f
o
ca
l
lo
s
s
i
s
a
h
ig
h
-
q
u
ali
t
y
c
lass
i
f
ier
w
it
h
9
8
.
3
8
%
p
r
ec
is
io
n
,
9
8
.
2
7
%
s
en
s
iti
v
it
y
,
an
d
9
9
.
8
2
%
s
p
ec
if
icit
y
.
Sev
er
al
f
u
tu
r
e
s
t
u
d
ies
s
h
o
u
ld
b
e
b
u
ilt
o
n
t
h
i
s
r
esear
c
h
i
n
s
e
v
er
al
asp
ec
ts
.
Firs
t,
u
s
i
n
g
th
e
f
o
ca
l
-
lo
s
s
f
u
n
ctio
n
o
n
i
m
b
alan
ce
d
d
atase
ts
s
h
o
u
ld
b
e
ev
al
u
ated
b
y
co
m
p
ar
i
n
g
th
e
m
w
it
h
v
ar
io
u
s
d
atasets
.
I
n
th
i
s
r
esear
ch
,
t
h
e
in
f
iltra
tio
n
at
tack
cla
s
s
w
a
s
eli
m
i
n
ated
,
w
h
ic
h
b
eh
a
v
es
i
n
t
h
e
s
a
m
e
w
a
y
a
s
b
en
i
g
n
tr
a
f
f
i
c.
Ho
w
ev
er
,
f
u
t
u
r
e
s
tu
d
ie
s
s
h
o
u
ld
i
m
p
r
o
v
e
a
d
ee
p
lear
n
in
g
m
o
d
el
t
h
at
u
s
es t
w
o
s
tag
e
s
to
f
ilter
t
h
e
in
f
il
tr
atio
n
attac
k
s
.
ACK
NO
WL
E
D
G
E
M
E
NT
S
T
h
is
r
esear
ch
h
as
r
ec
eiv
e
d
f
u
n
d
in
g
f
r
o
m
I
n
d
o
n
esia
Mi
n
is
tr
y
o
f
R
esear
c
h
,
T
ec
h
n
o
lo
g
y
,
a
n
d
Hig
h
er
E
d
u
ca
tio
n
(
g
r
a
n
t
a
g
r
ee
m
e
n
t
1
7
0
/SP
2
H/L
T
/DR
P
M/2
0
2
0
)
.
T
h
e
au
t
h
o
r
s
w
o
u
ld
li
k
e
to
t
h
an
k
d
ata
s
cien
ce
r
e
s
ea
r
ch
g
r
o
u
p
,
Un
i
v
er
s
ita
s
B
in
a
Dar
m
a
,
an
d
Un
i
v
er
s
it
y
f
o
r
s
u
p
p
o
r
t a
n
d
f
ac
ili
ties
.
RE
F
E
R
E
NC
E
S
[1
]
F
.
L
i,
R.
X
ie,
Z
.
W
a
n
g
,
L
.
G
u
o
,
J.
Ye
,
P
.
M
a
,
a
n
d
W
.
S
o
n
g
,
“
On
li
n
e
Distri
b
u
ted
Io
T
S
e
c
u
rit
y
M
o
n
i
to
ri
n
g
w
it
h
M
u
lt
i
d
im
e
n
sio
n
a
l
S
trea
m
in
g
Big
Da
ta,”
IEE
E
In
ter
n
e
t
T
h
i
n
g
s
J
o
u
rn
a
l
,
v
o
l.
7
,
n
o
.
5
,
p
p
.
4
3
8
7
-
4
3
9
4
,
M
a
y
20
20
,
d
o
i:
1
0
.
1
1
0
9
/JIOT
.
2
0
1
9
.
2
9
6
2
7
8
8
.
[2
]
K.
R.
S
o
l
li
n
s,
“
I
o
T
Big
Da
ta
S
e
c
u
rit
y
a
n
d
P
riv
a
c
y
V
e
rsu
s
In
n
o
v
a
ti
o
n
,
”
IEE
E
In
ter
n
e
t
T
h
i
n
g
s
J
.
,
v
o
l.
