I
nte
rna
t
io
na
l J
o
urna
l o
f
E
lect
rica
l a
nd
Co
m
pu
t
er
E
ng
ineering
(
I
J
E
CE
)
Vo
l.
11
,
No
.
2
,
A
p
r
il
2
0
2
1
,
p
p
.
1
4
9
8
~1
5
0
9
I
SS
N:
2
0
8
8
-
8
7
0
8
,
DOI
: 1
0
.
1
1
5
9
1
/
ijece
.
v
1
1
i
2
.
p
p
1
4
9
8
-
1
5
0
9
1498
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
Forg
ing
a deep le
a
rning
neural ne
t
wo
rk int
r
usio
n de
tect
io
n
framewo
rk
to cur
b t
he
distrib
uted
denia
l of servi
ce
attack
Arno
ld Adim
a
bu
a
O
j
ug
o
1
,
Rum
e
E
liza
bet
h Yo
ro
2
1
De
p
a
rtme
n
t
o
f
Co
m
p
u
ter
S
c
ien
c
e
,
F
e
d
e
ra
l
Un
i
v
e
rsity
o
f
P
e
tr
o
leu
m
Re
so
u
rc
e
s E
ffu
ru
n
,
Warri,
Nig
e
ria
2
De
p
a
rtme
n
t
o
f
Co
m
p
u
ter
S
c
ien
c
e
,
De
lt
a
S
tate
P
o
l
y
tec
h
n
ic O
g
wa
sh
i
-
Uk
u
,
Nig
e
ria
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ap
r
2
2
,
2
0
2
0
R
ev
is
ed
May
2
8
,
2
0
2
0
Acc
ep
ted
J
u
l 1
0
,
2
0
2
0
To
d
a
y
’s
p
o
p
u
lari
ty
o
f
th
e
i
n
tern
e
t
h
a
s
sin
c
e
p
ro
v
e
n
a
n
e
ffe
c
ti
v
e
a
n
d
e
fficie
n
t
m
e
a
n
s
o
f
i
n
fo
rm
a
ti
o
n
s
h
a
rin
g
.
H
o
we
v
e
r,
t
h
is
h
a
s
c
o
n
se
q
u
e
n
tl
y
a
d
v
a
n
c
e
d
t
h
e
p
ro
li
fe
ra
ti
o
n
o
f
a
d
v
e
rsa
ries
wh
o
a
im
a
t
u
n
a
u
th
o
rize
d
a
c
c
e
ss
to
i
n
fo
rm
a
ti
o
n
b
e
in
g
s
h
a
re
d
o
v
e
r
t
h
e
i
n
tern
e
t
m
e
d
iu
m
.
T
h
e
se
a
re
a
c
h
iev
e
d
v
ia v
a
rio
u
s m
e
a
n
s
o
n
e
o
f
wh
ic
h
is
th
e
d
istri
b
u
ted
d
e
n
ial
o
f
se
rv
ice
a
tt
a
c
k
s
-
wh
ich
h
a
s
b
e
c
o
m
e
a
m
a
jo
r
th
re
a
t
t
o
t
h
e
e
lec
tro
n
ic
s
o
c
iety
.
T
h
e
se
a
re
c
a
re
fu
ll
y
c
ra
fted
a
tt
a
c
k
s
o
f
larg
e
m
a
g
n
i
tu
d
e
t
h
a
t
p
o
ss
e
ss
t
h
e
c
a
p
a
b
il
it
y
to
wre
a
k
h
a
v
o
c
a
t
v
e
ry
h
i
g
h
lev
e
ls
a
n
d
n
a
ti
o
n
a
l
in
fra
stru
c
t
u
re
s.
Th
is
stu
d
y
p
o
sits
in
tell
ig
e
n
t
sy
ste
m
s
v
ia
th
e
u
se
o
f
m
a
c
h
in
e
lea
rn
i
n
g
fra
m
e
wo
rk
s
to
d
e
tec
t
su
c
h
.
We
e
m
p
lo
y
th
e
d
e
e
p
lea
rn
in
g
a
p
p
ro
a
c
h
to
d
ist
in
g
u
is
h
b
e
twe
e
n
b
e
n
ig
n
e
x
c
h
a
n
g
e
o
f
d
a
ta
a
n
d
m
a
li
c
io
u
s
a
tt
a
c
k
s
fro
m
d
a
ta
traffi
c
.
Re
su
lt
s
sh
o
ws
c
o
n
se
q
u
e
n
t
su
c
c
e
ss
in
th
e
e
m
p
lo
y
m
e
n
t
o
f
d
e
e
p
lea
rn
i
n
g
n
e
u
ra
l
n
e
two
r
k
to
e
ffe
c
ti
v
e
l
y
d
iffere
n
ti
a
te
b
e
twe
e
n
a
c
c
e
p
tab
le
a
n
d
n
o
n
-
a
c
c
e
p
tab
le
d
a
ta
p
a
c
k
e
ts
(i
n
tru
si
o
n
)
o
n
a
n
e
two
rk
d
a
ta t
ra
ffic.
K
ey
w
o
r
d
s
:
Data
s
ec
u
r
ity
DDo
S
Dee
p
n
eu
r
al
n
etwo
r
k
I
n
tr
u
s
io
n
d
etec
tio
n
s
y
s
tem
Ma
ch
in
e
lear
n
in
g
Sp
am
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
:
Ar
n
o
ld
Ad
im
a
b
u
a
Oju
g
o
Dep
ar
tm
en
t o
f
C
o
m
p
u
ter
Scie
n
ce
Fed
er
al
Un
iv
er
s
ity
o
f
Petr
o
le
u
m
R
eso
u
r
ce
s
E
f
f
u
r
u
n
P.M
.
B
1
2
2
1
,
E
f
f
u
r
u
n
,
W
ar
r
i,
Delta
State,
Nig
er
ia
E
m
ail:
o
ju
g
o
.
a
r
n
o
ld
@
f
u
p
r
e.
ed
u
.
n
g
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
r
ap
id
ad
v
an
ce
m
en
t in
tec
h
n
o
lo
g
y
o
v
e
r
th
e
y
ea
r
s
-
h
as b
e
en
g
ea
r
ed
to
war
d
s
ef
f
ec
tiv
ely
im
p
r
o
v
in
g
th
e
way
an
d
h
o
w
we
liv
e
in
a
b
id
to
m
ee
ts
s
p
ec
if
ic
tar
g
eted
h
u
m
an
n
ee
d
s
.
T
ec
h
n
o
lo
g
y
s
ee
k
s
to
b
etter
ad
v
a
n
c
e
o
u
r
liv
in
g
ex
is
ten
ce
u
n
to
h
ig
h
er
p
lain
s
cu
m
lev
els
o
f
s
o
p
h
is
ticatio
n
with
ea
s
e.
T
r
em
en
d
o
u
s
ad
o
p
tio
n
a
nd
in
teg
r
atio
n
o
f
t
h
e
I
n
ter
n
et
h
as
s
ig
n
if
ican
tly
ad
v
an
ce
d
th
e
u
s
e
o
f
d
ata
s
h
ar
in
g
p
r
o
g
r
am
s
th
at
s
ee
k
s
to
ef
f
ec
tiv
ely
d
is
s
em
in
ate
d
ata
f
r
o
o
m
o
n
e
u
s
er
to
a
n
o
th
e
r
[
1
]
.
T
h
is
ad
o
p
tio
n
an
d
in
teg
r
atio
n
h
as
b
ee
n
attr
i
b
u
te
d
to
its
u
s
ag
e
ea
s
e,
u
b
iq
u
ity
o
f
its
n
atu
r
e
,
lo
w
-
co
s
t
o
f
tr
an
s
a
ctio
n
an
d
tr
u
s
t
in
co
m
m
u
n
ica
tio
n
ch
an
n
el
-
all
o
f
w
h
i
c
h
c
o
n
t
i
n
u
e
s
t
o
a
d
v
a
n
c
e
i
t
s
p
o
p
u
l
a
r
i
t
y
,
a
d
o
p
t
i
o
n
e
a
s
e
a
n
d
u
s
a
g
e
.
T
h
i
s
g
r
o
w
t
h
h
a
s
e
q
u
a
l
l
y
a
t
t
r
a
c
t
e
d
s
p
a
m
s
[
2
,
3
]
an
o
r
g
an
ized
b
u
s
in
ess
aim
ed
at
m
a
k
in
g
m
o
n
ey
v
ia
u
s
e
o
f
m
ess
ag
es
with
o
u
t
t
h
e
co
n
s
en
t
o
f
u
s
er
s
.
T
h
eir
s
er
v
ices
ar
e
u
n
s
o
licited
a
d
v
er
ts
,
p
h
is
h
in
g
an
d
m
al
war
e
d
is
tr
ib
u
tio
n
ca
lled
s
p
am
s
.
Sp
am
s
ar
e
u
n
s
o
licited
/u
n
wan
ted
m
ess
ag
e
s
s
en
t to
u
s
er
s
.
W
i
th
s
p
am
s
o
n
th
e
r
is
e,
it h
as p
r
o
v
en
a
g
r
ea
t
co
n
ce
r
n
to
s
ec
u
r
ity
ex
p
er
ts
[
4
-
7
]
.
Su
ch
co
m
p
r
o
m
is
es
d
esig
n
ed
to
ev
ad
e
s
ec
u
r
ity
,
o
b
s
cu
r
e
d
ata
p
r
iv
ac
y
an
d
wea
k
e
n
n
etwo
r
k
in
f
r
astru
ctu
r
e
h
av
e
b
ec
o
m
e
a
g
r
ea
t
c
o
n
ce
r
n
with
n
eg
ativ
e
im
p
ac
ts
o
n
th
e
a
d
o
p
tio
n
o
f
tech
n
o
lo
g
y
.
T
h
is
in
clu
d
es
(
n
o
t
lim
ited
to
)
atta
ck
s
o
n
d
ata,
s
tealin
g
o
f
p
r
i
v
a
te
d
ata,
in
tr
u
s
io
n
,
s
er
v
ice
d
en
ial
an
d
o
u
tag
e
[
8
]
.
R
ep
o
r
ts
co
n
tin
u
es
to
s
tr
ess
o
f
in
tr
u
s
io
n
to
n
etwo
r
k
s
th
at
ef
f
ec
tiv
e
ly
attac
k
s
an
y
g
iv
en
tar
g
e
t
at
an
y
g
iv
en
tim
e
[9
-
1
1
]
.
T
h
e
e
x
p
o
n
en
tial
r
ate
o
f
attac
k
s
is
as
b
r
o
a
d
as
t
h
e
r
an
g
e
o
f
c
o
n
s
tr
u
ctiv
e
tech
n
o
lo
g
y
it
s
elf
-
lead
in
g
to
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2
0
8
8
-
8
7
0
8
F
o
r
g
in
g
a
d
ee
p
le
a
r
n
in
g
n
e
u
r
a
l n
etw
o
r
k
in
tr
u
s
io
n
d
etec
tio
n
fr
a
mewo
r
k
to
cu
r
b
…
(
A
r
n
o
ld
A
d
ima
b
u
a
Oju
g
o
)
1499
de
n
ial
o
f
s
er
v
ice
attac
k
s
am
o
n
g
s
t
o
th
er
s
.
T
h
is
ca
lls
f
o
r
u
r
g
en
t
n
ee
d
s
to
ter
m
in
ate
as
clo
s
e
an
d
as
f
ast
as
p
o
s
s
ib
le
to
its
s
o
u
r
ce
,
th
e
attac
k
s
.
Ma
n
y
o
f
th
ese
attac
k
s
o
n
n
etwo
r
k
r
eso
u
r
ce
s
ar
e
co
o
r
d
in
a
ted
an
d
tar
g
eted
at
a
clien
t
s
y
s
tem
an
d
lau
n
ch
e
d
ag
ain
s
t
th
e
s
er
v
er
v
ia
a
n
u
m
b
er
o
f
co
m
p
r
o
m
is
ed
s
y
s
tem
s
[
1
2
-
1
5
]
.
DDo
S
th
r
ea
ten
s
to
d
a
y
’
s
n
etwo
r
k
as
th
ey
ar
e
ca
r
e
f
u
lly
cr
af
ted
to
tar
g
et
a
lar
g
e
n
u
m
b
er
o
f
u
s
er
s
as
well
as
wr
ea
k
s
h
av
o
c
at
v
a
r
io
u
s
s
ec
u
r
ity
lev
e
ls
.
T
h
e
ea
s
e
with
wh
ich
th
ey
ar
e
p
er
p
etr
ate
d
h
as
also
b
ec
o
m
e
a
g
r
ea
t
co
n
c
er
n
,
ev
en
with
a
p
leth
o
r
a
o
f
a
v
ailab
le
to
o
ls
.
T
h
u
s
,
m
o
s
t
s
tu
d
ies
u
s
e
o
f
m
ac
h
i
n
e
lea
r
n
i
n
g
m
eth
o
d
s
to
ef
f
ec
tiv
ely
d
if
f
er
en
tiate
b
etwe
en
g
o
o
d
an
d
m
alicio
u
s
d
ata
-
p
ac
k
ets th
at
attem
p
ts
to
p
er
f
o
r
m
in
tr
u
s
io
n
[
1
6
-
1
9
]
.
Dis
tr
ib
u
ted
d
en
ial
o
f
s
er
v
ice
(
D
Do
S)
is
u
s
u
ally
a
ca
r
ef
u
lly
cr
af
ted
attac
k
,
in
itiated
to
tar
g
e
t
n
etwo
r
k
r
eso
u
r
ce
(
s
)
.
T
ar
g
eted
d
ir
ec
tly
at
co
m
p
r
o
m
is
ed
clien
ts
o
n
a
n
etwo
r
k
[
1
2
,
2
0
,
2
1
]
,
an
a
d
v
er
s
ar
y
g
ain
s
ac
ce
s
s
o
f
s
u
ch
co
m
p
r
o
m
is
ed
clien
t,
s
e
izin
g
u
p
r
eso
u
r
ce
s
s
u
ch
C
PU
tim
e,
b
a
n
d
wid
th
,
m
e
m
o
r
y
etc
-
to
ex
p
l
o
it
a
n
etwo
r
k
’
s
wea
k
n
ess
.
An
ad
v
e
r
s
ar
y
ac
h
iev
es
th
is
,
b
y
in
s
er
tin
g
m
alwa
r
e
th
at
s
o
u
g
h
t
to
o
v
er
wh
elm
th
e
n
etwo
r
k
with
r
eq
u
ests
[
2
0
-
23
]
.
Sin
ce
t
h
e
DDo
S
is
ca
r
ef
u
lly
cr
af
te
d
an
d
well
-
co
o
r
d
in
ated
,
th
e
m
a
g
n
itu
d
e
d
ep
en
d
s
o
n
th
e
b
o
t
n
et
s
ize
-
wh
ich
co
r
r
esp
o
n
d
s
also
to
th
e
s
ev
er
ity
o
f
th
e
attac
k
[
1
9
-
21
]
.
A
DDo
S
s
ee
k
s
to
e
x
h
au
s
t
tar
g
eted
r
eso
u
r
c
es,
d
en
y
‘
u
n
c
o
m
p
r
o
m
is
ed
’
clien
ts
’
ac
ce
s
s
t
o
s
er
v
ices
an
d
en
s
u
r
e
a
co
m
p
r
o
m
is
ed
n
etwo
r
k
o
n
a
lar
g
er
-
s
ca
le.
DDo
S
attac
k
s
ar
e
ea
s
y
to
f
ix
b
y
m
an
u
ally
d
is
co
n
n
ec
tin
g
af
f
ec
te
d
clien
ts
f
r
o
m
th
e
n
etwo
r
k
,
wh
en
d
etec
ted
.
Dete
ctio
n
s
ch
e
m
es
s
ee
k
to
s
to
p
a
d
etec
ted
atta
ck
as
clo
s
e
an
d
as
f
ast
a
s
p
o
s
s
ib
le
to
its
s
o
u
r
ce
[
1
5
,
2
0
,
2
1
,
2
4
-
27
].
