I
AE
S
I
nte
rna
t
io
na
l J
o
urna
l o
f
Art
if
icia
l In
t
ellig
ence
(
I
J
-
AI
)
Vo
l.
1
3
,
No
.
4
,
Dec
em
b
er
2
0
2
4
,
p
p
.
4
8
9
5
~
4
9
0
5
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
13
.i
4
.
p
p
4
8
9
5
-
4
9
0
5
4895
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
A nov
el deep
ano
ma
ly
det
ection
appro
a
ch f
o
r in
trusi
o
n
detec
tion in
fut
u
r
istic ne
twork
Sa
i
K
rish
na
L
a
k
s
hm
ina
ra
y
a
na
,
P
ra
bh
ug
o
ud
I
.
B
a
s
a
rk
o
d
S
c
h
o
o
l
o
f
E
l
e
c
t
r
o
n
i
c
s
a
n
d
C
o
mm
u
n
i
c
a
t
i
o
n
E
n
g
i
n
e
e
r
i
n
g
,
R
EV
A
U
n
i
v
e
r
si
t
y
,
B
a
n
g
a
l
o
r
e
,
I
n
d
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
No
v
3
,
2
0
2
3
R
ev
is
ed
Ma
r
2
2
,
2
0
2
4
Acc
ep
ted
Ap
r
1
7
,
2
0
2
4
In
a
n
e
ra
wh
e
re
n
e
two
rk
s
a
re
in
c
re
a
sin
g
ly
h
e
tero
g
e
n
e
o
u
s
a
n
d
m
u
lt
i
-
d
o
m
a
in
,
e
sta
b
li
sh
in
g
r
o
b
u
st
se
c
u
rit
y
m
o
d
e
ls
t
o
p
ro
tec
t
d
a
ta
a
n
d
n
e
two
r
k
in
fra
stru
c
tu
re
is
b
e
c
o
m
in
g
e
v
e
r
m
o
re
c
o
m
p
lex
.
Trad
it
i
o
n
a
l
in
tru
si
o
n
d
e
tec
ti
o
n
sy
ste
m
s
(IDS)
o
ften
str
u
g
g
le
wit
h
n
o
v
e
l
o
r
v
a
rian
t
a
tt
a
c
k
s
t
h
a
t
fa
ll
o
u
tsi
d
e
p
re
d
e
fin
e
d
ru
le
se
ts,
re
su
lt
in
g
in
sig
n
ifi
c
a
n
t
d
e
tec
ti
o
n
c
h
a
ll
e
n
g
e
s.
Th
is
p
a
p
e
r
p
r
o
p
o
se
s
a
m
e
th
o
d
o
l
o
g
ica
ll
y
re
fin
e
d
a
p
p
r
o
a
c
h
lev
e
ra
g
in
g
d
a
ta
-
d
riv
e
n
in
si
g
h
ts
a
n
d
sta
ti
stica
ll
y
ro
b
u
st
fe
a
tu
re
se
lec
ti
o
n
t
o
e
n
h
a
n
c
e
th
e
trai
n
in
g
d
a
tas
e
t.
T
h
e
stu
d
y
p
re
se
n
ts
a
l
o
n
g
sh
o
r
t
-
term
m
e
m
o
ry
-
a
u
to
e
n
c
o
d
e
r
(LS
TM
-
AE)
b
a
se
d
lea
rn
i
n
g
m
o
d
e
l
d
e
sig
n
e
d
f
o
r
m
u
lt
i
-
c
las
s
a
n
o
m
a
ly
d
e
tec
ti
o
n
.
T
h
e
m
o
d
e
l'
s
n
o
v
e
lt
y
li
e
s
in
it
s
a
p
p
li
c
a
ti
o
n
o
f
d
istan
c
e
m
e
tri
c
s
to
d
e
fin
e
d
isti
n
c
t
th
re
sh
o
l
d
s
fo
r
v
a
ried
a
tt
a
c
k
c
las
sifica
ti
o
n
s,
a
stra
teg
y
th
a
t
sig
n
ifi
c
a
n
t
ly
a
m
p
l
ifi
e
s
d
e
tec
ti
o
n
p
re
c
isio
n
.
Ex
p
e
rime
n
tal
re
su
lt
s
v
a
li
d
a
te
th
e
su
p
e
rio
r
p
e
rfo
rm
a
n
c
e
o
f
t
h
e
p
r
o
p
o
se
d
sy
ste
m
,
a
c
h
ie
v
in
g
9
4
.
8
2
%
a
c
c
u
ra
c
y
ra
te,
o
u
t
p
e
rfo
rm
in
g
sim
il
a
r
e
x
i
stin
g
w
o
rk
s.