6
,
n
o
.
2
,
p
p
.
1
6
2
8
-
1
6
3
5
,
A
p
r.
2
0
1
9
,
d
o
i
:
1
0
.
1
1
0
9
/JIOT
.
2
0
1
9
.
2
8
9
8
1
1
3
.
[3
]
D.
S
ti
a
w
a
n
,
M
o
h
d
.
Y.
I
d
ris,
R.
F
.
M
a
li
k
,
S
.
Nu
rm
a
in
i,
N.
A
lsh
a
ri
f
,
a
n
d
R.
Bu
d
iarto
,
“
In
v
e
stig
a
ti
n
g
Bru
te
F
o
rc
e
A
tt
a
c
k
P
a
tt
e
r
n
s
in
I
o
T
Ne
tw
o
rk
,
”
J
o
u
r
n
a
l
o
f
El
e
c
trica
l
a
n
d
C
o
mp
u
ter
En
g
in
e
e
rin
g
,
v
o
l.
2
0
1
9
,
p
p
.
1
–
1
3
,
A
p
r.
2
0
1
9
,
d
o
i:
1
0
.
1
1
5
5
/2
0
1
9
/
4
5
6
8
3
6
8
.
[4
]
K.
Yo
sh
ig
o
e
,
W
.
Da
i,
M
.
A
b
ra
m
so
n
,
a
n
d
A
.
Ja
c
o
b
s,
“
Ov
e
rc
o
m
in
g
in
v
a
sio
n
o
f
p
riv
a
c
y
in
sm
a
rt
h
o
m
e
e
n
v
iro
n
m
e
n
t
w
it
h
s
y
n
th
e
ti
c
p
a
c
k
e
t
in
jec
t
io
n
,
”
2
0
1
5
T
RON
S
y
mp
o
siu
m
(
T
RONS
HO
W
)
,
De
c
.
2
0
1
5
,
p
p
.
1
-
7
,
d
o
i:
1
0
.
1
1
0
9
/T
RON
S
HO
W
.
2
0
1
4
.
7
3
9
6
8
7
5
.
[5
]
B.
Ya
n
g
,
L
.
G
u
o
,
F
.
L
i,
J.
Ye
,
a
n
d
W
.
S
o
n
g
,
“
V
u
ln
e
ra
b
il
i
ty
a
ss
e
ss
m
e
n
ts
o
f
e
lec
tri
c
d
riv
e
s
y
ste
m
s
d
u
e
to
se
n
so
r
d
a
ta
i
n
teg
rit
y
a
tt
a
c
k
s
,
”
IEE
E
T
ra
n
s.
I
n
d
.
I
n
f.
,
v
o
l.
1
6
,
n
o
.
5
,
p
p
.
3
3
0
1
-
3
3
1
0
,
M
a
y
2
0
2
0
,
d
o
i:
1
0
.
1
1
0
9
/T
II.
2
0
1
9
.
2
9
4
8
0
5
6
.
[6
]
M
.
M
.
Ha
ss
a
n
,
A
.
G
u
m
a
e
i,
S
.
Hu
d
a
,
a
n
d
A
.
A
l
m
o
g
re
n
,
“
In
c
re
a
sin
g
th
e
tru
stw
o
rth
in
e
ss
in
t
h
e
in
d
u
stri
a
l
Io
T
n
e
t
w
o
rk
s
th
ro
u
g
h
a
re
li
a
b
le
c
y
b
e
ra
tt
a
c
k
d
e
tec
ti
o
n
m
o
d
e
l
,
”
IEE
E
T
r
a
n
s.
I
n
d
.
In
f.
,
v
o
l
.
1
6
,
n
o
.
9
,
p
p
.
6
1
5
4
-
6
1
6
2
,
S
e
p
.
2
0
2
0
,
d
o
i:
1
0
.
1
1
0
9
/T
II.
2
0
2
0
.
2
9
7
0
0
7
4
.
[7
]
N.
Ch
a
a
b
o
u
n
i,
M
.
M
o
sb
a
h
,
A
.
Zem
m
a
ri,
C.