T
h
er
e
ar
e
two
c
o
m
m
o
n
f
o
r
m
s
o
f
a
DDo
S
attac
k
:
(
i
)
A
n
ad
v
e
r
s
ar
y
s
ee
k
s
b
y
ex
p
l
o
it
d
esig
n
,
t
o
f
lo
o
d
a
n
etwo
r
k
with
clien
t
r
e
q
u
ests
th
at
ex
h
au
s
ts
o
r
s
eize
u
p
C
PU
-
tim
e,
p
o
wer
,
b
a
n
d
with
etc
-
m
ak
in
g
it
d
if
f
ic
u
lt
f
o
r
o
th
er
clien
ts
to
ac
ce
s
s
th
ese
r
eso
u
r
ce
s
(
i.e
.
f
lo
o
d
in
g
)
;
an
d
(
ii
)
A
d
v
er
s
ar
y
s
en
d
s
lar
g
e
v
o
lu
m
e
o
f
m
alicio
u
s
p
ac
k
ets
to
a
s
er
v
e
r
(
i.e
.
p
r
o
to
co
l
attac
k
)
.
A
DDo
S
attac
k
ca
n
ev
a
d
e
d
etec
tio
n
if
th
e
a
d
v
er
s
ar
y
s
p
o
o
f
s
th
e
s
o
u
r
ce
ad
d
r
ess
to
m
ask
p
ac
k
e
ts
an
d
m
ak
e
it
d
if
f
icu
lt
to
d
i
f
f
er
en
tiate
g
en
u
i
n
e
f
r
o
m
m
alic
io
u
s
d
ata
[
2
8
,
2
9
]
.
Dete
ctio
n
s
ch
em
es a
r
e
u
s
u
ally
g
r
o
u
p
ed
b
ased
o
n
t
h
eir
l
o
ca
lit
y
o
f
d
ep
lo
y
m
en
t a
s
[
1
4
,
2
0
,
2
1
,
3
0
]:
−
W
h
en
a
clien
t
s
y
s
tem
(
k
n
o
wn
as
s
o
u
r
ce
d
ev
ice)
h
as
s
ec
u
r
ity
m
ec
h
an
is
m
t
h
at
h
elp
s
it
id
en
tify
a
m
alicio
u
s
m
ess
ag
e
in
a
n
o
u
tg
o
in
g
p
ac
k
et
a
n
d
f
ilter
s
it.
Su
c
h
d
etec
tio
n
is
s
aid
,
t
o
h
a
v
e
b
e
en
lau
n
c
h
e
d
at
th
e
s
o
u
r
ce
o
f
th
e
attac
k
-
o
r
e
v
en
tin
g
n
etwo
r
k
clien
ts
g
en
er
a
tin
g
a
DDo
S
attac
k
.
T
h
is
d
et
ec
tio
n
tr
ies
to
s
to
p
a
DDo
S
as
clo
s
e
an
d
as
f
ast
as
p
o
s
s
ib
le
to
th
e
s
o
u
r
ce
o
f
th
e
attac
k
(
a
b
est
p
r
ac
tice)
a
n
d
m
in
im
izes
h
av
o
c
o
n
th
e
n
etwo
r
k
as we
ll
as o
n
o
th
er
u
n
co
m
p
r
o
m
is
ed
le
g
itima
te
p
ac
k
ets
cu
m
tr
af
f
ic
[
2
0
,
2
1
].
−
W
h
en
a
co
m
p
r
o
m
is
ed
s
y
s
tem
d
etec
ts
in
co
m
in
g
m
alicio
u
s
p
ac
k
et,
it
ca
n
clea
r
ly
d
is
t
in
g
u
is
h
in
g
g
en
u
i
n
e
‘
u
n
co
m
p
r
o
m
is
ed
’
p
ac
k
ets
f
r
o
m
‘
co
m
p
r
o
m
is
ed
’
attac
k
p
ac
k
ets
f
r
o
m
eit
h
er
th
e
m
is
u
s
e
o
f
i
n
tr
u
s
io
n
,
o
r
v
i
a
an
an
o
m
aly
-
in
tr
u
s
io
n
d
etec
tio
n
s
c
h
em
e.
T
h
is
is
ca
lled
a
v
ictim
-
en
d
d
etec
tio
n
.
An
d
an
attac
k
p
ac
k
et
th
a
t
r
ea
ch
es a
v
ictim
m
ay
d
en
ied
/d
eg
r
ad
ed
s
er
v
ices
an
d
b
a
n
d
wid
th
s
atu
r
atio
n
[
2
0
,
2
1
].
−
W
h
en
a
n
etwo
r
k
r
o
u
ter
ca
n
in
d
ep
en
d
e
n
tly
attem
p
t
to
id
en
tif
y
a
m
alicio
u
s
p
ac
k
et
b
y
r
ate
-
l
im
it
o
n
d
ata
-
tr
y
in
g
t
o
b
alan
ce
b
etwe
en
th
e
ac
cu
r
ac
y
o
f
th
e
d
etec
tio
n
an
d
th
e
co
n
s
u
m
p
tio
n
b
an
d
wid
th
o
f
an
attac
k
.
Su
ch
d
etec
tio
n
tr
ac
e
-
b
ac
k
b
e
co
m
es
ea
s
y
-
b
ec
au
s
e
th
e
p
ac
k
et
tr
af
f
ic
a
r
e
th
e
n
a
g
g
r
eg
ated
b
y
p
lacin
g
a
r
ate
-
lim
it
o
n
all
tr
af
f
ic
d
ata
s
i
n
ce
b
o
th
g
en
u
in
e
an
d
attac
k
p
ac
k
ets
ar
r
iv
e
at
th
e
r
o
u
ter
[
2
0
,
2
1
]
.
T
h
is
is
u
s
u
ally
ca
lled
th
e
co
r
e
-
en
d
d
et
ec
tio
n
.
K
n
o
w
led
g
e
-
d
r
ive
n
meth
o
d
s
o
f
d
etec
tio
n
-
r
eso
u
r
ce
s
a
r
e
a
s
tr
e
am
o
f
ev
e
n
ts
,
ch
ec
k
ed
o
n
th
e
b
ac
k
d
r
o
p
o
f
p
r
ed
ef
in
e
d
attac
k
r
u
les
a
n
d
p
atter
n
s
.
A
g
en
e
r
al
v
iew
o
f
k
n
o
wn
attac
k
s
a
r
e
f
o
r
m
u
lated
,
s
o
a
s
y
s
tem
ea
s
ily
id
en
tifie
s
o
cc
u
r
r
e
n
ce
o
f
an
at
tack
s
u
s
in
g
eith
er
o
f
s
ig
n
atu
r
e/an
am
o
lay
a
n
aly
s
is
,
s
elf
-
o
r
g
an
izin
g
m
a
p
s
,
an
d
s
tate
tr
an
s
iti
o
n
an
aly
s
i
s
.
Gil
a
n
d
Po
letto
[
3
1
]
u
s
ed
a
m
u
lti
-
lev
el
tr
ee
f
o
r
o
n
lin
e
p
ac
k
et
s
tatis
tics
t
o
m
o
n
ito
r
tr
af
f
ic
f
ea
ts
o
n
d
ev
ices,
an
d
t
o
d
etec
t/elim
in
ate
DDo
S.
I
t
a
g
g
r
eg
ates
an
d
r
ates
p
ac
k
ets
s
tatis
tic
s
at
v
a
r
io
u
s
lev
els,
ex
p
an
d
/co
n
tr
ac
t
p
ac
k
et
s
to
s
u
cc
ess
f
u
lly
d
etec
t o
n
g
o
in
g
attac
k
v
ia
a
d
is
p
r
o
p
o
r
tio
n
al
d
if
f
er
en
ce
b
etwe
en
th
e
r
ates
in
/
o
u
t
a
n
etwo
r
k
.
I
t
i
s
s
et
u
p
at
l
o
ca
tio
n
s
th
at
allo
w
s
th
e
eq
u
ip
p
e
d
d
ev
ice
t
o
eith
e
r
d
etec
t,
o
r
f
ail
to
m
o
n
ito
r
b
an
d
wid
t
h
attac
k
.
S
h
r
iv
ar
aj
[
3
2
]
Attack
er
s
ca
n
ev
ad
e
s
u
ch
d
etec
tio
n
m
o
d
e
b
y
r
a
n
d
o
m
izin
g
th
e
s
o
u
r
ce
ad
d
r
ess
I
P f
o
r
s
u
ch
m
alicio
u
s
d
ata
an
d
/o
r
p
ac
k
ets.
T
h
o
m
as
et
a
l
.
[
3
3
]
u
s
ed
Net
B
o
u
n
ce
r
to
d
is
tin
g
u
is
h
‘
u
n
c
o
m
p
r
o
m
is
ed
’
f
r
o
m
‘
c
o
m
p
r
o
m
i
s
ed
’
clien
ts
.
T
h
e
clien
t
ar
e
u
p
d
ated
o
n
a
li
s
t
-
s
o
th
at
o
n
ly
clien
ts
o
n
th
is
lis
t
ar
e
allo
wed
ac
ce
s
s
to
n
etwo
r
k
r
eso
u
r
ce
s
.
I
f
a
clien
t
n
o
t
in
th
e
lis
t
s
en
d
a
p
a
ck
ets,
th
e
NetBo
u
n
ce
r
test
f
o
r
leg
itima
cy
if
s
u
ch
a
clien
t
is
co
m
p
r
o
m
is
ed
.
I
f
th
e
clien
t
p
ass
es
th
e
te
s
ts
,
it
is
ad
d
ed
to
th
e
leg
itima
cy
l
is
t
an
d
s
u
ch
clien
t
is
g
r
an
ted
ac
ce
s
s
t
o
r
eso
u
r
ce
s
u
n
til
th
e
win
d
o
w
f
o
r
leg
itima
cy
ex
p
ir
es.
W
ith
th
e
lis
t’
s
ex
p
ir
atio
n
,
clien
ts
ar
e
r
ev
alid
ated
.
E
v
o
lu
tio
n
a
r
y
d
r
iv
en
a
n
d
m
ac
h
in
e
lear
n
in
g
(
ML
)
f
r
am
ewo
r
k
s
-
lear
n
in
g
al
g
o
r
ith
m
s
ef
f
ec
tiv
el
y
class
if
y
attac
k
s
with
h
eu
r
is
tics
th
at
c
an
to
ler
ate
u
n
ce
r
tain
ty
,
n
o
is
e,
am
b
u
ities
,
an
d
im
p
r
ec
is
io
n
wh
ile
y
ield
in
g
an
o
p
tim
al
s
o
lu
tio
n
.
I
t
m
o
d
els
tr
af
f
ic
d
ata
as
a
s
et
o
f
t
est
f
o
r
s
tatis
tical
in
f
er
en
ce
,
wh
ich
s
ee
k
s
to
d
eter
m
i
n
e
if
a
n
ew
in
s
tan
ce
b
elo
n
g
s
to
th
e
class
.
I
n
s
tan
ce
s
th
at
d
o
n
o
t
c
o
n
f
o
r
m
to
th
e
t
r
ain
ed
m
o
d
el
is
class
if
ied
as
an
an
o
m
aly
.
Ng
u
y
en
[
3
4
]
u
s
ed
a
p
r
o
ac
tiv
e
d
etec
tio
n
to
class
if
y
n
etwo
r
k
s
tatu
s
,
wh
ich
b
r
ea
k
a
n
attac
k
in
to
p
h
ases
-
s
o
th
at
it c
an
b
e
in
v
esti
g
ated
b
ased
o
n
s
elec
ted
f
ea
ts
o
f
in
ter
est.
I
t th
en
u
s
es k
-
n
ea
r
est n
eig
h
b
o
r
(
KNN)
to
g
r
o
u
p
d
ata
f
ea
ts
o
f
th
e
n
etwo
r
k
s
tatu
s
in
to
ea
ch
p
h
ase
o
f
th
e
DDo
S a
ttack
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
11
,
No
.
2
,
Ap
r
il 2
0
2
1
:
1
4
9
8
-
1509
1500
W
u
et
a
l
.
[3
5
]
u
s
ed
a
d
ec
is
io
n
tr
ee
s
th
at
d
etec
ts
a
ttack
u
s
in
g
1
5
p
a
r
am
eter
s
to
m
o
n
ito
r
p
ac
k
ets
an
d
f
lag
r
ates
in
an
d
o
u
t
a
s
y
s
tem
,
to
d
escr
ib
e
a
t
r
af
f
ic
f
lo
w
p
att
er
n
.
I
t
d
etec
ts
ab
n
o
r
m
al
tr
af
f
i
c
f
lo
w
v
ia
a
m
atch
s
ch
em
e
th
at
id
en
tifie
s
tr
af
f
ic
f
lo
w
s
im
ilar
to
an
attac
k
f
lo
w
as
well
as
to
tr
ac
e
b
ac
k
th
e
o
r
i
g
in
o
f
an
attac
k
b
ased
o
n
t
h
is
s
im
ilar
ity
.
L
ee
et
a
l
.
[3
6
]
u
s
ed
clu
s
ter
an
aly
s
is
o
n
DARP
A
2
0
0
0
d
ataset.
T
h
ei
r
r
esu
lts
s
h
o
wed
th
at
ea
ch
attac
k
was
p
ar
titi
o
n
ed
,
an
d
th
eir
m
eth
o
d
ca
n
ef
f
ec
t
iv
ely
d
etec
t
p
r
ec
u
r
s
o
r
s
o
f
a
D
Do
S,
an
d
th
e
attac
k
its
elf
.
Oju
g
o
et
a
l
.
[3
7
]
u
s
ed
a
s
ig
n
atu
r
e
-
b
ased
m
e
m
etic
alg
o
r
ith
m
to
d
etec
t
attac
k
as
a
class
if
icatio
n
p
r
o
b
lem
.
I
t
u
s
es
s
ev
en
p
ar
am
eter
s
to
m
o
n
ito
r
p
ac
k
et
r
ate
a
n
d
tr
af
f
ic
p
atter
n
.
I
t
u
s
es
a
m
atch
m
eth
o
d
to
id
en
tify
tr
af
f
ic
f
lo
w(
s
)
in
to
class
es
an
d
tr
ac
e
th
em
b
ac
k
to
a
n
atta
ck
’
s
o
r
ig
in
v
ia
th
e
s
im
ilar
ity
.
Kar
in
m
az
ad
an
d
Far
aa
h
i
[
3
8
]
u
s
ed
an
o
m
aly
-
b
ased
d
etec
tio
n
with
p
ac
k
et
f
ea
ts
,
an
aly
ze
d
v
i
a
a
r
ad
ial
b
asis
f
u
n
ctio
n
n
etwo
r
k
th
at
was a
p
p
lied
to
an
ed
g
e
-
r
o
u
ter
s
o
n
a
v
ictim
n
etwo
r
k
s
.
I
t
u
s
es
s
ev
en
-
f
ea
ts
to
tr
ain
a
R
B
F
-
n
et
an
d
cl
ass
if
ies
d
ata
in
to
n
o
r
m
al
an
d
attac
k
class
es.
I
f
m
o
d
el
r
ec
o
g
n
izes
an
in
co
m
in
g
tr
a
f
f
ic
as
attac
k
,
its
s
o
u
r
ce
p
ac
k
ets
ar
e
s
en
t
to
a
f
ilter
/attack
-
alar
m
r
o
u
tin
e
f
o
r
f
u
r
th
er
ac
tio
n
s
.
E
ls
e,
it is
s
en
t to
its
d
esti
n
atio
n
.
Mo
o
r
e
[
2
6
]
p
r
o
p
o
s
ed
a
d
etec
t
io
n
s
ch
em
e
w
h
er
e
ea
ch
r
o
u
ter
d
etec
ts
tr
af
f
ic
an
o
m
alies
u
s
in
g
p
r
o
f
iles
o
f
a
n
o
r
m
al
tr
af
f
ic
co
n
s
tr
u
cte
d
v
ia
s
tr
ea
m
s
am
p
lin
g
alg
o
r
it
h
m
s
.