T
h
e
stu
d
y
a
lso
p
ro
a
c
ti
v
e
ly
a
d
d
re
ss
e
s
c
o
m
m
o
n
issu
e
s
o
f
c
las
s
imb
a
lan
c
e
a
n
d
sk
e
we
d
d
a
ta
re
p
r
e
se
n
tatio
n
in
b
e
n
c
h
m
a
rk
d
a
tas
e
ts
b
y
stra
teg
i
c
a
ll
y
train
i
n
g
t
h
e
m
o
d
e
l
o
n
n
o
rm
a
l
traffic,
e
n
h
a
n
c
in
g
it
s c
a
p
a
b
il
it
y
t
o
g
e
n
e
ra
li
z
e
a
n
d
i
d
e
n
ti
f
y
a
n
o
m
a
li
e
s e
ffe
c
ti
v
e
ly
.
K
ey
w
o
r
d
s
:
An
o
m
aly
d
etec
tio
n
Fu
tu
r
is
tic
n
etwo
r
k
I
n
ter
n
et
o
f
th
in
g
s
L
o
n
g
s
h
o
r
t
-
ter
m
m
e
m
o
r
y
-
a
u
to
en
co
d
er
Secu
r
ity
attac
k
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
:
Sai
Kr
is
h
n
a
L
ak
s
h
m
in
ar
ay
a
n
a
Sch
o
o
l
of
E
lectr
o
n
ics
an
d
C
o
m
m
u
n
icatio
n
E
n
g
in
ee
r
in
g
,
R
E
VA
Un
iv
er
s
ity
B
en
g
alu
r
u
,
I
n
d
ia
E
m
ail:
s
k
ln
ar
ay
an
a@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
ev
o
lu
tio
n
of
co
m
m
u
n
icati
o
n
tech
n
o
l
o
g
y
f
r
o
m
3G
to
4G
h
as
alr
ea
d
y
r
ev
o
lu
tio
n
ized
h
o
w
h
u
m
an
s
in
ter
ac
t
with
d
ig
ital
tech
n
o
lo
g
ies
[
1
]
.
T
h
is
tr
an
s
itio
n
h
as
also
f
ac
ilit
ated
u
n
p
r
ec
ed
e
n
ted
in
n
o
v
atio
n
s
in
th
e
ac
tu
al
r
ea
lizatio
n
of
p
r
ac
tical
ap
p
licatio
n
s
s
u
ch
as
s
m
ar
t
h
o
m
es,
s
m
ar
t
cities,
s
m
ar
t
h
ea
lth
ca
r
e,
a
u
to
n
o
m
o
u
s
v
eh
icles,
an
d
in
d
u
s
tr
ial
au
to
m
atio
n
.
W
ith
th
e
ad
v
en
t
of
f
if
th
-
g
en
er
atio
n
5G
tech
n
o
lo
g
y
,
we
ar
e
on
th
e
cu
s
p
of
an
o
th
er
u
ltra
-
m
o
d
er
n
tr
a
n
s
f
o
r
m
atio
n
th
at
will
d
r
am
atica
lly
ch
an
g
e
how
h
u
m
an
s
liv
e
a
n
d
wo
r
k
[
2
]
.
T
h
e
ex
p
ec
tatio
n
f
r
o
m
th
e
5G
is
not
o
n
ly
lim
ited
to
o
f
f
e
r
in
g
f
a
s
ter
s
p
ee
d
an
d
s
ea
m
less
co
n
n
ec
tiv
ity
.
B
u
t
it
will
u
ltima
tely
p
r
o
life
r
ate
th
e
cu
r
r
en
t
ec
o
s
y
s
tem
of
co
n
n
ec
te
d
d
ev
ices
to
th
e
n
ex
t
lev
el,
in
cr
ea
s
in
g
th
e
m
atu
r
ity
of
cu
r
r
en
t
tec
h
n
o
lo
g
ies.
T
h
e
i
n
te
g
r
atio
n
of
5G
tech
n
o
lo
g
y
with
th
e
in
te
r
n
et
of
th
in
g
s
(
I
o
T
)
,
clo
u
d
co
m
p
u
tin
g
,
co
g
n
itiv
e
r
ad
io
n
etwo
r
k
s
(
C
R
N)
,
an
d
s
o
f
twar
e
-
d
ef
in
ed
n
et
wo
r
k
in
g
(
SDN)
s
y
s
tem
s
,
to
g
eth
er
will
f
o
r
m
t
h
e
f
o
u
n
d
atio
n
of
t
h
e
f
u
tu
r
is
tic
n
etwo
r
k
ec
o
s
y
s
tem
[
3
]
,
[
4
]
.
T
h
e
f
u
tu
r
is
tic
n
etwo
r
k
ca
n
be
ch
ar
ac
ter
ized
by
ex
em
p
lify
in
g
th
e
in
te
r
n
et
of
e
v
er
y
th
in
g
(
I
o
E
)
c
o
n
ce
p
t,
w
h
ich
will
h
av
e
r
eliab
le
c
o
n
n
ec
tiv
i
ty
,
ex
p
e
d
itio
u
s
d
ata
tr
an
s
m
is
s
io
n
,
an
d
u
tm
o
s
t
ad
ap
tab
ilit
y
.