S
a
u
v
ig
n
a
c
,
a
n
d
P
.
F
a
ru
k
i,
“
Ne
t
w
o
rk
in
tru
sio
n
d
e
tec
ti
o
n
f
o
r
Io
T
se
c
u
rit
y
b
a
se
d
o
n
lea
rn
i
n
g
tec
h
n
iq
u
e
s,”
IE
EE
Co
mm
u
n
ica
t
io
n
s
S
u
rv
e
y
s &
T
u
to
ri
a
ls
,
v
o
l
.
2
1
,
n
o
.
3
,
p
p
.
2
6
7
1
-
2
7
0
1
,
Ja
n
.
2
0
1
9
,
d
o
i:
1
0
.
1
1
0
9
/COM
S
T
.
2
0
1
9
.
2
8
9
6
3
8
0
.
[8
]
J.
A
.
Ju
p
in
,
T
.
S
u
ti
k
n
o
,
M
.
A
.
Ism
a
il
,
M
.
S
.
M
o
h
a
m
a
d
,
S
.
Ka
sim
,
a
n
d
D.
S
ti
a
w
a
n
,
“
Re
v
ie
w
o
f
th
e
m
a
c
h
in
e
lea
rn
in
g
m
e
th
o
d
s
in
t
h
e
c
las
sif
i
c
a
ti
o
n
o
f
p
h
ish
i
n
g
a
tt
a
c
k
,
”
Bu
ll
e
ti
n
o
f
El
e
c
t
ric
a
l
E
n
g
i
n
e
e
rin
g
a
n
d
I
n
fo
rm
a
ti
c
s
,
v
o
l.
8
,
n
o
.
4
,
p
p
.
1
5
4
5
-
1
5
5
5
,
De
c
.
2
0
1
9
,
d
o
i
:
1
0
.
1
1
5
9
1
/ee
i.
v
8
i
4
.
1
3
4
4
.
[9
]
R.
V
in
a
y
a
k
u
m
a
r,
M
.
A
laz
a
b
,
K.
P
.
S
o
m
a
n
,
P
.
P
o
o
rn
a
c
h
a
n
d
ra
n
,
A
.
A
l
-
Ne
m
ra
t,
a
n
d
S
.
V
e
n
k
a
tra
m
a
n
,
“
De
e
p
L
e
a
rn
in
g
A
p
p
ro
a
c
h
f
o
r
In
telli
g
e
n
t
In
tr
u
si
o
n
De
tec
ti
o
n
S
y
ste
m
,
”
IEE
E
Acc
e
ss
,
v
o
l.
7
,
p
p
.
4
1
5
2
5
-
4
1
5
5
0
,
A
p
r.
2
0
1
9
,
d
o
i:
1
0
.
1
1
0
9
/A
CCES
S
.
2
0
1
9
.
2
8
9
5
3
3
4
.
[1
0
]
J.
M
a
ji
d
p
o
u
r
a
n
d
H.
Ha
sa
n
z
a
d
e
h
,
“
A
p
p
li
c
a
ti
o
n
o
f
d
e
e
p
lea
rn
in
g
t
o
e
n
h
a
n
c
e
th
e
a
c
c
u
ra
c
y
o
f
in
tru
si
o
n
d
e
tec
ti
o
n
in
m
o
d
e
rn
c
o
m
p
u
ter
n
e
tw
o
rk
s,”
Bu
ll
e
ti
n
o
f
El
e
c
trica
l
E
n
g
in
e
e
rin
g
a
n
d
In
f
o
rm
a
ti
c
s
,
v
o
l
.
9
,
n
o
.
3
,
p
p
.
1
1
3
7
–
1
1
4
8
,
J
u
n
.
2
0
2
0
,
d
o
i:
1
0
.
1
1
5
9
1
/ee
i.
v
9
i3
.
1
7
2
4
.
[
1
1
]
M
.
M
.
N
a
j
a
f
a
b
a
d
i
,
F
.
V
i
l
l
a
n
u
s
t
r
e
,
T
.
M
.
K
h
o
s
h
g
o
f
t
a
a
r
,
N
.