T
h
eir
r
esu
lts
in
d
icate
s
:
(
i
)
we
ca
n
p
r
o
f
ile
a
n
o
r
m
al
tr
a
f
f
ic
r
ea
s
o
n
ab
ly
ac
c
u
r
ately
;
(
ii
)
id
en
tify
an
o
m
alies
with
lo
w
f
alse
-
p
o
s
itiv
e
a
n
d
f
alse
-
n
eg
ativ
e
r
ates
;
an
d
(
iii
)
b
e
co
s
t
ef
f
ec
tiv
e
with
m
em
o
r
y
c
o
n
s
u
m
p
tio
n
p
er
p
ac
k
et
co
m
p
u
tatio
n
.
Als
o
,
th
e
r
o
u
ter
s
ex
ch
an
g
es
d
ata
with
ea
c
h
o
th
er
to
i
n
cr
ea
s
e
co
n
f
id
e
n
ce
in
th
eir
d
etec
tio
n
.
R
esu
lts
s
h
o
w
th
at
ea
ch
r
o
u
t
er
p
r
o
f
iles
ca
p
tu
r
e
k
ey
ch
ar
ac
ter
is
tics
o
f
th
e
tr
af
f
ic
ef
f
ec
tiv
ely
an
d
id
en
tif
y
an
o
m
alies
with
lo
w
f
alse
p
o
s
itiv
e
an
d
f
alse
n
eg
ativ
e
r
ate.
Oju
g
o
et
a
l
.
[
1
5
]
ex
ten
d
ed
Oju
g
o
et
a
l
.
[3
7
]
v
ia
a
g
e
n
etic
alg
o
r
ith
m
s
ig
n
atu
r
e
r
u
le
-
b
ased
m
o
d
el,
with
10
-
f
ea
ts
to
m
o
n
ito
r
in
/o
u
t
p
ac
k
et
r
ates.
J
alili
et
al
.
[3
9
]
ad
v
an
ce
d
th
is
p
o
s
itio
n
u
s
in
g
SP
UNNI
D
-
an
u
n
s
u
p
er
v
is
ed
n
e
u
r
al
n
et
to
e
x
tr
ac
t
tr
af
f
ic
f
ea
ts
,
an
aly
s
e
a
n
d
class
if
y
tr
af
f
ic
p
atter
n
s
as
eith
er
a
n
o
r
m
al
o
r
DDo
S a
ttack
.
C
h
en
an
d
Delis
[
40
]
u
s
ed
a
d
i
s
tr
ib
u
ted
ch
an
g
e
p
o
in
t (
DC
P)
d
etec
tio
n
th
at
ad
o
p
ts
ch
an
g
e
a
g
g
r
eg
atio
n
tr
ee
s
(
C
AT
s
)
.
T
h
is
n
o
n
-
p
ar
a
m
etr
ic
m
o
d
el
d
escr
ib
es
d
is
tr
ib
u
t
io
n
o
f
p
r
e/p
o
s
t
ch
a
n
g
e
in
tr
af
f
ic.
W
h
en
a
DDo
S
f
lo
o
d
-
attac
k
is
lau
n
ch
ed
,
th
e
c
u
m
u
lativ
e
d
e
v
iatio
n
is
n
o
ticea
b
ly
h
ig
h
e
r
th
an
r
an
d
o
m
f
l
u
ctu
atio
n
s
.
T
h
e
C
AT
is
d
esig
n
ed
s
o
a
r
o
u
ter
d
etec
ts
ab
r
u
p
t
ch
a
n
g
es
in
tr
af
f
ic.
A
d
o
m
ain
s
er
v
er
u
s
es
th
e
tr
af
f
i
c
ch
an
g
e
p
atter
n
s
d
etec
ted
at
attac
k
-
t
r
an
s
it
r
o
u
ter
s
to
co
n
s
tr
u
ct
C
AT
s
.
I
t
w
o
r
k
s
in
in
lin
e
-
m
o
d
e
to
in
s
p
e
ct
an
d
m
an
ip
u
late
o
n
g
o
in
g
tr
af
f
ic
i
n
r
ea
l
tim
e.
I
t
co
n
tin
u
o
u
s
ly
m
o
n
ito
r
s
b
o
th
at
tack
s
an
d
l
eg
itima
te
tr
af
f
ic
b
y
in
s
p
ec
tin
g
p
ac
k
ets
an
d
co
r
r
elatin
g
ev
e
n
ts
am
o
n
g
d
if
f
er
en
t ses
s
io
n
s
.
I
t p
r
o
ac
tiv
e
ly
ter
m
in
ates a
s
ess
io
n
wh
en
it d
etec
ts
an
attac
k
.
I
n
tr
u
s
io
n
s
ch
em
es
h
av
e
b
ee
n
d
ev
is
ed
to
m
in
im
ize
t
h
e
h
a
v
o
c
b
y
in
tr
u
s
io
n
ac
tiv
ities
[
4
1
,
4
2
]
,
an
d
f
o
r
n
etw
o
r
k
s
,
s
o
m
e
b
e
h
av
io
r
ex
is
ts
with
an
ex
ter
n
al
ev
e
n
t.
T
h
e
ar
ch
itectu
r
e
o
f
in
tr
u
s
io
n
d
etec
tio
n
s
y
s
tem
s
(
I
DS)
s
ee
k
s
to
u
n
m
ask
m
alicio
u
s
p
r
o
ce
s
s
es e
ith
er
v
ia
th
e
s
ig
n
atu
r
e
o
f
s
u
ch
attac
k
,
o
r
v
ia
an
an
o
m
aly
o
n
th
e
n
etwo
r
k
tr
af
f
ic.
An
I
DS
g
o
al
is
to
s
ec
u
r
e
n
etwo
r
k
r
eso
u
r
ce
s
an
d
g
r
a
n
t
a
u
s
er
s
y
tem
co
n
f
i
d
en
tiality
,
d
ata
in
teg
r
ity
,
an
d
r
eso
u
r
ce
av
ailab
ilit
y
[
1
5
,
1
6
]
.
An
I
D
S
ca
n
r
etr
iev
e
d
ata
f
r
o
m
a
n
etwo
r
k
s
ec
tio
n
f
o
r
an
aly
s
is
to
iu
n
v
eil
in
tr
u
s
io
n
af
f
ec
te
d
co
m
p
o
n
e
n
t(
s
)
u
s
in
g
a
v
a
r
io
u
s
tech
n
i
q
u
es.
T
h
ese
tech
n
iq
u
es
ar
e
ch
ar
ac
te
r
ized
to
d
e
p
en
d
o
n
3
-
m
ain
asp
ec
ts
[
1
5
,
2
5
, 4
3
,
4
4
]
in
F
ig
u
r
e
1:
Fig
u
r
e
1
.
Stru
ctu
r
al
ar
ch
itectu
r
e
an
d
class
if
icatio
n
o
f
a
n
I
DS
-
Data
s
o
u
r
ce
-
a
n
etwo
r
k
I
DS
ex
am
in
es
th
e
tr
af
f
ic;
wh
ile,
a
h
o
s
t
I
DS
ex
am
in
es
n
etwo
r
k
co
m
p
o
n
en
ts
s
u
ch
as th
e
o
p
er
atin
g
s
y
s
tem
.
C
o
n
v
er
s
ely
,
a
h
y
b
r
id
I
DS su
p
p
o
r
ts
b
o
th
s
o
u
r
ce
s
o
f
d
ata.
-
I
n
tr
u
s
io
n
m
o
d
el
d
ea
ls
with
m
is
u
s
e
d
etec
tio
n
.
Sig
n
atu
r
e
d
etec
tio
n
aim
s
to
v
er
if
y
th
e
s
ig
n
a
tu
r
e
o
n
d
ata
tr
af
f
ic;
W
h
ile,
an
o
m
aly
d
etec
tio
n
s
ee
k
s
to
v
er
if
y
th
e
s
y
s
tem
b
eh
av
io
r
.
Hy
b
r
i
d
I
DS
m
o
n
ito
r
s
b
o
t
h
d
etec
tio
n
m
o
d
es.
-
Au
d
it
co
llectio
n
a
n
d
a
n
aly
s
is
is
im
p
lem
en
ted
u
s
in
g
2
-
m
eth
o
d
s
n
a
m
ely
:
(
i
)
ce
n
tr
ali
ze
d
-
co
n
tr
o
lled
r
eso
u
r
ce
I
DS
;
an
d
(
ii
)
d
ec
en
tr
alize
d
I
DS
is
co
n
tr
o
lled
f
r
o
m
a
lo
ca
l
co
n
tr
o
l
n
o
d
e
with
h
ier
ar
ch
ical
r
ep
o
r
tin
g
to
o
n
e
o
r
m
o
r
e
ce
n
tr
al
lo
ca
tio
n
(
s
)
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2
0
8
8
-
8
7
0
8
F
o
r
g
in
g
a
d
ee
p
le
a
r
n
in
g
n
e
u
r
a
l n
etw
o
r
k
in
tr
u
s
io
n
d
etec
tio
n
fr
a
mewo
r
k
to
cu
r
b
…
(
A
r
n
o
ld
A
d
ima
b
u
a
Oju
g
o
)
1501
Mo
tiv
atio
n
f
o
r
th
e
s
tu
d
y
:
-
DDo
S
attac
k
s
ar
e
r
is
in
g
with
ea
s
e
at
an
alar
m
in
g
g
r
o
wth
r
at
e
o
f
m
alicio
u
s
task
s
g
u
is
ed
to
ex
p
lo
it
u
s
er
s
.
T
h
ey
h
a
v
e
ca
u
s
ed
u
s
er
’
s
lo
s
s
in
tr
u
s
t
-
lev
el
o
f
tech
ad
o
p
tio
n
an
d
f
i
n
an
ce
s
.
Usi
n
g
p
a
r
ticu
lar
tech
n
iq
u
e
t
o
r
eso
lv
e
DDo
S
with
I
DS
[
2
]
,
an
d
s
tatis
tical
m
o
d
els
[
4
4
]
h
a
v
e
p
r
o
v
e
n
s
u
cc
ess
f
u
l
o
n
m
ali
cio
u
s
tr
af
f
ics.
Ho
wev
er
,
co
m
b
atin
g
DDo
S
is
an
‘
in
co
n
clu
s
iv
e’
an
d
c
o
n
ti
n
u
o
u
s
task
as
m
an
y
o
f
m
o
d
el
s
’
p
er
f
o
r
m
a
n
ce
ar
e
h
am
p
e
r
ed
b
y
s
elec
tio
n
o
f
p
ar
am
eter
s
th
at
o
f
te
n
r
esu
lts
in
m
o
d
el
o
v
er
-
f
itti
n
g
an
d
o
v
e
r
-
tr
ain
in
g
[
*
]
.
-
Ma
n
y
m
o
d
el(
s
)
em
p
lo
y
h
ill
-
cl
im
b
in
g
m
eth
o
d
s
;
an
d
t
h
u
s
,
o
f
t
en
g
ets tr
ap
p
e
d
at
lo
ca
l m
ax
im
a.
-
DDo
S
attac
k
p
r
ev
en
ts
leg
it
im
ate
u
s
er
s
f
r
o
m
ac
ce
s
s
in
g
r
eso
u
r
ce
s
.
I
t
co
n
s
u
m
es
all
av
ailab
le
r
eso
u
r
ce
s
,
o
v
er
wh
elm
s
th
e
n
etwo
r
k
wit
h
r
eq
u
ests
o
v
er
lo
ad
,
b
l
o
ck
s
u
n
co
m
p
r
o
m
is
ed
u
s
er
s
’
ac
ce
s
s
to
p
r
o
v
is
io
n
ed
s
er
v
ices
with
a
v
iew
to
co
m
p
r
o
m
is
e
th
e
en
tire
n
etwo
r
k
u
n
t
il
co
u
n
ter
m
ea
s
u
r
es
a
r
e
em
p
l
o
y
ed
.
T
h
u
s
,
th
e
u
r
g
en
t
n
ee
d
to
id
e
n
tify
t
h
eir
s
o
u
r
ce
,
m
an
a
g
e
an
d
p
r
ev
e
n
t
th
em
.
T
h
is
is
ef
f
ec
tiv
ely
ac
h
iev
ed
v
i
a
s
tatis
t
ical
m
ea
n
s
an
d
g
u
id
es a
u
s
er
to
ef
f
icien
tly
d
if
f
er
e
n
tiate
b
etwe
en
leg
itima
te
an
d
m
alicio
u
s
ac
t
s.
-
Fo
r
m
u
latin
g
an
ef
f
ec
tiv
e
d
ete
ctio
n
s
ch
em
e
h
as
its
s
etb
ac
k
(
s
)
-
as
m
alicio
u
s
tr
af
f
ics
ar
e
p
o
is
ed
b
y
th
eir
d
esig
n
ar
ch
itectu
r
e
to
ev
a
d
e
f
ilter
s
,
wh
o
s
e
p
er
f
o
r
m
an
ce
a
r
e
h
in
d
er
ed
b
y
th
e
lim
ited
s
ize
o
f
ch
ar
ac
ter
s
,
non
-
a
v
ailab
ilit
y
o
f
m
alicio
u
s
tr
af
f
ic
d
ata
etc
-
c
r
ea
tin
g
im
p
e
d
im
en
ts
in
s
elec
tin
g
p
a
r
am
ete
r
s
f
o
r
tr
ai
n
in
g
.
An
d
,
u
ltima
tely
,
lead
i
n
g
to
b
o
t
h
p
o
o
r
lear
n
i
n
g
an
d
class
if
icatio
n
o
f
t
h
e
lear
n
in
g
alg
o
r
ith
m
.
T
o
o
v
er
c
o
m
e
th
ese
s
h
o
r
tf
alls
in
d
etec
tin
g
m
alicio
u
s
tr
af
f
i
cs,
we
ad
o
p
t
a
d
ee
p
n
e
u
r
al
n
etwo
r
k
to
r
ed
u
ce
n
o
is
e
v
ia
p
r
e
-
p
r
o
ce
s
s
in
g
o
f
tr
af
f
ic
p
ac
k
ets
an
d
f
in
e
-
tu
n
in
g
m
ess
ag
es
s
en
t
as
r
eq
u
ests
s
en
t
to
/f
r
o
m
a
s
er
v
er
,
to
en
h
an
ce
a
d
eq
u
ate
cl
ass
if
icatio
n
.
2.
RE
S
E
ARCH
M
E
T
H
O
D
2
.
1
.
Dee
p neura
l net
wo
rk
s
(
DNN
)
DNN
u
s
es
d
ee
p
lear
n
in
g
to
ad
ap
t
u
s
ef
u
l
s
elec
ted
f
ea
ts
o
f
in
t
er
est
an
d
p
ar
am
eter
s
,
ca
r
ef
u
lly
co
n
s
tr
u
ctin
g
a
m
u
lti
-
lay
er
n
et
wo
r
k
f
r
o
m
v
ast
am
o
u
n
t
o
f
d
a
ta.
I
ts
d
ee
p
ar
ch
itectu
r
e
at
its
in
p
u
t,
h
id
d
e
n
an
d
o
u
tp
u
t
lay
e
r
s
-
h
elp
s
to
im
p
r
o
v
e
its
p
r
ed
ictio
n
ac
c
u
r
ac
y
.
I
ts
h
id
d
en
lay
er
tr
an
s
f
o
r
m
s
n
o
n
-
lin
ea
r
ly
f
r
o
m
a
p
r
ev
io
u
s
lay
e
r
to
t
h
e
n
e
x
t
[
2
1
,
4
5
]
.
Pro
p
o
s
e
d
b
y
Hin
to
et
a
l
.
[4
6
]
,
a
DNN
is
tr
ain
ed
v
ia
two
p
h
ases
:
p
r
e
-
tr
ain
ed
,
an
d
f
in
e
-
t
u
n
ed
p
r
o
ce
s
s
es [
2
1
,
4
7
].
T
h
e
au
to
-
en
co
d
er
is
an
u
n
s
u
p
er
v
is
ed
m
u
lti
-
lay
er
e
d
n
e
u
r
al
n
etwo
r
k
co
n
s
is
tin
g
b
o
th
an
e
n
c
o
d
er
a
n
d
a
d
ec
o
d
er
n
etwo
r
k
.