In
th
is
co
n
tex
t,
not
o
n
ly
ev
er
y
d
ev
ice
but
m
ay
b
e
ev
er
y
o
b
ject
will
be
v
ir
tu
ally
co
n
n
ec
ted
,
in
clu
d
in
g
h
o
m
e
,
o
f
f
ice,
an
d
ev
e
n
o
u
r
b
o
d
y
will
g
e
n
er
ate
a
n
d
s
h
ar
e
d
ata,
th
er
eb
y
tr
an
s
f
o
r
m
in
g
h
u
m
an
life
s
ty
le
s
with
in
f
in
ite
p
o
s
s
ib
ilit
ies
[
5
]
,
[
6
]
.
Ho
wev
er
,
th
is
tr
an
s
itio
n
to
th
is
f
u
tu
r
is
tic
n
etwo
r
k
ec
o
s
y
s
tem
will
also
r
aise
s
ig
n
if
ican
t
co
n
ce
r
n
s
ab
o
u
t
s
ec
u
r
ity
an
d
p
r
iv
ac
y
is
s
u
es.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
13
,
No
.
4
,
Dec
em
b
er
20
24
:
4
8
9
5
-
4
9
0
5
4896
As
tech
n
o
lo
g
y
d
ev
elo
p
s
an
d
a
d
v
an
ce
s
,
it
also
o
p
en
s
n
ew
wa
y
s
f
o
r
attac
k
er
s
to
ex
p
l
o
it
v
u
ln
er
ab
ilit
ies
r
elate
d
to
n
etwo
r
k
s
y
s
tem
s
.
T
h
e
co
m
p
lex
n
atu
r
e
of
f
u
tu
r
is
tic
n
etwo
r
k
s
o
p
er
atin
g
on
th
e
i
n
ter
n
et
an
d
wir
eless
co
m
m
u
n
icatio
n
ch
a
n
n
els
will
ca
r
r
y
in
h
er
en
t
v
u
ln
e
r
ab
ilit
ies,
wh
ich
m
ak
es
it
o
p
en
to
i
n
v
ite
p
o
ten
tial
s
ec
u
r
ity
th
r
ea
ts
[
7
]
,
[
8
]
.
If
s
ec
u
r
ity
r
e
q
u
ir
em
en
ts
ar
e
n
o
t
en
s
u
r
e
d
c
o
m
p
r
eh
e
n
s
iv
ely
,
it
will
h
av
e
a
s
ig
n
if
ican
t
im
p
ac
t
th
at
can
tu
r
n
it
f
r
o
m
a
p
r
o
m
is
in
g
b
en
e
f
it
to
a
p
o
ten
tial
cu
r
s
e.
As
I
o
E
ev
o
lv
es,
we
will
f
ac
e
m
an
y
n
ew
s
ec
u
r
ity
ch
allen
g
es
th
at
will
s
ig
n
if
ican
tly
ex
ce
ed
t
h
e
ca
p
ab
ilit
ies
of
tr
ad
itio
n
al
s
ec
u
r
ity
s
y
s
tem
s
.
T
h
e
wo
r
ld
h
as
r
ec
en
tly
s
ee
n
ca
s
es
of
m
o
d
e
r
n
cy
b
er
-
attac
k
s
,
wh
ich
h
a
v
e
ca
u
s
ed
ec
o
n
o
m
ic
l
o
s
s
an
d
d
am
a
g
e
to
r
ep
u
tatio
n
a
n
d
liv
es
[
9
]
.
T
h
er
ef
o
r
e,
it
is
im
p
er
ativ
e
th
at
we
lea
r
n
f
r
o
m
p
a
s
t
cy
b
er
-
attac
k
i
n
cid
en
ts
an
d
put
ef
f
o
r
ts
to
war
d
s
d
ev
elo
p
in
g
s
o
p
h
is
ticated
s
ec
u
r
ity
s
o
lu
tio
n
s
th
at
m
ee
t
th
e
s
e
cu
r
ity
r
eq
u
ir
e
m
en
ts
of
a
f
u
tu
r
i
s
tic
n
etwo
r
k
with
a
m
ax
im
u
m
lay
e
r
of
p
r
o
tectio
n
.
Mu
ch
r
esear
ch
h
as
b
ee
n
d
o
n
e
in
liter
atu
r
e
to
ex
p
lo
r
e
th
e
p
o
s
s
ib
le
s
ec
u
r
ity
v
u
ln
er
ab
ilit
ies
ass
o
ciate
d
wi
th
th
e
I
o
T
ec
o
s
y
s
tem
,
an
d
d
if
f
er
en
t
co
u
n
ter
m
ea
s
u
r
e
s
ch
em
es
h
av
e
b
ee
n
in
tr
o
d
u
ce
d
to
p
r
o
tect
n
etwo
r
k
s
.