S
e
l
i
y
a
,
R
.
W
a
l
d
,
a
n
d
E
.
M
u
h
a
r
e
m
a
g
i
c
,
“
D
e
e
p
l
e
a
r
n
i
n
g
a
p
p
l
i
c
a
t
i
o
n
s
a
n
d
c
h
a
l
l
e
n
g
e
s
i
n
b
i
g
d
a
t
a
a
n
a
l
y
t
i
c
s
,
”
J
o
u
r
n
a
l
o
f
B
i
g
D
a
t
a
,
v
o
l
.
2
,
n
o
.
1
,
D
e
c
.
2
0
1
5
,
d
o
i
:
1
0
.
1
1
8
6
/
s
4
0
5
3
7
-
014
-
0007
-
7.
[1
2
]
Y.
L
e
Cu
n
,
Y.
Be
n
g
io
,
a
n
d
G
.
Hin
to
n
,
“
De
e
p
lea
rn
in
g
,
”
Na
t
u
re
,
v
o
l.
5
2
1
,
n
o
.
7
5
5
3
,
p
p
.
4
3
6
-
4
4
4
,
M
a
y
2
0
1
5
,
d
o
i:
1
0
.
1
0
3
8
/n
a
t
u
re
1
4
5
3
9
.
[1
3
]
K.
No
d
a
,
Y.
Ya
m
a
g
u
c
h
i,
K.
Na
k
a
d
a
i,
H.
G
.
Ok
u
n
o
,
a
n
d
T
.
Og
a
ta,
“
A
u
d
io
-
v
isu
a
l
s
p
e
e
c
h
re
c
o
g
n
i
ti
o
n
u
si
n
g
d
e
e
p
lea
rn
in
g
,
”
Ap
p
li
e
d
I
n
telli
g
e
n
c
e
,
v
o
l.
4
2
,
n
o
.
4
,
p
p
.
7
2
2
–
7
3
7
,
Ju
n
.
2
0
1
5
,
d
o
i:
1
0
.
1
0
0
7
/s1
0
4
8
9
-
0
1
4
-
0
6
2
9
-
7.
[1
4
]
R.
K.
M
e
lep
p
a
t
,
C.
S
h
e
a
rw
o
o
d
,
S
.
L
.
Ke
e
y
,
a
n
d
M
.
V.
M
a
t
h
a
m
,
“
Q
u
a
n
ti
tativ
e
o
p
ti
c
a
l
c
o
h
e
re
n
c
e
m
icr
o
sc
o
p
y
f
o
r
th
e
in
sit
u
in
v
e
stig
a
ti
o
n
o
f
th
e
b
io
f
il
m
,
”
J
.
Bi
o
me
d
.
Op
t
,
v
o
l.
2
1
,
n
o
.
1
2
,
p
.
1
2
7
0
0
2
,
De
c
.
2
0
1
6
,
d
o
i:
1
0
.
1
1
1
7
/1
.
JBO
.
2
1
.
1
2
.
1
2
7
0
0
2
.
[1
5
]
K.
M
.
Ra
th
e
e
sh
,
L
.
K.
S
e
a
h
,
a
n
d
V
.
M
.
M
u
r
u
k
e
sh
a
n
,
“
S
p
e
c
tral
p
h
a
se
-
b
a
se
d
a
u
to
m
a
ti
c
c
a
li
b
ra
ti
o
n
sc
h
e
m
e
f
o
r
s
we
p
t
so
u
rc
e
-
b
a
se
d
o
p
t
ica
l
c
o
h
e
re
n
c
e
to
m
o
g
r
a
p
h
y
s
y
ste
m
s,”
Ph
y
s.
M
e
d
.
Bi
o
l.
,
v
o
l
.
6
1
,
n
o
.
2
1
,
p
p
.
7
6
5
2
-
7
6
6
3
,
N
o
v
.
2
0
1
6
,
d
o
i:
1
0
.
1
0
8
8
/0
0
3
1
-
9
1
5
5
/
6
1
/
2
1
/
7
6
5
2
.
Evaluation Warning : The document was created with Spire.PDF for Python.