I
ts
en
co
d
er
s
ee
k
s
to
tr
an
s
f
o
r
m
in
p
u
ts
d
ata
-
p
o
in
ts
f
r
o
m
a
h
ig
h
u
n
to
a
lo
w
-
d
im
en
s
io
n
v
ia
an
en
co
d
in
g
f
u
n
ctio
n
f
encoder
as
in
(
1
)
wh
er
e
x
m
is
a
d
ata
p
o
in
t,
an
d
h
m
is
th
e
en
co
d
in
g
v
ec
to
r
o
b
tain
e
d
.
C
o
n
v
er
s
ely
,
its
d
ec
o
d
er
n
etwo
r
k
s
ee
k
s
to
r
ec
o
n
s
tr
u
ct
th
e
f
u
n
ctio
n
u
s
in
g
f
decoder
as
in
(
2
)
with
x
m
as
d
ec
o
d
in
g
v
e
c
t
o
r
f
r
o
m
h
m
.
T
h
u
s
,
r
e
v
e
r
t
s
t
h
e
o
p
e
r
a
t
i
o
n
s
o
f
t
h
e
e
n
c
o
d
e
r
[
48
]
.
O
j
u
g
o
a
n
d
E
b
o
k
a
[
21
]
i
n
G
i
l
r
o
t
a
n
d
B
en
g
io
[
47
]
d
etails s
p
ec
if
ic
alg
o
r
ith
m
s
f
o
r
en
co
d
in
g
an
d
d
ec
o
d
in
g
f
u
n
cti
o
n
s
r
esp
ec
tiv
ely
.
ℎ
=
(
)
(
1
)
=
(
ℎ
)
(
2
)
At
th
e
p
r
e
-
tr
ain
in
g
p
h
ase,
N
au
to
en
co
d
er
s
ca
n
b
e
s
tack
ed
o
n
to
an
N
-
h
i
d
d
en
-
la
y
er
s
o
th
at
with
in
p
u
t
ac
ce
p
ted
,
th
e
in
p
u
t
lay
er
an
d
f
ir
s
t
h
id
d
en
lay
er
ac
ts
an
en
co
d
er
o
f
th
e
f
ir
s
t
au
to
-
en
co
d
er
.
T
h
ey
ar
e
tr
ain
ed
as
th
u
s
,
to
m
i
n
im
ize
th
e
r
ec
o
n
s
tr
u
ctio
n
e
r
r
o
r
.
T
r
ai
n
in
g
p
ar
am
eter
(
s
)
o
f
th
e
e
n
co
d
e
r
ar
e
u
s
e
d
to
i
n
itialize
f
ir
s
t
h
id
d
en
lay
er
b
ef
o
r
e
p
r
o
ce
e
d
in
g
to
s
ec
o
n
d
h
id
d
e
n
lay
e
r
.
T
h
e
r
e,
th
e
f
ir
s
t
an
d
s
ec
o
n
d
h
i
d
d
en
lay
er
s
ar
e
s
elec
ted
as
en
co
d
er
(
s
)
an
d
as
in
th
e
ea
r
lier
s
tag
e,
th
e
s
ec
o
n
d
h
id
d
e
n
lay
er
is
in
itialized
b
y
th
e
s
ec
o
n
d
tr
ain
ed
au
t
o
-
en
co
d
er
.
T
h
is
p
r
o
ce
s
s
co
n
tin
u
es
till
th
e
ℎ
au
to
-
en
c
o
d
er
is
tr
ain
ed
a
n
d
in
itializes
th
e
f
in
a
l
h
id
d
en
lay
er
.
W
ith
all
h
id
d
e
n
lay
e
r
s
s
tack
e
d
in
th
e
au
to
-
e
n
co
d
e
r
at
ea
ch
tr
ain
in
g
N
-
tim
es,
th
e
y
ar
e
th
u
s
r
eg
ar
d
ed
as
p
r
e
-
tr
ain
ed
.
T
h
is
f
ea
t
h
as
p
r
o
v
e
n
to
b
e
s
ig
n
if
ican
tly
b
etter
th
an
r
an
d
o
m
in
itializa
tio
n
.
I
t
also
ac
h
iev
es
b
etter
g
en
er
aliza
tio
n
[
2
0
,
2
1
,
4
6
,
4
9
].
Fin
e
-
tu
n
in
g
is
a
s
u
p
e
r
v
i
s
e
d
p
h
a
s
e
t
h
a
t
s
e
e
k
s
t
o
o
p
t
i
m
i
z
e
a
D
N
N
’
s
p
e
r
f
o
r
m
a
n
c
e
b
y
r
etr
ain
in
g
th
e
n
etwo
r
k
lab
eled
tr
ain
in
g
d
ata.
I
t
co
m
p
u
tes
th
e
er
r
o
r
s
as
a
d
if
f
e
r
en
ce
in
r
ea
l
v
er
s
u
s
p
r
ed
icte
d
v
al
u
es
v
ia
b
ac
k
-
p
r
o
p
a
g
ated
s
to
ch
asti
c
g
r
ad
ien
t
d
escen
t
(
SGD)
,
w
h
ich
r
a
n
d
o
m
ly
s
elec
ts
d
ata,
an
d
iter
ativ
e
ly
u
p
d
ates
g
r
ad
ie
n
t
d
ir
ec
tio
n
with
th
e
weig
h
t
p
ar
am
eter
s
.
A
m
er
it
o
f
th
e
SGD
is
th
at
it
co
n
v
er
g
es
f
aster
an
d
d
o
es
n
o
t
r
eq
u
ir
e
th
e
en
tire
d
ataset.
T
h
is
m
ak
e
s
it
s
u
itab
le
f
o
r
co
m
p
lex
n
eu
r
al
n
et
wo
r
k
s
as
g
iv
e
n
in
(
3
)
with
E
a
s
lo
s
s
f
u
n
ctio
n
,
y
is
lab
el
an
d
t
is
o
u
tp
u
t o
f
th
e
n
et
wo
r
k
[
2
0
,
2
1
]:
2
1
1
()
2
ii
j
E
M
y
t
=
=−
(
3
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
11
,
No
.
2
,
Ap
r
il 2
0
2
1
:
1
4
9
8
-
1509
1502
T
h
e
g
r
a
d
ien
t
o
f
th
e
weig
h
t
w
is
o
b
tain
ed
as
a
d
er
iv
ativ
e
o
f
th
e
er
r
o
r
e
q
u
atio
n
-
s
o
th
at
a
n
u
p
d
ated
SGD
is
g
iv
en
b
y
(
4
)
with
ŋ
is
s
t
ep
-
s
ize,
h
is
n
u
m
b
er
o
f
h
id
d
en
lay
er
s
[
2
0
,
2
1
]:
(
)
(
)
.
.
1
.
n
e
w
o
l
d
i
j
i
j
j
j
j
i
i
W
W
y
t
y
y
h
=
−
−
−
ŋ
(
4
)
T
h
is
p
r
o
ce
s
s
is
o
p
tim
ize
d
b
y
th
e
weig
h
ts
an
d
th
r
esh
o
ld
b
ased
o
n
co
r
r
ec
tly
lab
elled
d
at
a.
T
h
u
s
,
a
DNN
ca
n
lear
n
ac
cu
r
ately
at
its
f
in
al
o
u
tp
u
t
an
d
d
ir
ec
t
th
u
s
,
task
all
n
etwo
r
k
p
ar
am
eter
s
to
p
er
f
o
r
m
c
o
r
r
ec
t
class
if
icatio
n
s
[
2
0
,
2
1
].
2
.
2
.
T
he
deep
lea
rn
ing
f
ra
m
ewo
r
k
/a
lg
o
rit
h
m
Dee
p
lear
n
in
g
s
o
lv
es
task
s
b
y
:
(
a)
d
iv
id
in
g
tr
ai
n
in
g
d
ata
in
t
o
clu
s
ter
s
,
co
m
p
u
tin
g
ce
n
ter
p
o
in
ts
f
r
o
m
ea
ch
clu
s
ter
p
o
in
t,
(
b
)
ea
c
h
clu
s
ter
is
tr
ain
e
d
an
d
s
ca
led
s
o
th
at
ea
ch
DNN
lear
n
s
th
e
v
ar
io
u
s
attr
ib
u
tes o
f
ea
ch
s
u
b
s
et,
(
c)
th
e
test
d
ata
ap
p
lies
th
e
p
r
ev
i
o
u
s
clu
s
te
r
ce
n
ter
s
in
its
f
ir
s
t
s
tep
to
d
etec
t
o
u
tlier
(
s
)
b
y
t
h
e
p
r
e
-
tr
ain
ed
DNNs,
an
d
(
d
)
o
u
tp
u
t
o
f
ea
ch
DNN
is
ag
g
r
eg
ated
f
o
r
th
e
f
in
al
r
esu
lt
d
ata/o
u
tlier
s
[
7
,
2
0
,
2
1
]
.
Pro
p
o
s
ed
s
o
lu
tio
n
is
d
iv
id
ed
i
n
to
3
-
s
tep
s
[
1
0
,
2
0
,
2
1
,
50
]:
−
Step
1
d
iv
id
es
d
ata
in
to
tr
ain
an
d
test
clu
s
ter
s
o
r
p
ar
titi
o
n
s
.
DNN
s
to
r
es
co
m
p
u
ted
clu
s
ter
ce
n
ter
s
,
u
s
ed
as
in
itializat
io
n
ce
n
ter
(
s
)
to
g
en
er
ate
test
d
ataset
s
.
Data
s
e
t
attr
ib
u
tes
ar
e
f
o
r
m
atted
as
d
ata
-
p
o
in
ts
f
o
r
s
elec
ted
p
ar
am
eter
s
,
a
n
d
t
h
e
d
ata
-
p
o
in
ts
in
t
h
e
tr
ain
i
n
g
d
ataset
ar
e
alig
n
e
d
in
t
o
g
r
o
u
p
s
o
f
s
am
e
class
.
T
o
im
p
r
o
v
e
th
e
p
er
f
o
r
m
an
ce
o
f
t
h
e
DNN,
m
o
d
el
r
ev
is
es
clu
s
ter
n
u
m
b
e
r
s
(
to
b
etwe
en
2
to
6
)
an
d
s
ig
m
a
v
alu
es
(
i.e
.
0
.
1
t
o
1
.
0
)
.
T
h
e
m
i
n
im
u
m
d
is
tan
ce
f
r
o
m
a
d
ata
p
o
in
t
to
ea
c
h
clu
s
ter
ce
n
ter
is
m
ea
s
u
r
ed
,
a
n
d
a
d
ata
-
p
o
in
t’
s
n
ea
r
n
ess
to
a
clu
s
ter
,
ass
ig
n
s
it
to
th
at
clu
s
te
r
-
class
.
T
r
ain
in
g
s
ets
g
en
er
ate
d
b
y
clu
s
ter
s
ar
e
tak
en
u
p
as
in
p
u
t
to
DNNs.
Fo
r
tr
ain
in
g
,
t
h
e
n
u
m
b
e
r
o
f
DNNs
s
h
o
u
ld
eq
u
al
th
e
n
u
m
b
er
o
f
clu
s
ter
s
.
DNN
ar
ch
itectu
r
e
co
n
s
is
ts
o
f
f
iv
e
lay
e
r
s
:
an
in
p
u
t,
two
h
id
d
e
n
,
a
s
o
f
tm
ax
an
d
an
o
u
tp
u
t
lay
er
r
esp
ec
tiv
ely
.
T
h
e
h
id
d
en
la
y
er
s
lear
n
f
ea
ts
f
r
o
m
ea
ch
tr
a
in
in
g
s
u
b
s
et,
a
n
d
t
h
e
to
p
la
y
er
is
a
f
i
v
e
-
d
im
en
s
io
n
al
o
u
tp
u
t
v
ec
to
r
.
E
ac
h
tr
ain
in
g
s
u
b
s
et
g
en
er
ated
f
r
o
m
th
e
k
th
clu
s
ter
ce
n
ter
i
s
r
eg
ar
d
ed
as
in
p
u
t
d
ata
to
f
ee
d
in
to
k
t
h
DNN
r
esp
ec
tiv
ely
.
T
r
ain
ed
s
u
b
-
DNN
m
o
d
els
ar
e
m
ar
k
ed
s
u
b
-
DNN
1
to
k
[
2
0
,
2
1
,
5
0
].
−
Step
2
u
s
es
test
d
ataset
to
g
en
er
ate
k
-
d
atasets
with
th
e
p
r
e
v
i
o
u
s
clu
s
ter
ce
n
ter
o
b
tain
ed
f
r
o
m
clu
s
ter
s
in
Step
1
.
T
h
e
test
s
u
b
-
d
ataset
ar
e
d
en
o
ted
as
t
est
1
th
r
o
u
g
h
tes
t
k
[
2
0
,
2
1
,
5
0
].
−
Step
3
:
T
h
e
k
-
test
d
ata
s
u
b
s
ets
ar
e
f
ed
in
to
k
s
u
b
-
DNNs,
w
h
ich
wer
e
co
m
p
leted
b
y
th
e
k
tr
ain
in
g
d
ata
s
u
b
s
ets
in
Step
1
.
Ou
tp
u
t
o
f
ea
ch
s
u
b
-
DNN
is
in
teg
r
ated
as
f
in
al
o
u
t
p
u
t
an
d
em
p
lo
y
ed
to
an
aly
s
e
p
o
s
itiv
e
d
et
ec
tio
n
r
ates.
T
h
en
,
co
n
f
u
s
io
n
m
atr
i
x
is
u
s
ed
to
an
aly
s
e
m
in
in
g
p
er
f
o
r
m
a
n
c
e
o
f
g
en
er
ate
d
r
u
les [
2
0
,
2
1
,
5
0
].
Pro
p
o
s
ed
DNN
class
if
ies
d
a
ta
v
ia
b
ac
k
-
p
r
o
p
a
g
atio
n
lear
n
in
g
th
at
m
a
p
s
in
p
u
t
s
ig
n
al
s
to
lo
w
-
d
im
en
s
io
n
al
s
p
ac
e
th
at
s
ee
k
s
t
o
d
is
co
v
er
p
atter
n
s
in
th
e
d
ata
s
et
s
.
Alg
o
r
ith
m
is
th
u
s
[
2
0
,
2
1
,
50
-
56
]:
I
n
p
u
t:
Data
s
et
,
clu
s
ter
n
u
m
b
er
,
n
u
m
b
er
o
f
h
id
d
en
-
lay
e
r
n
o
d
e
s
HL
N,
n
u
m
b
er
o
f
h
i
d
d
en
la
y
e
r
s
HL
.
Ou
tp
u
t: Fin
al
p
r
ed
ictio
n
r
esu
lts
-
Div
id
e
r
aw
d
ataset
in
to
two
co
m
p
o
n
e
n
ts
: tr
ain
in
g
an
d
a
test
in
g
d
ataset.
/*
g
et
th
e
la
r
g
est m
atr
ix
eig
en
v
ec
to
r
s
an
d
tr
ain
in
g
d
ata
s
u
b
s
ets*
/
-
Ob
tain
clu
s
ter
ce
n
ter
an
d
cl
u
s
ter
r
esu
lts
.
Her
e,
th
e
clu
s
ter
in
g
r
esu
lts
ar
e
r
eg
ar
d
ed
as
tr
ain
in
g
d
ata
s
u
b
s
ets.
/*
T
r
ain
ea
ch
DNN
with
ea
ch
tr
ain
in
g
d
ata
s
u
b
s
et*
/
-
L
ea
r
n
in
g
r
ate,
d
e
-
n
o
is
in
g
a
n
d
s
p
ar
s
ity
p
ar
am
eter
s
a
r
e
s
e
t
an
d
th
e
weig
h
t
a
n
d
b
ias
ar
e
r
an
d
o
m
ly
in
itialized
.
-
HL
N
is
s
et
4
0
-
n
o
d
es f
o
r
f
ir
s
t a
n
d
2
0
-
n
o
d
es f
o
r
s
ec
o
n
d
h
id
d
e
n
lay
er
.