T
h
e
s
ec
u
r
ity
s
o
lu
tio
n
s
b
ased
on
th
e
cr
y
p
to
g
r
ap
h
ic
p
r
i
m
itiv
e
h
av
e
s
h
o
wn
p
r
o
m
is
in
g
o
u
tco
m
es
in
en
s
u
r
in
g
d
ata
c
o
n
f
id
e
n
tially
,
in
teg
r
ity
,
an
d
u
s
er
p
r
iv
ac
y
[
1
0
]
.
B
u
t
with
s
ig
n
if
ican
t
tech
n
o
lo
g
ical
im
p
r
o
v
em
en
ts
,
cy
b
er
attac
k
s
h
a
v
e
b
ec
o
m
e
c
o
m
p
lex
an
d
in
c
r
ea
s
in
g
ly
i
n
tellig
en
t.
T
h
e
ex
is
tin
g
cr
y
p
to
g
r
ap
h
ic
s
ch
em
es
p
r
im
a
r
ily
f
o
cu
s
on
s
ec
u
r
ity
d
ata
r
ath
er
th
an
m
o
n
ito
r
in
g
an
d
p
r
o
t
ec
tin
g
th
e
n
etwo
r
k
,
wh
ich
m
ay
not
be
s
u
f
f
icien
t
to
co
u
n
ter
s
o
p
h
is
ticated
an
d
ev
o
lv
in
g
th
r
ea
ts
ef
f
ec
tiv
el
y
[
1
1
]
.
T
h
er
ef
o
r
e,
to
p
r
o
tect
th
e
e
n
tire
n
etwo
r
k
in
f
r
astru
ctu
r
e
a
h
ig
h
ly
r
esp
o
n
s
iv
e,
d
y
n
am
ic,
a
n
d
in
tellig
en
t
d
ef
en
ce
s
y
s
tem
is
r
eq
u
ir
ed
to
be
d
esig
n
an
d
d
e
v
elo
p
.
In
t
h
is
r
eg
ar
d
,
with
th
e
ad
v
en
t
of
m
ac
h
in
e
lear
n
in
g
tech
n
o
lo
g
ies,
th
e
d
ev
elo
p
m
e
n
t
of
n
etwo
r
k
in
tr
u
s
io
n
d
etec
tio
n
s
y
s
tem
(
I
DS)
is
one
of
th
e
ac
tiv
e
ar
ea
s
in
t
h
e
cu
r
r
en
t
r
esear
ch
s
ce
n
ar
io
[
1
2
]
.
In
liter
atu
r
e,
t
h
e
I
DS
b
ased
s
ec
u
r
ity
s
o
lu
tio
n
s
ar
e
eith
er
b
ased
on
th
e
s
ig
n
atu
r
e
-
b
ased
or
an
o
m
aly
-
b
ased
ap
p
r
o
ac
h
es.
Ho
wev
er
,
u
n
lik
e
s
ig
n
atu
r
e
-
b
ased
I
DS,
wh
ich
r
ely
on
p
r
ed
ef
in
ed
p
atter
n
s
to
id
en
tify
o
n
ly
k
n
o
wn
s
attac
k
s
,
an
o
m
aly
-
b
ased
I
DS
ar
e
in
cr
ea
s
in
g
ly
g
ain
in
g
tr
ac
tio
n
d
u
e
to
its
ab
ilit
y
to
m
o
n
ito
r
n
etwo
r
k
a
n
d
f
la
g
an
y
b
eh
av
i
o
r
th
at
d
ev
iates
f
r
o
m
th
e
estab
lis
h
ed
n
o
r
m
[
1
3
]
.
T
h
is
I
DS
s
y
s
tem
is
q
u
ick
in
m
a
k
in
g
r
esp
o
n
s
e
th
a
n
s
ig
n
atu
r
e
-
b
ased
I
DS
an
d
o
wn
s
ab
ilit
y
to
d
etec
t
ze
r
o
-
d
a
y
attac
k
s
,
a
cr
itically
im
p
o
r
tan
t
f
ea
t
u
r
e
in
t
h
e
f
ac
e
of
ev
o
lv
in
g
s
o
p
h
is
ticated
th
r
ea
t
s
in
th
e
co
n
tex
t
of
f
u
tu
r
is
tic
I
o
T
n
etwo
r
k
.
Alza
h
r
an
i
an
d
Alen
az
i
[
1
4
]
u
s
ed
X
G
B
o
o
s
t
class
if
ier
s
to
cr
ea
te
in
tr
u
s
io
n
d
etec
tio
n
m
o
d
els
f
o
r
h
ig
h
-
o
r
d
er
n
etwo
r
k
en
v
ir
o
n
m
en
ts
th
at
in
teg
r
ate
SDN
an
d
wir
eless
s
en
s
o
r
n
etwo
r
k
s
(
W
SN)
.