-
C
o
m
p
u
te
s
p
ar
s
ity
co
s
t f
u
n
ctio
n
-
Par
am
e
ter
weig
h
ts
an
d
b
ias ar
e
u
p
d
ated
-
T
r
ain
k
s
u
b
-
DNNs c
o
r
r
esp
o
n
d
in
g
to
th
e
tr
ai
n
in
g
d
a
ta
s
u
b
s
ets.
-
Fin
e
-
tu
n
e
th
e
s
u
b
-
DNNs b
y
u
s
in
g
b
ac
k
p
r
o
p
ag
atio
n
t
o
tr
ain
t
h
em
.
-
Fin
al
s
tr
u
ctu
r
e
o
f
tr
ain
e
d
s
u
b
-
DNNs is o
b
tain
ed
an
d
lab
elled
with
ea
ch
tr
ain
in
g
d
ata
s
u
b
s
et
.
-
Div
id
e
test
d
ataset
in
to
s
u
b
s
e
ts
with
SC
.
C
lu
s
ter
ce
n
ter
p
a
r
am
eter
s
f
r
o
m
th
e
tr
ain
i
n
g
d
a
ta
clu
s
ter
s
ar
e
u
s
ed
.
-
T
est
d
ata
s
u
b
s
ets
is
u
s
ed
to
t
est
co
r
r
esp
o
n
d
in
g
s
u
b
-
DNNs,
b
ased
o
n
ea
c
h
clu
s
ter
ce
n
ter
b
etwe
en
t
h
e
test
in
g
an
d
tr
ain
in
g
d
ata
s
u
b
s
e
ts
/*
ag
g
r
eg
ate
ea
ch
p
r
ed
ictio
n
r
e
s
u
lt*
/
-
R
esu
lts
ar
e
g
en
er
ated
b
y
ea
c
h
s
u
b
-
DNN,
ar
e
in
teg
r
ated
an
d
t
h
e
f
in
al
o
u
tp
u
ts
ar
e
o
b
tain
ed
.
-
r
etu
r
n
class
if
icatio
n
r
esu
lt=
f
in
al
o
u
tp
u
t
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2
0
8
8
-
8
7
0
8
F
o
r
g
in
g
a
d
ee
p
le
a
r
n
in
g
n
e
u
r
a
l n
etw
o
r
k
in
tr
u
s
io
n
d
etec
tio
n
fr
a
mewo
r
k
to
cu
r
b
…
(
A
r
n
o
ld
A
d
ima
b
u
a
Oju
g
o
)
1503
2
.
3
.
M
o
del o
ptim
iza
t
io
n
A
m
ajo
r
is
s
u
e
in
ML
im
p
lem
en
tatio
n
is
f
in
e
tu
n
in
g
f
ea
tu
r
e
s
th
at
liv
e
o
u
ts
id
e
th
e
m
o
d
el.
T
h
ese
f
ea
ts
o
f
ten
in
f
lu
en
ce
th
e
m
o
d
el’
s
b
eh
av
io
r
-
r
ip
p
l
in
g
ef
f
ec
ts
ac
r
o
s
h
id
d
en
elem
en
ts
ca
lled
h
y
p
er
-
p
ar
am
eter
s
.
Hy
p
er
-
p
ar
am
eter
s
ar
e
c
r
itical
s
ettin
g
s
th
at
ca
n
b
e
t
u
n
ed
to
co
n
tr
o
l
a
m
o
d
el
’
s
b
eh
av
i
o
r
.
T
h
ey
a
r
e
p
ar
am
eter
s
wh
ich
ar
e
s
p
ec
if
ic
to
t
h
e
ty
p
e
o
f
lea
r
n
in
g
m
o
d
el
we
wis
h
to
o
p
ti
m
ize
[
7
]
.
I
f
a
m
o
d
el
s
ee
k
s
to
lear
n
th
ese
s
ettin
g
s
d
ir
ec
tly
f
r
o
m
a
tr
ain
in
g
d
atas
et
-
th
er
e
is
th
e
lik
elih
o
o
d
f
o
r
th
e
m
o
d
el
to
tr
y
to
m
ax
im
ize
th
ese
p
ar
am
eter
s
-
wh
ich
will
lead
to
o
v
er
-
f
itti
n
g
.
An
d
th
u
s
,
will
r
esu
lt
in
p
o
o
r
g
en
er
aliza
tio
n
[
5
7
,
5
8
]
.
Ma
jo
r
cr
it
er
ia
o
f
h
y
p
er
-
p
ar
am
eter
s
ar
e
[
5
9
-
61
]:
−
L
ea
r
n
in
g
r
ate
is
a
h
y
p
e
r
-
p
ar
a
m
eter
th
at
co
n
tr
o
ls
h
o
w
m
u
c
h
an
d
wh
at
weig
h
ts
n
ee
d
s
to
b
e
ad
ju
s
ted
o
n
o
u
r
n
etwo
r
k
in
lie
u
o
f
g
r
ad
ie
n
t
lo
s
s
.
T
h
e
lo
wer
th
e
v
alu
e
,
th
e
s
l
o
wer
we
tr
av
el
o
n
d
o
wn
war
d
s
lo
p
e.
L
ea
r
n
in
g
r
ate
co
n
n
o
tes
h
o
w
q
u
ick
ly
a
n
et
ab
an
d
o
n
s
o
ld
b
elief
s
f
o
r
n
ew
o
n
es.
I
t
ca
n
eith
er
b
e
u
n
s
u
p
er
v
is
ed
a
n
d
/o
r
s
u
p
er
v
is
ed
lear
n
i
n
g
.
Als
o
,
with
s
m
all/lar
g
e
lear
n
i
n
g
r
ate,
th
e
n
et
q
u
ick
ly
d
if
f
er
e
n
tiates
b
e
twee
n
im
p
o
r
ta
n
t
f
ea
ts
an
d
o
th
er
wis
e
.
Hig
h
e
r
le
ar
n
in
g
r
ate
m
ea
n
s
th
e
n
etwo
r
k
ca
n
c
h
an
g
e
an
d
ad
a
p
t
s
f
lex
ib
l
y
,
m
o
r
e
ea
s
ily
.
T
h
e
m
o
d
el
m
u
s
t b
e
ab
le
t
o
ad
e
q
u
ately
ad
ju
s
t its
lear
n
in
g
r
ate
to
av
o
id
o
v
er
-
f
itti
n
g
an
d
o
v
er
tr
ain
in
g
.
−
B
atch
s
ize
is
th
e
n
u
m
b
er
o
f
tr
ain
in
g
u
tili
ze
d
in
o
n
e
iter
atio
n
.
W
e
ca
n
ad
o
p
t
o
n
e
-
of
-
th
r
ee
o
p
tio
n
s
:
(
i
)
b
atch
m
o
d
e
w
h
er
e
t
h
e
iter
atio
n
an
d
ep
o
c
h
v
alu
es
ar
e
eq
u
al
;
(
ii
)
m
in
i
-
b
atch
u
s
es
a
b
atch
s
ize
g
r
ea
ter
th
a
n
o
n
e
;
(
iii
)
s
to
ch
asti
c
in
wh
ich
th
e
g
r
ad
ien
t a
n
d
t
h
e
n
eu
r
al
n
etwo
r
k
p
ar
am
eter
s
ar
e
u
p
d
ated
a
f
ter
e
ac
h
s
am
p
le.
−
E
p
o
ch
m
e
asu
r
es
th
e
n
u
m
b
er
o
f
tim
es
all
o
f
th
e
tr
ai
n
in
g
v
ec
to
r
s
ar
e
u
s
ed
o
n
ce
to
u
p
d
ate
th
e
weig
h
ts
.
I
t
is
a
s
in
g
le
s
tep
in
tr
ain
in
g
a
n
etwo
r
k
.
T
h
u
s
,
if
a
n
etwo
r
k
is
tr
ain
ed
o
n
ev
e
r
y
tr
ain
in
g
d
ataset
s
am
p
les
in
ju
s
t
one
-
p
ass
,
th
en
an
e
p
o
ch
is
ex
h
au
s
ted
.
A
tr
ain
i
n
g
m
a
y
co
n
s
is
t
o
f
m
o
r
e
th
an
o
n
e
ep
o
c
h
s
.
I
n
b
atch
tr
ain
in
g
,
a
ll
s
am
p
les
f
ilter
th
r
o
u
g
h
t
h
e
lear
n
in
g
m
o
d
el
s
im
u
ltan
eo
u
s
ly
in
o
n
e
e
p
o
ch
with
weig
h
ts
u
p
d
ated
.
C
o
n
v
er
s
ely
,
in
s
eq
u
e
n
tial tr
ain
in
g
,
all
weig
h
ts
ar
e
u
p
d
ated
a
f
ter
ea
ch
tr
ain
in
g
.
3.
RE
SU
L
T
S
A
ND
D
I
SCU
S
S
I
O
NS
3
.
1
.
Da
t
a
s
a
m
pli
ng
A
m
ajo
r
ch
allen
g
e
is
to
g
et
a
d
ataset
p
r
o
p
er
ly
f
o
r
m
atted
f
o
r
th
e
task
at
h
an
d
.
D
ataset
u
s
ed
f
o
r
tr
ain
in
g
(
to
f
it
th
e
m
o
d
el)
m
u
s
t
b
e
s
am
e
f
o
r
ev
alu
atin
g
th
e
m
o
d
el.
Her
e,
we
ad
o
p
t
th
e
Ho
ch
s
ch
u
le
C
o
b
u
r
g
I
DS
d
atasets
(
C
I
DDS
-
2017)
-
a
s
e
t
o
f
lab
ele
d
an
o
m
aly
-
b
ased
I
DS
d
ataset,
s
p
lit
as
th
u
s
:
tr
ain
in
g
(
7
0
%)
an
d
test
in
g
(
3
0
%)
[
2
0
,
2
1
]
.
W
e
th
e
n
ad
o
p
t 8
-
p
ar
a
m
eter
s
to
ad
ju
s
t
weig
h
ts
an
d
co
e
f
f
icien
ts
in
m
in
im
izin
g
er
r
o
r
s
as
in
T
ab
le
1:
T
ab
le
1
.
Selecte
d
f
ea
tu
r
es a
n
d
th
eir
d
ata
ty
p
es
F
e
a
t
u
r
e
s
F
o
r
mat
D
a
t
a
Ty
p
e
s
S
o
u
r
c
e
I
P
a
.
b
.
c
.
d
O
b
j
e
c
t
S
o
u
r
c
e
P
o
r
t
N
u
meri
c
I
n
t
e
g
e
r
D
e
st
i
n
a
t
i
o
n
I
P
a
.
b
.
c
.
d
O
b
j
e
c
t
D
e
st
i
n
a
t
i
o
n
P
o
r
t
N
u
meri
c
F
l
o
a
t
P
r
o
t
o
c
o
l
S
t
r
i
n
g
O
b
j
e
c
t
D
u
r
a
t
i
o
n
H
:
M
:
S
F
l
o
a
t
P
a
c
k
e
t
s
N
u
meri
c
I
n
t
e
g
e
r
A
t
t
a
c
k
N
a
me
/
T
y
p
e
S
t
r
i
n
g
O
b
j
e
c
t
3
.
2
.
E
nco
din
g
s
c
hem
es us
e
d
Un
class
if
ied
an
d
u
n
f
o
r
m
atted
d
ata
a
r
e
o
f
ten
am
b
u
i
g
u
o
s
,
in
co
m
p
lete,
r
ip
p
led
with
n
o
is
e,
im
p
r
ec
is
e
an
d
in
co
n
s
is
ten
t.
E
n
co
d
in
g
s
ee
k
s
to
f
ilter
th
e
d
ataset,
m
ap
p
in
g
it
u
n
t
o
th
e
r
eq
u
ir
ed
f
o
r
m
at
th
e
m
o
d
el
ca
n
ea
s
ily
u
n
d
er
s
tan
d
.
T
o
en
c
o
d
e
th
e
s
elec
te
d
f
ea
ts
,
we
t
r
a
n
s
f
o
r
m
o
u
r
d
ataset
u
s
in
g
th
e
f
ea
ts
o
f
in
ter
est
as
i
n
T
ab
le
1
.
T
h
is
m
o
d
e
will
s
ee
k
to
m
o
d
u
late
th
e
r
aw
d
ata
u
n
t
o
th
e
r
eq
u
ir
e
d
ataset
-
s
o
th
at
d
ata
g
ath
er
ed
f
r
o
m
v
ar
y
in
g
s
o
u
r
ce
s
,
is
ad
eq
u
ate
f
o
r
an
aly
s
is
.
W
e
em
p
lo
y
d
ata
ty
p
e
in
Pan
d
as
lib
r
ar
y
d
is
p
lay
ed
b
y
l
is
tin
g
1
alg
o
r
ith
m
[
2
0
,
2
1
].
Input
: Selected Feature
Output
: Converted Feature Data type
1.
Select Feature
2.
For each Selected Feature
3.
If Selected Feature is Non
-
Numerical then
4.
Generate Category Data type
5.
End if
6.
End For each
List
ing 1: Algori
thm to Convert Data type to Category
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
11
,
No
.
2
,
Ap
r
il 2
0
2
1
:
1
4
9
8
-
1509
1504
3
.
3
.
P
a
ra
m
e
t
er
t
un
ing
W
e
m
o
d
eled
th
e
n
etwo
r
k
u
s
in
g
8
-
n
eu
r
o
n
s
at
th
e
in
p
u
t
lay
er
(
a
n
eu
r
o
n
f
o
r
ea
ch
f
ea
t)
.
2
-
n
e
u
r
o
n
s
wer
e
u
s
ed
f
o
r
o
u
tp
u
t
lay
er
(
a
n
e
u
r
o
n
f
o
r
ea
ch
p
o
s
s
ib
le
class
)
.
T
h
e
p
ar
am
ete
r
s
f
o
r
th
e
d
e
e
p
lea
r
n
in
g
ar
e
th
e
n
u
m
b
er
o
f
ep
o
ch
s
,
th
e
ac
tiv
atio
n
f
u
n
ct
io
n
,
its
lear
n
in
g
r
ate
a
n
d
t
h
e
h
id
d
en
lay
e
r
to
p
o
lo
g
y
.
W
e
em
p
lo
y
ed
th
e
r
ec
tifie
d
lin
ea
r
u
n
it
(
R
eL
U)
ac
tiv
atio
n
f
u
n
ctio
n
with
5
0
0
-
e
p
o
ch
s
(
th
o
u
g
h
o
p
tim
al
v
alu
es
wer
e
r
ea
c
h
ed
at
1
0
0
,
3
0
0
an
d
5
0
0
ep
o
c
h
s
tak
in
g
in
to
ac
co
u
n
t
ac
cu
r
ac
y
a
n
d
tim
e
to
tr
ain
th
e
m
o
d
el)
.
T
h
er
e
is
n
o
b
e
s
t
p
r
ac
tice
in
s
elec
tin
g
th
e
n
u
m
b
er
o
f
h
i
d
d
en
lay
er
s
/n
eu
r
o
n
s
th
er
ein
an
d
u
s
in
g
m
o
r
e
h
id
d
e
n
lay
e
r
(
s
)
g
r
an
ts
th
e
m
o
d
el
g
r
ea
ter
ca
p
ab
ilit
y
to
p
er
f
o
r
m
m
o
r
e
c
o
m
p
lex
f
u
n
ctio
n
o
n
th
e
d
ata
[
1
,
2
]
.
W
e
s
ee
k
m
in
im
u
m
tr
a
in
in
g
er
r
o
r
th
at
will
also
r
esu
lt
in
th
e
b
est
f
i
t,
s
ele
ctin
g
th
e
n
u
m
b
er
o
f
h
i
d
d
en
la
y
er
s
(
an
d
n
e
u
r
o
n
s
f
o
r
ea
ch
la
y
er
)
was
estab
lis
h
ed
v
ia
a
tr
ail
-
an
d
-
e
r
r
o
r
m
eth
o
d
,
an
d
ex
am
in
i
n
g
th
e
r
esu
lts
.
T
h
e
b
est
p
o
s
s
ib
le
n
u
m
b
er
o
f
lay
er
s
was
d
eter
m
in
ed
b
y
r
u
n
n
in
g
test
s
o
n
a
s
in
g
le
lay
er
with
1
to
2
0
n
e
u
r
o
n
s
at
th
e
f
ir
s
t
in
s
tan
ce
s
-
wh
ich
y
ield
ed
th
e
g
r
ea
test
f
-
s
co
r
e
with
th
e
least
(
co
n
s
tan
t)
am
o
u
n
t
o
f
tr
ain
in
g
lo
s
s
tim
e.
Ad
d
iti
o
n
o
f
a
s
ec
o
n
d
h
id
d
e
n
lay
er
o
f
n
eu
r
o
n
s
f
r
o
m
1
to
2
0
y
ield
e
d
s
co
r
es.
F
in
ally
,
th
e
ad
d
itio
n
o
f
a
th
ir
d
h
i
d
d
en
lay
er
u
s
in
g
th
e
b
est
p
o
s
s
ib
le
n
u
m
b
e
r
o
f
n
eu
r
o
n
s
p
r
o
d
u
ce
d
th
e
g
r
ea
test
f
-
s
co
r
e
an
d
th
u
s
,
was
s
elec
ted
as
th
e
o
v
er
all
b
est
p
o
s
s
ib
le
h
id
d
en
lay
er
co
n
f
ig
u
r
atio
n
.
R
esu
lts
o
f
th
e
f
ir
s
t
h
id
d
en
lay
er
ar
e
s
ee
n
in
T
a
b
le
2
.
T
a
b
l
e
2
s
h
o
ws
r
esu
lt
o
f
th
e
f
ir
s
t
h
id
d
en
lay
e
r
with
co
n
f
ig
u
r
atio
n
o
f
9
-
n
eu
r
o
n
s
a
n
d
f
-
s
co
r
e
o
f
9
2
%
at
1
8
th
-
ite
r
atio
n
an
d
tr
ain
in
g
lo
s
s
o
f
1
.
1
4
0
.
F
-
s
co
r
e
s
h
o
ws
ac
cu
r
ac
y
o
f
ea
ch
r
u
n
-
s
in
ce
w
e
u
s
ed
an
u
n
b
alan
ce
d
d
ataset
to
tr
ain
/tes
t
m
o
d
el
with
m
o
r
e
r
ec
o
r
d
s
in
n
o
r
m
al
class
th
an
in
m
alicio
u
s
class
.
T
ab
le
2
.
First h
id
d
e
n
lay
er
c
o
n
f
ig
u
r
at
io
n
an
aly
s
is
H
i
d
d
e
n
L
a
y
e
r
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
-
S
c
o
r
e
I
t
e
r
a
t
i
o
n
Tr
a
i
n
i
n
g
L
o
ss
Ep
o
c
h
1
0
.
8
4
0
.
9
2
0
.
8
8
44
0
.
2
9
4
5
0
0
2
0
.
8
4
0
.
9
2
0
.
8
7
24
0
.
2
7
8
5
0
0
3
0
.
8
4
0
.
9
2
0
.
8
8
26
0
.
2
9
3
5
0
0
4
0
.
8
4
0
.
9
2
0
.
8
8
9
0
.
5
0
1
5
0
0
5
0
.
8
9
0
.
5
5
0
.
6
4
19
1
.
4
9
6
5
0
0
6
0
.
9
4
0
.
9
4
0
.
9
2
18
1
.
4
0
0
5
0
0
7
0
.
8
6
0
.
5
3
0
.
6
3
4
2
.
2
3
0
5
0
0
8
0
.
9
0
0
.
8
4
0
.
8
6
16
2
.
0
7
1
5
0
0
9
0
.
9
2
0
.
9
3
0
.
9
2
18
1
.
1
4
0
5
0
0
10
0
.
9
2
0
.
9
2
0
.
9
0
16
1
.
7
7
9
5
0
0
11
0
.
8
8
0
.
9
1
0
.
8
9
7
2
.
1
3
4
5
0
0
12
0
.
9
1
0
.
9
2
0.
89
8
2
.
3
2
0
5
0
0
13
0
.
8
7
0
.
8
7
0
.
8
7
13
2
.
0
0
6
5
0
0
14
0
.
9
2
0
.
9
2
0
.
9
0
8
1
.
9
7
0
5
0
0
15
0
.
9
2
0
.
9
2
0
.
9
0
5
1
.
7
3
0
5
0
0
16
0
.
8
5
0
.
8
5
0
.
8
5
10
1
.
5
4
0
5
0
0
17
0
.
9
0
0
.
8
4
0
.
8
6
15
1
.
4
4
0
5
0
0
18
0
.
9
1
0
.
9
2
0
.
9
0
8
2
.
3
2
0
5
0
0
19
0
.
9
2
0
.
9
3
0
.
9
0
14
2
.
1
6
0
5
0
0
20
0
.
9
1
0
.
9
1
0
.
9
1
5
1
.
7
7
2
5
0
0
T
ab
le
3
s
h
o
ws
f
ir
s
t
lay
er
h
av
in
g
9
-
n
eu
r
o
n
s
an
d
o
t
h
er
s
n
e
u
r
o
n
s
v
ar
y
in
g
f
r
o
m
1
to
2
0
.
W
ith
h
id
d
e
n
lay
er
o
f
9
an
d
1
1
n
eu
r
o
n
s
y
i
eld
in
g
f
-
s
co
r
e
o
f
9
3
%
an
d
tr
ain
in
g
lo
s
s
o
f
0
.
3
9
.
T
h
e
s
ec
o
n
d
h
id
d
en
lay
er
is
f
av
o
r
e
d
as
it
y
ield
s
g
r
ea
ter
f
-
s
co
r
e.
T
ab
l
e
4
s
h
o
ws
th
ir
d
co
n
f
ig
u
r
atio
n
with
f
ir
s
t
an
d
s
ec
o
n
d
lay
er
h
av
in
g
9
an
d
1
1
n
o
d
es
an
d
v
ar
y
in
g
th
i
r
d
h
id
d
en
lay
er
.
B
est
co
n
f
ig
u
r
a
tio
n
is
9
-
11
-
1
4
n
e
u
r
o
n
s
,
y
ield
i
n
g
f
-
s
co
r
e
o
f
9
2
%
with
a
tr
ain
in
g
lo
s
s
at
0
.
5
6
0
.
3
.
4
.
M
o
del e
v
a
lua
t
io
n
W
e
u
s
e
th
e
ac
cu
r
ac
y
,
r
ec
all
an
d
er
r
o
r
r
at
e
(
s
)
to
e
v
alu
ate
m
o
d
el
p
er
f
o
r
m
an
ce
as in
(
5
)
to
(
7
)
:
=
+
+
+
+
(
5
)
=
+
(
6
)
=
+
+
+
+
(
7
)
On
ev
alu
atin
g
o
u
r
p
a
r
am
ete
r
s
,
r
esu
lt
o
f
th
e
m
o
d
el
is
g
iv
en
in
a
co
n
f
u
s
io
n
m
atr
i
x
an
d
th
e
class
if
icatio
n
r
ep
o
r
t.
T
h
e
r
esu
ltin
g
class
if
icatio
n
r
ep
o
r
t
a
n
d
co
n
f
u
s
io
n
m
atr
ix
is
g
i
v
en
i
n
t
h
e
T
ab
les
5
an
d
6
r
esp
ec
tiv
ely
.
T
ab
le
5
s
h
o
ws
m
o
d
el
h
as
p
r
ed
ictio
n
ac
cu
r
ac
y
o
f
9
4
-
p
e
r
ce
n
t
(
0
.
9
4
)
with
an
i
m
p
r
o
v
e
m
en
t
r
ate
o
f
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2
0
8
8
-
8
7
0
8
F
o
r
g
in
g
a
d
ee
p
le
a
r
n
in
g
n
e
u
r
a
l n
etw
o
r
k
in
tr
u
s
io
n
d
etec
tio
n
fr
a
mewo
r
k
to
cu
r
b
…
(
A
r
n
o
ld
A
d
ima
b
u
a
Oju
g
o
)
1505
97
-
p
er
ce
n
t.
I
t
also
h
as
a
m
is
cl
ass
if
icatio
n
er
r
o
r
r
at
e
o
f
4
1
-
p
er
ce
n
t
f
o
r
d
ata
in
clu
s
io
n
th
at
wer
e
n
o
t
o
r
ig
in
ally
u
s
ed
to
tr
ain
th
e
m
o
d
el.
T
ab
le
3
.
Seco
n
d
h
id
d
en
lay
e
r
co
n
f
ig
u
r
atio
n
a
n
aly
s
is
H
i
d
d
e
n
L
a
y
e
r
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
-
S
c
o
r
e
I
t
e
r
a
t
i
o
n
Tr
a
i
n
i
n
g
L
o
ss
Ep
o
c
h
9
,
1
0
.
8
4
0
.
9
2
0
.
8
8
25
0
.
2
9
3
5
0
0
9
,
2
0
.
8
4
0
.
9
2
0
.
8
8
29
0
.
2
9
2
5
0
0
9
,
3
0
.
9
1
0
.
9
2
0
.
9
1
15
0
.
5
8
3
5
0
0
9
,
4
0
.
8
7
0
.
8
7
0
.
8
7
5
1
.
0
5
8
5
0
0
9
,
5
0
.
9
2
0
.
9
2
0
.
9
0
13
1
.
6
2
8
5
0
0
9
,
6
0
.
9
1
0
.
9
2
0
.
8
9
10
1
.
9
9
6
5
0
0
9
,
7
0
.
8
4
0
.
9
2
0
.
8
8
24
0
.
2
8
1
5
0
0
9
,
8
0
.
9
3
0
.
9
3
0
.
9
2
11
1
.
8
8
4
5
0
0
9
,
9
0
.
9
2
0
.
9
2
0
.
8
9
12
1
.
5
9
0
5
0
0
9
,
1
0
0
.
9
0
0
.
9
2
0
.
9
0
12
1
.
7
3
1
5
0
0
9
,
1
1
0
.
9
5
0
.
9
4
0
.
9
3
14
0
.
3
9
0
5
0
0
9
,
1
2
0
.
9
3
0
.
9
3
0
.
9
1
12
1
.
1
3
0
5
0
0
9
,
1
3
0
.
9
1
0
.
9
2
0
.
9
1
20
1
.
9
2
9
5
0
0
9
,
1
4
0
.
9
2
0
.
9
3
0
.
9
0
13
2
.
2
3
7
5
0
0
9
,
1
5
0
.
9
4
0
.
9
4
0
.
9
2
7
1
.
7
6
5
5
0
0
9
,
1
6
0
.
8
5
0
.
5
2
0
.
6
2
7
2
.
0
1
0
5
0
0
9
,
1
7
0
.
9
4
0
.
9
4
0
.
9
4
6
1
.
6
2
0
5
0
0
9
,
1
8
0
.
9
3
0
.
9
4
0
.
9
2
7
1
.
7
6
0
5
0
0
9
,
1
9
0
.
8
6
0
.
.
7
4
0
.
7
9
13
2
.
0
5
9
5
0
0
9
,
2
0
0
.
9
2
0
.
9
2
0
.
8
9
8
2
.
4
2
1
5
0
0
T
ab
le
4
.
T
h
ir
d
h
i
d
d
en
la
y
er
co
n
f
ig
u
r
atio
n
an
aly
s
is
H
i
d
d
e
n
L
a
y
e
r
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
-
S
c
o
r
e
I
t
e
r
a
t
i
o
n
Tr
a
i
n
i
n
g
L
o
ss
Ep
o
c
h
9
,
1
1
,
1
0
.
8
3
0
.
9
1
0
.
8
7
32
0
.
2
8
7
5
0
0
9
,
1
1
,
2
0
.
9
1
0
.
9
2
0
.
8
9
6
1
.
5
9
2
5
0
0
9
,
1
1
,
3
0
.
8
3
0
.
9
1
0
.
8
7
29
0
.
2
8
0
5
0
0
9
,
1
1
,
4
0
.
9
0
0
.
9
1
0
.
9
0
16
1
.
5
6
4
5
0
0
9
,
1
1
,
5
0
.
9
2
0
.
9
2
0
.
9
0
18
0
.
7
4
1
5
0
0
9
,
1
1
,
6
0
.
9
3
0
.
9
2
0
.
8
9
21
0
.
2
8
2
5
0
0
9
,
1
1
,
7
0
.
9
2
0
.
9
3
0
.
9
0
6
1
.
3
2
2
5
0
0
9
,
1
1
,
8
0
.
9
0
0
.
8
6
0
.
8
8
6
1
.
2
3
9
5
0
0
9
,
1
1
,
9
0
.
9
0
0
.
9
1
0
.
9
0
7
1
.
8
8
6
5
0
0
9
,
1
1
,
1
0
0
.
8
8
0
.
9
1
0
.
8
9
8
0
.
6
2
3
5
0
0
9
,
1
1
,
1
1
0
.
9
2
0
.
9
3
0
.
9
1
5
2
.
0
0
0
5
0
0
9
,
1
1
,
1
2
0
.
8
6
0
.
8
3
0
.
8
5
11
2
.
3
7
0
5
0
0
9
,
1
1
,
1
3
0
.
8
6
0
.
8
3
0
.
8
4
8
2
.
3
5
0
5
0
0
9
,
1
1
,
1
4
0
.
9
3
0
.
9
2
0
.
9
2
15
0
.
5
6
0
5
0
0
9
,
1
1
,
1
5
0
.
9
3
0
.
9
3
0
.
9
1
8
1
.
2
0
4
5
0
0
9
,
1
1
,
1
6
0
.
9
4
0
.
9
4
0
.
9
2
8
1
.
7
3
0
5
0
0
9
,
1
1
,
1
7
0
.
8
7
0
.
5
4
0
.
6
3
12
1
.
7
3
0
5
0
0
9
,
1
1
,
1
8
0
.
9
3
0
.
9
4
0
.
9
3
6
1
.
8
5
0
5
0
0
9
,
1
1
,
1
9
0
.
9
3
0
.
9
3
0
.
9
0
9
0
.
6
6
0
5
0
0
9
,
1
1
,
2
0
0
.
9
2
0
.
9
2
0
.
9
0
28
1
.
1
8
0
5
0
0
T
ab
le
5
.
C
lass
if
icatio
n
r
ep
o
r
t
b
ef
o
r
e
p
r
e
-
p
r
o
ce
s
s
in
g
test
d
ata
s
et
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
-
S
c
o
r
e
S
u
p
p
o
r
t
0
0
.
9
4
1
.
0
0
0
.
9
7
1
1
4
1
1
1
0
.
9
0
0
.
2
7
0
.
4
1
1
.
0
5
9
A
v
g
/
T
o
t
a
l
0
.
9
3
0
.
9
4
0
.
9
2
12
.
500
Als
o
,
T
ab
le
6
s
h
o
ws
th
at
1
1
.
4
1
0
in
s
tan
ce
s
o
f
th
e
d
ataset
we
r
e
co
r
r
ec
tly
class
if
ied
.
T
h
at
is
,
r
esu
lts
o
f
th
e
test
d
ataset
(
with
1
2
.
5
0
0
p
o
in
ts
)
s
h
o
w
th
at
we
h
av
e
1
1
.
4
1
1
b
en
ig
n
in
s
tan
ce
s
in
th
e
f
ir
s
t
class
(
lab
el
0
)
.
T
h
e
m
o
d
el
s
u
cc
ess
f
u
lly
id
e
n
tifie
d
1
1
.
4
1
0
c
o
r
r
ec
tly
class
if
ied
an
d
i
d
en
tifie
d
b
en
ig
n
in
s
tan
ce
s
as
tr
u
e
-
p
o
s
itiv
es;
but
,
3
1
-
ca
s
es
in
co
r
r
ec
tly
id
en
tifie
d
b
en
ig
n
in
s
tan
ce
s
wer
e
m
ar
k
ed
as
f
alse
-
p
o
s
itiv
e.
Similar
ly
,
on
th
e
s
ec
o
n
d
r
o
w,
th
er
e
wer
e
1
.