T
h
e
au
th
o
r
s
ex
tr
ac
ted
ap
p
r
o
x
im
ately
40
f
e
atu
r
es
f
r
o
m
th
e
NSL
-
KDD
d
ataset
an
d
tr
ain
ed
an
Xg
B
o
o
s
t
class
if
ier
to
d
etec
t
m
u
ltip
le
class
es
of
in
tr
u
s
io
n
s
.
E
x
p
er
im
en
tal
r
esu
lts
d
em
o
n
s
tr
ate
th
e
ef
f
ec
tiv
en
ess
of
th
e
p
r
o
p
o
s
ed
s
ch
em
e
r
elativ
e
to
v
a
r
io
u
s
s
im
ilar
cl
ass
ical
lear
n
in
g
m
o
d
els
s
u
ch
as
d
ec
is
io
n
tr
ee
(
DT
)
,
r
an
d
o
m
f
o
r
est
(
R
F),
a
n
d
ar
tific
ial
n
eu
r
al
n
etwo
r
k
(
ANN)
m
o
d
els.
An
o
th
er
r
esear
ch
ef
f
o
r
t
u
s
in
g
tr
ee
-
b
ased
class
if
ier
s
can
be
s
ee
n
in
th
e
s
tu
d
y
by
Sar
k
er
et
a
l.
[
1
5
]
,
wh
ich
ad
d
r
ess
es
th
e
ch
allen
g
e
of
h
ig
h
c
o
m
p
u
tatio
n
al
co
s
t
in
d
ev
elo
p
in
g
I
DS.
T
h
ey
p
r
o
p
o
s
ed
a
d
etec
tio
n
m
e
th
o
d
u
s
in
g
a
DT
m
o
d
el
th
at
not
o
n
ly
p
er
f
o
r
m
s
class
if
icatio
n
but
also
p
r
io
r
itizes
f
ea
tu
r
es
b
ased
on
th
eir
im
p
o
r
tan
ce
,
th
e
r
eb
y
ef
f
ec
tiv
el
y
r
e
d
u
cin
g
p
r
o
f
ile
d
im
en
s
io
n
ality
.
T
h
is
s
tr
ateg
y
ca
n
s
h
o
r
ten
tr
ain
in
g
tim
e
an
d
p
o
te
n
tially
in
cr
ea
s
e
f
u
n
ctio
n
al
g
en
er
ality
to
d
ev
el
o
p
co
s
t
-
ef
f
ec
tiv
e
an
d
r
o
b
u
s
t
I
DS.
R
esear
ch
ef
f
o
r
ts
to
r
ed
u
ce
f
al
s
e
p
o
s
itiv
es
d
u
r
in
g
attac
k
d
et
ec
tio
n
ar
e
s
h
o
wn
[
1
6
]
,
wh
e
r
e
th
ey
s
u
g
g
ested
an
in
tellig
en
t
f
ilter
in
g
m
eth
o
d
th
at
u
s
es
ed
g
e
co
m
p
u
tin
g
d
ev
ic
es
with
th
r
ea
t
d
etec
tio
n
m
o
d
u
les
to
jo
in
tly
s
o
lv
e
th
e
p
r
o
ce
s
s
in
g
b
u
r
d
e
n
p
r
o
b
le
m
an
d
o
b
tain
b
etter
d
etec
tio
n
ac
cu
r
ac
y
.
Alh
ajjar
et
a
l.
[
1
7
]
h
ig
h
lig
h
ted
an
e
x
citin
g
ar
ea
of
r
esear
ch
ex
p
lo
r
in
g
s
u
s
ce
p
tib
ilit
ies
in
d
if
f
e
r
en
t
s
u
p
er
v
is
ed
lear
n
i
n
g
s
ch
em
es
an
d
th
eir
wea
k
n
ess
to
m
a
n
ip
u
latio
n
by
attac
k
er
s
s
ee
k
in
g
to
ev
ad
e
in
tr
u
s
io
n
d
etec
tio
n
.
T
h
ey
u
s
ed
d
if
f
er
e
n
t
n
atu
r
e
-
i
n
s
p
ir
ed
o
p
tim
izatio
n
alg
o
r
ith
m
s
an
d
g
e
n
er
ativ
e
a
d
v
er
s
ar
ial
n
etwo
r
k
s
(
GAN)
to
g
e
n
er
ate
ad
v
er
s
ar
ial
s
am
p
les.
T
h
e
r
esu
lts
f
o
u
n
d
th
at
m
an
y
lear
n
in
g
m
o
d
els
ex
p
er
ien
ce
d
a
s
ig
n
if
ican
t
in
cr
ea
s
e
in
f
alse
p
o
s
itiv
e
r
ates
wh
en
ex
p
o
s
ed
to
th
ese
ad
v
er
s
ar
ial
p
atter
n
s
.
A
r
esear
ch
ef
f
o
r
t
to
en
s
u
r
e
th
e
f
air
n
ess
of
I
DS
on
b
iased
d
atasets
was
co
n
d
u
cted
by
[
1
8
]
,
w
h
er
e
th
ey
a
d
o
p
ted
th
e
s
y
n
t
h
etic
m
in
o
r
ity
o
v
er
s
am
p
lin
g
tech
n
iq
u
e
to
s
o
lv
e
th
e
class
im
b
alan
ce
p
r
o
b
lem
in
n
etwo
r
k
in
tr
u
s
io
n
d
atasets
.
T
h
e
p
r
esen
ted
s
ch
em
e
f
u
r
th
er
u
t
ilized
th
e
Gin
i
i
n
d
ex
to
s
e
lect
r
elev
an
t
f
ea
tu
r
es
f
o
r
t
r
ain
in
g
th
e
m
o
d
el.
E
x
p
er
im
en
tal
r
esu
lts
on
th
e
U
NSW
-
N
B
1
5
d
ataset
im
p
r
o
v
e
I
DS
an
d
en
s
u
r
e
its
s
u
itab
ilit
y
f
o
r
class
im
b
alan
ce
p
r
o
b
lem
s
.
Me
b
aw
o
n
d
u
et
a
l.
[
1
9
]
aim
e
d
to
im
p
r
o
v
e
th
e
ef
f
icien
cy
of
n
etwo
r
k
I
DS
by
p
r
io
r
itizin
g
ac
cu
r
ac
y
an
d
f
ast
r
esp
o
n
s
e
tim
e.
T
h
e
r
esear
ch
er
s
in
tr
o
d
u
ce
d
a
n
eu
r
al
n
etwo
r
k
m
o
d
el
t
h
at
in
teg
r
ates
an
u
n
ce
r
tain
ty
-
awa
r
e
m
ec
h
a
n
is
m
to
id
en
tify
th
e
m
o
s
t
r
elev
an
t
f
ea
tu
r
es
in
th
e
lear
n
in
g
p
r
o
ce
s
s
.
T
h
e
m
o
d
el'
s
p
er
f
o
r
m
an
ce
was
v
alid
ated
u
s
in
g
th
e
UNSW
-
NB
1
5
d
ataset.
T
h
e
r
esu
lts
d
em
o
n
s
tr
ate
th
e
e
f
f
ec
tiv
en
ess
of
th
e
p
r
o
p
o
s
ed
I
DS
m
o
d
el
in
ter
m
s
of
h
ig
h
e
r
d
etec
tio
n
r
ate
an
d
co
m
p
u
tatio
n
al
ef
f
icien
c
y
.
Ho
wev
er
,
th
e
in
h
e
r
en
t
ch
allen
g
es
ass
o
ciate
d
with
th
e
m
ajo
r
ity
of
s
u
p
er
v
is
ed
I
DSs
ar
e
p
r
o
g
r
ess
iv
ely
g
iv
in
g
r
is
e
to
in
cr
ea
s
in
g
in
ter
est
in
an
o
m
aly
-
b
ased
ap
p
r
o
ac
h
es.
T
h
ese
ch
allen
g
es
in
clu
d
e
th
e
n
ee
d
f
o
r
h
u
g
e
,
lab
elled
tr
ain
in
g
d
ata,
th
e
n
ee
d
f
o
r
co
m
p
lex
f
ea
tu
r
e
en
g
i
n
ee
r
in
g
p
r
e
-
p
r
o
ce
s
s
in
g
s
tep
s
,
an
d
th
e
d
ep
en
d
e
n
ce
on
la
b
el
ac
cu
r
ac
y
f
o
r
ef
f
ec
tiv
e
m
o
d
el
tr
ain
in
g
.
On
e
of
th
e
m
o
s
t
p
r
o
m
is
in
g
ap
p
r
o
ac
h
es
f
o
r
an
o
m
aly
-
b
ased
I
DS
is
au
to
en
co
d
er
s
n
eu
r
al
n
etwo
r
k
m
o
d
el
wh
ich
is
an
u
n
s
u
p
er
v
is
ed
lear
n
in
g
tech
n
iq
u
e.
Au
to
e
n
co
d
er
s
lear
n
co
m
p
r
ess
ed
r
ep
r
esen
tatio
n
s
f
r
o
m
a
s
in
g
le
in
p
u
t,
ty
p
ically
r
ef
lectin
g
n
o
r
m
al
s
y
s
tem
b
eh
av
io
r
[
2
0
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
A
n
o
ve
l d
ee
p
a
n
o
ma
ly
d
etec
ti
o
n
a
p
p
r
o
a
ch
fo
r
in
tr
u
s
io
n
d
ete
ctio
n
in
…
(
S
a
i Kr
is
h
n
a
La
ksh
min
a
r
a
ya
n
a
)
4897
An
y
s
ig
n
if
ican
t
d
e
v
iatio
n
s
f
r
o
m
p
r
ev
io
u
s
ly
id
en
tifie
d
an
d
lear
n
ed
p
atter
n
s
ar
e
th
en
f
l
ag
g
ed
as
p
o
ten
tial
an
o
m
alies.