0
5
9
m
alicio
u
s
in
s
tan
ce
s
in
s
ec
o
n
d
class
(
lab
el
1
)
; Bu
t,
7
7
6
-
in
co
r
r
ec
tly
id
en
tifie
d
m
alicio
u
s
in
s
tan
ce
s
wer
e
m
ar
k
ed
as
f
alse
-
n
eg
ativ
e,
a
n
d
2
8
3
co
r
r
ec
tly
i
d
en
tifie
d
m
alicio
u
s
in
s
tan
ce
s
o
f
th
em
wer
e
m
ar
k
e
d
as
tr
u
e
-
n
e
g
ativ
e.
T
h
ese
ar
e
f
u
r
th
er
ex
p
lain
e
d
as:
(
i
)
Fo
r
tr
u
e
p
o
s
itiv
e,
t
h
e
m
o
d
el
p
r
ed
icted
p
o
s
itiv
e
an
d
it
was
tr
u
e
;
(
ii
)
Fo
r
tr
u
e
n
e
g
ativ
e,
th
e
m
o
d
el
p
r
ed
icted
n
e
g
ativ
e
an
d
it
was
tr
u
e
;
(
iii
)
Fo
r
f
alse
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
11
,
No
.
2
,
Ap
r
il 2
0
2
1
:
1
4
9
8
-
1509
1506
p
o
s
itiv
e,
th
e
m
o
d
el
p
r
ed
icted
p
o
s
itiv
e
a
n
d
it
was
f
alse
;
a
n
d
(
iv
)
f
o
r
f
alse
n
eg
ativ
e,
th
e
m
o
d
el
p
r
ed
icted
n
eg
ativ
e
an
d
it wa
s
f
alse (
as a
g
r
ee
d
b
y
[
6
2
-
64
]
)
.
T
ab
le
6
.
C
o
n
f
u
s
io
n
m
atr
ix
r
ep
o
r
t
A
c
t
u
a
l
V
a
l
u
e
s
P
r
e
d
i
c
t
e
d
11
.
410
31
V
a
l
u
e
s
7
7
6
2
8
3
T
h
u
s
,
we
ca
n
s
ay
t
h
at
th
e
m
o
d
el
p
r
ed
icts
th
e
r
esu
lts
o
f
eith
er
it’s
a
n
o
r
m
al
attac
k
o
r
DDo
S
attac
k
9
2
%
ac
cu
r
ately
u
s
in
g
th
e
to
t
al
v
alu
e
o
f
th
e
f
-
s
co
r
e
.
T
h
e
n
eu
r
al
n
etwo
r
k
in
Py
th
o
n
m
ay
h
av
e
d
if
f
ic
u
lty
co
n
v
er
g
in
g
b
ef
o
r
e
th
e
m
ax
im
u
m
n
u
m
b
e
r
o
f
iter
atio
n
s
allo
wed
if
th
e
d
ata
is
n
o
t
s
tan
d
ar
d
ized
.
Fo
r
a
m
o
r
e
m
ea
n
in
g
f
u
l
r
esu
lt,
we
d
ec
id
ed
to
s
ca
le
o
u
r
test
d
ata.
T
h
er
e
ar
e
a
lo
t
o
f
d
if
f
er
en
t
m
eth
o
d
s
f
o
r
s
tan
d
a
r
d
izatio
n
o
f
d
ata,
we
will
u
s
e
t
h
e
b
u
ilt
-
in
Stan
d
ar
d
Scaler
f
o
r
s
tan
d
ar
d
izatio
n
.
T
h
e
r
esu
lt
g
o
tten
af
t
er
th
i
s
p
r
o
ce
s
s
was
d
o
n
e
is
s
lated
as in
T
ab
le
7
b
e
lo
w:
T
ab
le
7
.
C
lass
if
icatio
n
r
ep
o
r
t
af
ter
p
r
e
-
p
r
o
ce
s
s
in
g
test
d
ataset
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
-
S
c
o
r
e
S
u
p
p
o
r
t
0
1
.
0
0
1
.
0
0
1
.
0
0
11
.
449
1
1
.
0
0
0
.
9
8
0
.
9
9
1
.
0
5
1
A
v
g
/
T
o
t
a
l
1
.
0
0
0
.
9
9
0
.
9
9
12
.
500
Fro
m
T
ab
le
8
(
co
n
f
u
s
io
n
m
atr
ix
)
i.e
.
th
e
p
r
ed
icted
r
esu
lts
,
u
s
in
g
o
u
r
test
d
ata
with
1
2
.
5
0
0
p
o
in
ts
we
h
av
e
1
1
.
4
4
9
b
en
ig
n
i
n
s
tan
ce
s
in
th
e
f
ir
s
t
class
(
lab
el
0
)
.
O
u
t
o
f
th
is
,
th
e
m
o
d
el
s
u
cc
ess
f
u
lly
id
en
tifie
d
1
1
.
449
co
r
r
ec
tly
id
en
tifie
d
b
en
ig
n
in
s
tan
ce
s
as
a
T
r
u
e
Po
s
itiv
e
b
u
t
th
er
e
was
n
o
in
co
r
r
ec
tly
id
en
tifie
d
b
e
n
ig
n
in
s
tan
ce
s
wh
ich
is
s
u
p
p
o
s
ed
t
o
b
e
m
ar
k
e
d
as
Fals
e
Po
s
itiv
e.
Similar
ly
,
lo
o
k
in
g
at
th
e
s
ec
o
n
d
r
o
w,
th
er
e
wer
e
1.
0
5
1
m
alicio
u
s
in
s
tan
ce
s
in
s
ec
o
n
d
class
(
lab
el
1
)
b
u
t
2
4
i
n
co
r
r
ec
tly
id
e
n
tifie
d
m
alicio
u
s
in
s
tan
ce
s
o
f
t
h
em
wer
e
m
ar
k
ed
as
f
alse
n
e
g
ativ
e
an
d
1
.
0
2
7
co
r
r
ec
tly
id
en
tifie
d
m
alicio
u
s
in
s
tan
ce
s
o
f
th
em
wer
e
m
ar
k
ed
as
tr
u
e
n
eg
ativ
e.
T
ab
le
8
.
C
o
n
f
u
s
io
n
m
atr
ix
r
ep
o
r
t
A
c
t
u
a
l
V
a
l
u
e
s
P
r
e
d
i
c
t
e
d
1
1
4
4
9
0
V
a
l
u
e
s
24
1
0
2
7
T
h
u
s
,
we
ca
n
s
ay
th
at
t
h
e
m
o
d
el
p
r
ed
icts
th
e
r
esu
lts
o
f
eith
er
it
is
a
‘
n
o
r
m
al’
attac
k
o
r
‘
DDo
S
’
attac
k
9
9
%
ac
cu
r
ately
u
s
in
g
th
e
to
t
al
v
alu
e
o
f
t
h
e
f
-
s
co
r
e
.
I
n
t
u
r
n
,
th
is
r
esu
lted
i
n
p
r
e
d
ictin
g
1
.
0
2
7
p
o
i
n
ts
as
m
alicio
u
s
s
am
p
les
an
d
1
1
.
4
4
9
p
o
in
ts
as
n
o
r
m
al
s
am
p
les
f
r
o
m
o
u
r
test
d
ata
.
Fu
r
th
e
r
m
o
r
e
,
th
e
s
tan
d
ar
d
izatio
n
o
f
o
u
r
test
d
ata
p
r
o
v
ed
t
o
b
e
m
o
r
e
ef
f
icien
t th
a
n
th
e
p
r
ev
io
u
s
test
r
u
n
th
at
was n
o
t stan
d
ar
d
ized
.
4.
CO
NCLU
SI
O
N
Ou
r
DNN
m
o
d
el
s
o
lu
tio
n
h
as
a
to
tal
o
f
5
6
-
r
u
les
with
to
p
r
u
les
f
o
u
n
d
to
h
av
e
class
if
icatio
n
ac
cu
r
ac
y
r
an
g
e
[
0
.
8
,
0
.
9
6
]
.
T
h
is
i
m
p
lies
th
at
an
esti
m
ated
o
v
er
8
0
%
o
f
th
e
r
u
les
ca
n
ad
e
q
u
ately
c
lass
if
y
th
e
d
ataset.
Ach
iev
in
g
a
s
et
o
f
g
o
o
d
r
u
les,
is
m
u
ch
b
etter
th
an
a
s
in
g
le
o
p
tim
u
m
r
u
le.
T
h
is
in
cr
ea
s
es
th
e
ch
an
ce
s
o
f
d
etec
tin
g
m
alicio
u
s
d
ata
p
ac
k
ets as we
ll a
s
also
im
p
r
o
v
es th
e
g
en
er
a
lity
o
f
r
u
les,
p
r
o
v
id
i
n
g
th
e
ab
ilit
y
f
o
r
n
ew
d
ataset
an
d
th
eir
co
r
r
esp
o
n
d
i
n
g
g
en
er
ate
d
r
u
les
to
b
e
a
d
d
ed
to
th
e
k
n
o
wled
g
eb
ase.
T
h
e
i
m
p
ac
t
o
f
th
e
DDo
S
attac
k
s
to
u
s
er
s
r
e
q
u
ir
es
a
c
o
n
ce
r
ted
e
f
f
o
r
t
to
d
etec
t
in
tr
u
s
io
n
.
Dete
ctio
n
s
ch
em
es
s
im
p
l
y
f
ilter
th
r
o
u
g
h
th
e
n
etwo
r
k
r
e
q
u
est,
an
aly
ze
th
e
m
to
d
ec
id
e
wh
ich
clien
ts
ar
e
u
n
co
m
p
r
o
m
is
ed
an
d
co
m
p
r
o
m
is
ed
,
an
d
u
ltima
tel
y
m
et
o
u
t
in
ten
d
ed
s
af
ety
m
ea
s
u
r
es
f
o
r
f
u
r
th
er
ac
tio
n
s
.
T
h
eir
p
er
f
o
r
m
a
n
ce
ca
n
b
e
h
in
d
er
ed
as
p
r
em
is
ed
o
n
th
eir
er
r
o
r
r
ate
f
o
r
in
c
o
r
r
e
ctly
cla
s
s
if
ied
a
n
d
u
n
id
en
tifie
d
d
ata
-
p
o
in
ts
th
at
s
ch
em
e/m
o
d
el
g
en
er
ates.
An
id
ea
l
s
ch
em
e
will
co
r
r
ec
tly
class
if
y
all
r
eq
u
est
an
d
p
ac
k
ets
with
alm
o
s
t
ze
r
o
er
r
o
r
r
at
es
o
f
f
alse
p
o
s
itiv
e/n
eg
ativ
e
th
r
o
u
g
h
tr
ad
e
o
f
f
s
b
etwe
en
t
h
e
n
u
m
b
er
o
f
f
alse p
o
s
itiv
es a
n
d
f
a
ls
e
n
eg
ativ
es.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2
0
8
8
-
8
7
0
8
F
o
r
g
in
g
a
d
ee
p
le
a
r
n
in
g
n
e
u
r
a
l n
etw
o
r
k
in
tr
u
s
io
n
d
etec
tio
n
fr
a
mewo
r
k
to
cu
r
b
…
(
A
r
n
o
ld
A
d
ima
b
u
a
Oju
g
o
)
1507
RE
F
E
R
E
NC
E
S
[1
]
Oju
g
o
,
A.
A.
a
n
d
E
b
o
k
a
,
A.
O.,
“
In
v
e
n
to
r
y
p
re
d
ictio
n
a
n
d
m
a
n
a
g
e
m
e
n
t
in
Nig
e
ria
u
sin
g
m
a
rk
e
t
b
a
sk
e
t
a
n
a
ly
si
s
a
ss
o
c
iativ
e
ru
le
m
in
in
g
:
m
e
m
e
ti
c
a
p
p
ro
a
c
h
,”
In
ter
n
a
ti
o
n
a
l
J
o
u
rn
a
l
o
f
In
f
o
rm
a
ti
c
s
a
n
d
C
o
mm
u
n
ic
a
t
io
n
T
e
c
h
n
o
l
o
g
y
(IJ
-
ICT
)
,
v
o
l.
8
,
n
o
.
3
,
p
p
.
1
2
8
-
1
3
8
,
2
0
1
9
.
[2
]
Oju
g
o
,
A
.
A.
a
n
d
Eb
o
k
a
,
A.
O.,
“
S
i
g
n
a
t
u
re
-
b
a
se
d
m
a
lwa
re
d
e
t
e
c
ti
o
n
u
sin
g
a
p
p
r
o
x
ima
te
Bo
y
e
r
M
o
o
re
str
in
g
m
a
tch
in
g
a
l
g
o
ri
th
m
,
”
I
n
ter
n
a
ti
o
n
a
l
J
o
u
rn
a
l
o
f
M
a
th
e
ma
ti
c
a
l
S
c
ie
n
c
e
s
a
n
d
C
o
mp
u
ti
n
g
,
v
o
l.
3
,
n
o
.
5
,
p
p
.
4
9
-
6
2
,
2
0
1
9
.
[3
]
P
a
x
so
n
,
V.
,
“
An
An
a
l
y
sis
o
f
Us
in
g
Re
flec
to
rs
fo
r
Distri
b
u
te
d
De
n
ial
-
of
-
S
e
rv
ice
At
tac
k
s,”
A
CM
S
IGCO
M
M
Co
mp
u
ter
C
o
mm
u
n
ica
ti
o
n
Rev
ie
w,
v
o
l.
3
1
,
n
o
.
3
,
p
p
.
3
8
-
4
7
,
2
0
0
1
.
[4
]
Nu
ru
z
z
a
m
a
n
,
T.
M
.
,
Lee
,
C.
,
Ab
d
u
ll
a
h
,
M
.
F
.
A.,
Ch
o
i,
D.,
“
S
imp
le
S
M
S
s
p
a
m
fil
terin
g
o
n
in
d
e
p
e
n
d
e
n
t
m
o
b
il
e
p
h
o
n
e
,
”
J
o
u
rn
a
l
o
f
S
e
c
u
rity
a
n
d
Co
mm
u
n
ica
ti
o
n
Ne
two
rk
s
,
v
o
l
.
5
,
n
o
.
1
0
,
p
p
.
1
2
0
9
-
1
2
2
0
,
2
0
1
2
.
[5
]
Na
ra
y
a
n
,
A.
a
n
d
S
a
x
e
n
a
,
P
.
,
“
T
h
e
c
u
rse
o
f
1
4
0
c
h
a
ra
c
ters
:
Ev
a
lu
a
ti
n
g
th
e
e
ffica
c
y
o
f
S
M
S
s
p
a
m
d
e
tec
ti
o
n
o
n
a
n
d
ro
i
d
,
”
S
PS
M
'1
3
:
Pro
c
e
e
d
in
g
s
o
f
th
e
T
h
ird
AC
M
wo
rk
sh
o
p
o
n
S
e
c
u
rity
a
n
d
p
riv
a
c
y
i
n
sm
a
rtp
h
o
n
e
s
&
mo
b
il
e
d
e
v
ice
s
,
2
0
1
3
,
p
p
.
3
3
-
4
2
.
[
6
]
D
a
d
k
h
a
h
,
M
.
,
a
n
d
S
u
t
i
k
n
o
,
T
.
,
“
P
h
i
s
h
i
n
g
o
r
h
i
j
a
c
k
i
n
g
?
F
o
r
g
e
r
s
h
i
j
a
c
k
e
d
D
U
j
o
u
r
n
a
l
b
y
c
o
p
y
i
n
g
c
o
n
t
e
n
t
o
f
a
n
o
t
h
e
r
a
u
t
h
e
n
t
i
c
a
t
e
j
o
u
r
n
a
l
,
”
I
n
d
o
n
e
s
i
a
n
J
o
u
r
n
a
l
o
f
E
l
e
c
t
r
i
c
a
l
E
n
g
i
n
e
e
r
i
n
g
a
n
d
I
n
f
o
r
m
a
t
i
c
s
(
I
J
E
E
I
)
,
v
o
l
.
3
,
n
o
.
3
,
p
p
.
1
1
9
-
1
2
0
,
2015.
[7
]
Oju
g
o
,
A.
A.
,
a
n
d
D.
O.