In
r
ec
en
t
y
ea
r
s
,
m
an
y
r
esear
ch
e
r
s
h
a
v
e
s
u
cc
ess
f
u
lly
em
p
lo
y
ed
v
a
r
io
u
s
au
to
e
n
co
d
er
m
o
d
els
f
o
r
an
o
m
aly
-
b
ased
I
D
S
d
ev
elo
p
m
en
t
[
2
1
]
.
L
o
p
es
et
a
l
.
[
2
2
]
ex
p
lo
r
ed
th
e
ef
f
ec
tiv
en
ess
of
d
e
n
o
is
in
g
au
to
en
co
d
er
f
o
r
a
d
d
r
ess
in
g
t
h
e
lim
ited
lab
elled
d
ata
s
am
p
les
r
eq
u
ir
ed
to
tr
ain
d
ee
p
le
ar
n
in
g
f
o
r
b
u
ild
i
n
g
ef
f
ec
tiv
e
I
DS
f
o
r
I
o
T
.
Her
e,
th
e
d
en
o
is
in
g
au
to
en
c
o
d
er
was
tr
ain
ed
with
an
in
p
u
t
s
a
m
p
le,
an
d
o
b
tain
e
d
co
m
p
ac
t
f
ea
t
u
r
es
wer
e
t
h
en
u
s
ed
to
tr
ain
th
e
d
ee
p
lea
r
n
in
g
m
o
d
els
in
a
s
u
p
e
r
v
is
ed
m
a
n
n
er
.
T
h
e
s
im
u
latio
n
r
esu
lts
claim
th
e
u
s
ef
u
ln
ess
of
th
e
p
r
esen
ted
a
p
p
r
o
ac
h
with
d
ata
r
ed
u
ctio
n
to
o
n
e
-
ten
th
of
th
e
in
p
u
t
tr
ai
n
in
g
s
am
p
le
s
ize,
alo
n
g
with
9
9
%
d
etec
tio
n
ac
cu
r
ac
y
.
R
esear
ch
wo
r
k
d
o
n
e
to
war
d
s
id
en
tify
in
g
b
o
th
k
n
o
wn
an
d
u
n
k
n
o
wn
in
tr
u
s
io
n
s
u
s
in
g
ad
ap
tiv
e
v
ar
iatio
n
al
a
u
to
en
c
o
d
e
r
is
p
r
esen
ted
[
2
3
]
.
T
h
is
wo
r
k
u
s
es
an
ex
tr
em
e
v
alu
e
to
d
if
f
er
en
tiate
b
etwe
en
k
n
o
wn
a
n
d
u
n
k
n
o
wn
in
tr
u
s
io
n
s
u
s
in
g
r
ec
o
n
s
tr
u
ctio
n
e
r
r
o
r
s
o
b
tain
e
d
f
r
o
m
th
e
tr
ain
ed
m
o
d
el
.
T
h
e
ex
p
er
im
e
n
ts
wer
e
co
n
d
u
cted
on
NSL
-
KDD
an
d
C
I
C
I
DS2
0
1
7
d
ataset,
an
d
th
e
o
u
tco
m
e
d
em
o
n
s
tr
ated
t
h
at
th
e
p
r
ese
n
ted
m
o
d
el
ac
h
ie
v
ed
a
m
in
im
al
f
alse
p
o
s
itiv
e
r
ate
of
ap
p
r
o
x
im
ately
1
%.
Su
n
et
a
l
.
[
2
4
]
p
r
esen
te
d
a
s
p
ar
s
e
f
ea
tu
r
e
r
ep
r
esen
tatio
n
m
o
d
el
b
ased
on
ap
p
l
y
in
g
th
e
laten
t
s
p
ac
e
of
a
v
ar
iatio
n
al
a
u
to
en
co
d
er
to
le
ar
n
c
o
n
tex
tu
al
attr
ib
u
tes
in
co
m
p
ac
t
v
ec
to
r
r
ep
r
esen
tatio
n
.
T
h
e
NSL
-
KDD
d
ataset
was
u
s
ed
to
v
alid
ate
t
h
e
p
er
f
o
r
m
a
n
ce
of
th
e
p
r
esen
t
ed
s
ch
em
e.
T
h
e
s
tu
d
y
o
u
tco
m
e
ex
h
ib
ited
a
h
ig
h
er
d
etec
tio
n
r
ate
an
d
s
ig
n
if
ican
t
tr
ain
in
g
,
s
to
r
ag
e,
an
d
m
em
o
r
y
co
s
t
m
i
n
im
izatio
n
.