Ota
k
o
r
e
,
“
Im
p
ro
v
e
d
e
a
rl
y
d
e
tec
ti
o
n
o
f
g
e
sta
ti
o
n
a
l
d
ia
b
e
tes
v
ia
in
telli
g
e
n
t
c
las
sifica
ti
o
n
m
o
d
e
ls:
a
c
a
se
o
f
Nig
e
r
De
lt
a
,
”
J
o
u
rn
a
l
o
f
C
o
mp
u
ter
S
c
ien
c
e
&
Ap
p
li
c
a
ti
o
n
,
v
o
l.
6
,
n
o
.
2
,
p
p
.
82
-
9
0
,
2
0
1
8
.
[8
]
Oju
g
o
,
A.
A.
,
a
n
d
E
b
o
k
a
,
A.
O.,
“
M
e
m
e
ti
c
a
lg
o
rit
h
m
fo
r
sh
o
r
t
m
e
ss
a
g
in
g
se
rv
ice
sp
a
m
fil
ter
tex
t
n
o
rm
a
li
z
a
ti
o
n
a
n
d
se
m
a
n
ti
c
a
p
p
ro
a
c
h
,
”
In
ter
n
a
ti
o
n
a
l
J
o
u
rn
a
l
o
f
I
n
fo
rm
a
ti
c
s
a
n
d
Co
mm
u
n
ica
ti
o
n
T
e
c
h
n
o
lo
g
y
(IJ
-
ICT
)
,
v
o
l
.
9
,
n
o
.
1
,
p
p
.
1
3
-
2
7
,
2
0
2
0
.
[9
]
Ha
sib
,
S
.
,
M
o
twa
n
i,
M
.
,
S
a
x
e
n
a
,
A.,
“
An
ti
-
S
p
a
m
M
e
th
o
d
o
lo
g
ies
:
A
Co
m
p
a
ra
ti
v
e
S
tu
d
y
,
”
In
ter
n
a
ti
o
n
a
l
J
o
u
rn
a
l
o
f
Co
mp
u
ter
S
c
ien
c
e
a
n
d
I
n
fo
rm
a
ti
o
n
T
e
c
h
n
o
l
o
g
ies
,
v
o
l
.
3
,
n
o
.
6
,
p
p
.
5
3
4
1
-
5
3
4
5
,
2
0
1
2
.
[1
0
]
Oju
g
o
,
A.
A
a
n
d
Eb
o
k
a
,
A.
O.
,
“
Co
m
p
a
ra
ti
v
e
e
v
a
l
u
a
ti
o
n
f
o
r
h
ig
h
in
tell
i
g
e
n
t
p
e
rfo
rm
a
n
c
e
a
d
a
p
ti
v
e
m
o
d
e
l
f
o
r
sp
a
m
p
h
ish
in
g
d
e
tec
ti
o
n
,
”
Di
g
it
a
l
T
e
c
h
n
o
l
o
g
ies
,
v
o
l
.
3
,
n
o
.
1
,
p
p
.
9
-
1
5
,
2
0
1
8
.
[1
1
]
Tri
n
it
y
S
e
c
u
rit
y
S
e
r
v
ice
s,
“
Re
tri
e
v
e
d
fro
m
T
h
e
Distrib
u
ted
De
n
i
a
l
o
f
S
e
rv
ice
Attac
k
,
”
2
0
0
3
.
[On
li
n
e
].
Av
a
il
a
b
le:
h
tt
p
:
//
a
rc
h
iv
e
.
n
e
two
r
k
n
e
wz
.
c
o
m
/
n
e
two
rk
n
e
wz
-
10
-
2
0
0
3
0
9
2
4
T
h
e
Distrib
u
te
d
De
n
ial
o
f
S
e
rv
ice
Attac
k
.
h
tml.
[1
2
]
Crisc
u
o
lo
P
.
J.,
“
Distri
b
u
te
d
De
n
ial
o
f
S
e
rv
ice
,
Tri
b
e
F
lo
o
d
Ne
t
wo
rk
,
a
n
d
S
tac
h
e
ld
ra
h
t
CIAC
-
2
3
1
9
,
”
L
a
wre
n
c
e
L
ive
rm
o
re
Na
ti
o
n
a
l
L
a
b
o
r
a
to
ry
,
2
0
1
0
.
[1
3
]
M
o
n
o
wa
r
H.
Bh
u
y
a
n
,
H.
Ka
sh
y
a
p
,
D.
K.
B
h
a
tt
a
c
h
a
ry
y
a
,
a
n
d
J.
K.
Ka
li
ta,
“
De
tec
ti
n
g
Distrib
u
ted
D
e
n
ial
o
f
S
e
rv
ice
Attac
k
s: M
e
th
o
d
s,
T
o
o
ls an
d
F
u
tu
re
Dire
c
ti
o
n
s,”
T
h
e
Co
mp
u
ter
J
o
u
rn
a
l
,
v
o
l.
5
7
,
n
o
.
4
,
p
p
.
4
3
7
-
5
5
6
,
2
0
1
4
.
[1
4
]
M
u
n
i
v
a
ra
,
P
.
K.,
Ra
m
a
,
M
.
,
M
o
h
a
n
,
R.
A.,
Ve
n
u
g
o
p
a
l,
R
.
K.
,
“
Do
S
a
n
d
DD
o
S
Att
a
c
k
s:
De
fe
n
se
,
De
tec
ti
o
n
a
n
d
T
r
a
c
e
b
a
c
k
-
A
S
u
r
v
e
y
,
”
G
l
o
b
a
l
J
o
u
r
n
a
l
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
T
e
c
h
n
o
l
o
g
y
(
G
J
C
S
T
)
,
v
o
l
.
1
4
,
n
o
.
7
,
p
p
.
1
5
-
3
1
,
2
0
1
4
.
[1
5
]
Oju
g
o
,
A.
A.,
E.
Be
n
-
Iw
h
iwh
u
,
O.
Ke
k
e
je.,
M
.
Ye
ro
k
u
n
.
,
I.
I
y
a
wa
h
,
“
M
a
lwa
re
p
r
o
p
a
g
a
ti
o
n
o
n
ti
m
e
v
a
ry
i
n
g
n
e
two
rk
s:
c
o
m
p
a
ra
ti
v
e
stu
d
y
,
”
I
n
ter
n
a
ti
o
n
a
l
J
o
u
rn
a
l
o
f
M
o
d
e
rn
Ed
u
c
a
ti
o
n
a
n
d
Co
mp
u
ter
S
c
ien
c
e
,
v
o
l
.
6
,
n
o
.
8
,
p
p
.
2
5
-
3
3
,
2
0
1
4
.
[1
6
]
Ale
x
a
n
d
e
r
S
.
,
“
An
a
n
o
m
a
ly
in
t
ru
sio
n
d
e
tec
ti
o
n
sy
ste
m
b
a
se
d
o
n
in
tel
li
g
e
n
t
u
se
r
re
c
o
g
n
it
i
o
n
,
”
P
h
.
D
T
h
e
sis,
Un
iv
e
rsity
o
f
J
y
v
ä
sk
y
lä,
F
a
c
u
l
ty
o
f
In
f
o
rm
a
ti
o
n
Tec
h
n
o
l
o
g
y
,
F
i
n
la
n
d
,
2
0
1
2
.
[1
7
]
Ah
m
e
d
,
E.
,
Clark
,
A.,
M
o
h
a
y
,
G
.
,
“
A
n
o
v
e
l
sli
d
i
n
g
wi
n
d
o
w
b
a
se
d
c
h
a
n
g
e
d
e
tec
ti
o
n
a
lg
o
rit
h
m
fo
r
a
sy
m
m
e
tri
c
traffic,”
2
0
0
8
IFI
P
In
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
Ne
two
rk
a
n
d
Pa
r
a
ll
e
l
Co
mp
u
ti
n
g
,
S
h
a
n
g
h
a
i
,
2
0
0
8
,
p
p
.
1
6
8
-
175
.
[1
8
]
C
o
o
k
,
D.,
e
t
a
l
.,
“
Ca
tch
in
g
S
p
a
m
b
e
fo
re
i
t
a
rriv
e
s:
Do
m
a
in
S
p
e
c
if
ic
Dy
n
a
m
ic
Blac
k
li
sts,”
T
h
e
p
ro
c
e
e
d
in
g
s
o
f
t
h
e
Fo
u
rt
h
Au
stra
la
si
a
n
S
y
mp
o
siu
m
o
n
Gr
id
Co
m
p
u
t
in
g
a
n
d
e
-
Res
e
a
rc
h
(Au
sG
rid
2
0
0
6
)
a
n
d
t
h
e
Fo
u
rt
h
Au
stra
l
a
sia
n
In
fo
rm
a
t
io
n
S
e
c
u
rity W
o
rk
sh
o
p
(
Ne
two
rk
S
e
c
u
rity) (A
IS
W
2
0
0
6
)
,
Ho
b
a
rt,
Tas
m
a
n
ia,
Au
stra
l
ia
,
2
0
0
6
.
[1
9
]
To
d
d
B.
,
“
Distrib
u
ted
De
n
i
a
l
o
f
S
e
r
v
ice
Attac
k
s,”
L
i
n
u
x
S
e
c
u
rity.c
o
m
,
2
0
1
2
.
[On
li
n
e
].
Av
a
il
a
b
le:
h
tt
p
:
//
ww
w.l
i
n
u
x
se
c
u
rit
y
.
c
o
m
/res
o
u
rc
e
fil
e
s/in
tru
si
o
n
d
e
tec
ti
o
n
/d
d
o
s
-
wh
it
e
p
a
p
e
r.
h
tml
[2
0
]
Oju
g
o
,
A.
A.,
Eb
o
k
a
,
A.
O.
,
“
Emp
iri
c
a
l
e
v
a
lu
a
ti
o
n
o
n
c
o
m
p
a
ra
ti
v
e
stu
d
y
o
f
m
a
c
h
in
e
lea
rn
i
n
g
tec
h
n
i
q
u
e
s
i
n
d
e
t
e
c
t
i
o
n
o
f
D
D
o
S
,
”
J
o
u
r
n
a
l
o
f
A
p
p
l
i
e
d
S
c
i
e
n
c
e
E
n
g
i
n
e
e
r
i
n
g
T
e
c
h
n
o
l
o
g
y
a
n
d
E
d
u
c
a
t
i
o
n
,
v
o
l
.
2
,
n
o
.
1
,
p
p
.
1
8
-
2
7
,
2
0
2
0
.
[2
1
]
Oju
g
o
,
A.
A.,
E
b
o
k
a
,
A.
O.,
“
M
o
d
e
li
n
g
s
o
lu
ti
o
n
o
f
m
a
rk
e
t
b
a
sk
e
t
a
ss
o
c
iativ
e
ru
le
m
in
i
n
g
a
p
p
r
o
a
c
h
e
s
u
sin
g
d
e
e
p
n
e
u
ra
l
n
e
tw
o
rk
,
”
Di
g
it
a
l
T
e
c
h
n
o
l
o
g
ies
,
v
o
l
.
3
,
n
o
.
1
,
p
p
.
1
-
8
,
2
0
1
8
.
[2
2
]
I.
P
.
Ok
o
b
a
h
a
n
d
A.
A.
Oj
u
g
o
,
“
E
v
o
l
u
ti
o
n
a
ry
m
e
m
e
ti
c
m
o
d
e
ls
fo
r
m
a
lwa
re
in
tru
sio
n
d
e
tec
ti
o
n
:
a
c
o
m
p
a
ra
ti
v
e
q
u
e
st
fo
r
c
o
m
p
u
tati
o
n
a
l
so
l
u
ti
o
n
a
n
d
c
o
n
v
e
r
g
e
n
c
e
,
”
I
n
ter
n
a
t
io
n
a
l
J
o
u
rn
a
l
o
f
C
o
mp
u
ter
A
p
p
li
c
a
ti
o
n
s
(I
J
CA)
,
v
o
l.
1
7
9
,
n
o
.
3
9
,
p
p
.
3
4
-
4
3
,
2
0
1
8
.
[2
3
]
M
irk
o
v
ic,
P
.
R.
,
“
A
Tax
o
n
o
m
y
o
f
D
Do
S
Attac
k
a
n
d
DD
o
S
De
fe
n
se
M
e
c
h
a
n
ism
s,”
ACM
S
IGC
OM
M
C
o
mp
u
ter
Co
mm
u
n
ica
ti
o
n
Rev
iew
,
v
o
l.
3
4
,
n
o
.
2
,
p
p
.
3
9
-
5
3
,
2
0
0
4
.
[2
4
]
Ke
rk
,
J.,
“
Esto
n
ia
re
c
o
v
e
rs
fr
o
m
m
a
ss
iv
e
DD
o
S
a
tt
a
c
k
,
”
c
o
m
p
u
terw
o
r
ld
.
c
o
m
,
2
0
1
7
.
[O
n
li
n
e
].
Av
a
il
a
b
le:
h
tt
p
:
//
ww
w.co
m
p
u
terw
o
rld
.
c
o
m
/s/article
/9
0
1
9
7
2
5
/E
sto
n
ia_
re
c
o
v
e
rs_
fro
m
_
m
a
ss
iv
e
_
DD
o
S
_
a
tt
a
c
k
.
[2
5
]
Ha
m
d
a
n
O.
A.,
Ra
fi
d
a
h
M
.
N.,
Zaid
a
n
B.
B.
,
Zaid
a
n
A.
A.
,
“
I
n
t
ru
sio
n
De
tec
ti
o
n
S
y
ste
m
:
Ov
e
r
v
i
e
w,”
J
o
u
rn
a
l
o
f
Co
mp
u
t
in
g
,
v
o
l
.
2
,
n
o
.
2
,
2
0
1
0
.
[
2
6
]
M
o
o
r
e
,
G
.
M
.
,
e
t
a
l
.,
“
I
n
f
e
r
r
i
n
g
I
n
t
e
r
n
e
t
D
e
n
i
a
l
-
of
-
S
e
r
v
i
c
e
A
c
t
i
v
i
t
y
,
”
A
C
M
T
r
a
n
s
a
c
t
i
o
n
s
o
n
C
o
m
p
u
t
e
r
S
y
s
t
e
m
s
,
2
0
0
6
.
[2
7
]
Ak
e
ll
a
,
A.,
Bh
a
ra
m
b
e
,
A.,
Re
it
e
r,
M
.
,
a
n
d
S
e
sh
a
n
,
S
.
,
“
De
tec
ti
n
g
DD
o
S
a
tt
a
c
k
s
o
n
IS
P
n
e
two
rk
s
,
”
Pro
c
e
e
d
in
g
s
o
f
th
e
W
o
rk
sh
o
p
o
n
M
a
n
a
g
e
me
n
t
a
n
d
Pro
c
e
ss
in
g
o
f
Da
t
a
S
tre
a
ms
,
2
0
13
,
p
p
.
1
-
2.
[2
8
]
Ap
o
o
r
v
,
K.,
“
Ho
w
to
d
e
a
l
with
IP
a
d
d
re
ss
e
s
in
M
a
c
h
in
e
Le
a
rn
in
g
a
l
g
o
rit
h
m
s,”
2
0
1
6
.
[O
n
li
n
e
].
Av
a
il
a
b
le:
ww
w.q
u
o
ra
.
c
o
m
/
h
o
w
-
can
-
IP
-
a
d
d
re
ss
e
s
-
in
-
m
a
c
h
in
e
-
lea
rn
in
g
-
a
lg
o
ri
th
m
s
-
in
-
traffic
-
a
n
a
ly
sis
-
a
n
d
-
a
n
o
m
a
ly
-
d
e
tec
ti
o
n
.
[2
9
]
Ed
d
y
,
W
.
,
“
TCP
S
YN
flo
o
d
i
n
g
At
tac
k
s
a
n
d
C
o
m
m
o
n
M
it
i
g
a
ti
o
n
s,”
2
0
1
7
.
[On
li
n
e
].
Av
a
il
a
b
le
:
h
tt
p
:
//
to
o
ls.i
e
tf.
o
rg
/h
tml/rfc
4
9
8
7
.
Evaluation Warning : The document was created with Spire.PDF for Python.