Z
a
v
r
ak
an
d
I
s
k
ef
iy
eli
[
2
5
]
in
v
esti
g
ated
th
e
u
s
e
of
au
to
e
n
co
d
er
s
f
o
r
an
o
m
al
y
d
etec
tio
n
in
n
etwo
r
k
tr
af
f
ic.
T
h
ey
em
p
lo
y
v
a
r
iatio
n
al
au
to
en
co
d
er
s
on
f
ea
t
u
r
es
ex
tr
a
cted
f
r
o
m
n
o
r
m
al
tr
a
f
f
ic
d
ata.
To
ass
ess
th
e
ef
f
ec
tiv
en
ess
of
an
o
m
aly
d
etec
tio
n
,
th
ey
co
m
p
ar
e
th
e
a
u
to
en
co
d
er
m
o
d
els
to
an
SVM
class
if
ier
.
T
h
e
m
o
d
els
wer
e
ev
alu
ated
on
s
ev
er
al
s
tan
d
a
r
d
d
atasets
,
an
d
th
e
r
esu
lts
d
em
o
n
s
tr
ate
th
at
th
e
v
ar
iatio
n
al
au
to
en
co
d
er
ac
h
iev
ed
lo
wer
f
alse
-
p
o
s
itiv
e
r
ates
co
m
p
ar
ed
to
th
e
SVM
an
d
s
tan
d
ar
d
au
to
e
n
co
d
e
r
s
.
Osad
a
et
a
l
.
[
2
6
]
also
s
ee
a
s
im
ilar
r
e
s
ea
r
ch
ef
f
o
r
t
in
th
e
s
am
e
r
esear
ch
d
ir
ec
tio
n
wh
er
e
v
ar
iatio
n
al
au
to
en
co
d
er
is
u
s
ed
alo
n
g
with
L
ap
lacia
n
r
eg
u
la
r
ized
least
s
q
u
ar
es
m
eth
o
d
to
ac
h
iev
e
p
r
ec
is
e
f
e
atu
r
es
an
d
r
ed
u
ce
tr
ain
i
n
g
c
o
s
ts
.
T
h
er
e
ar
e
o
th
er
s
im
ilar
a
p
p
r
o
ac
h
es,
s
u
ch
as
Dao
an
d
L
ee
[
2
7
]
,
th
e
au
th
o
r
s
h
av
e
p
r
esen
ted
a
lig
h
twei
g
h
t
I
DS
m
o
d
el
u
s
in
g
a
s
tack
ed
au
to
en
c
o
d
er
an
d
n
etwo
r
k
p
r
u
n
in
g
m
ec
h
an
is
m
.
An
au
to
en
co
d
er
lev
er
a
g
in
g
r
e
cu
r
r
en
t
n
eu
r
o
n
u
n
it
is
u
s
ed
[
2
8
]
,
[
2
9
]
to
d
ev
elo
p
r
ea
l
-
tim
e
I
DS
f
o
r
in
d
u
s
tr
ial
n
e
two
r
k
s
an
d
au
t
o
m
o
tiv
e
i
n
d
u
s
tr
ies,
r
esp
ec
tiv
ely
.
Ho
wev
er
,
th
er
e
ar
e
s
ev
e
r
al
a
p
p
r
o
ac
h
es
to
co
u
n
ter
cy
b
er
-
a
ttack
s
in
I
o
T
,
to
ac
h
iev
e
t
h
e
s
ec
u
r
ity
r
eq
u
ir
em
e
n
ts
f
o
r
n
etwo
r
k
s
in
th
e
f
u
tu
r
e
,
a
f
ew
s
ig
n
if
ica
n
t
co
n
ce
r
n
s
s
till
n
ee
d
to
be
f
ix
ed
.
C
o
n
tr
ar
y
to
s
ig
n
atu
r
e
-
b
ased
I
DS,
an
o
m
aly
-
b
ased
I
DS
ar
e
f
o
u
n
d
be
v
er
y
ef
f
ec
tiv
e
an
d
h
i
g
h
ly
r
esp
o
n
s
iv
e
ag
ain
s
t
m
alicio
u
s
f
lo
w
in
th
e
n
etwo
r
k
.
It
h
as
b
e
en
f
o
u
n
d
th
at
th
e
m
o
s
t
of
th
e
ex
is
tin
g
an
o
m
aly
d
etec
tio
n
s
c
h
em
es,
ar
e
not
m
u
ch
ap
p
licab
le
in
f
u
tu
r
is
tic
n
etwo
r
k
s
in
ce
en
o
r
m
o
u
s
am
o
u
n
t
of
d
ata
g
e
n
er
ate
due
th
ei
r
h
ig
h
-
d
im
en
s
io
n
al
an
d
h
eter
o
g
en
e
o
u
s
n
atu
r
e.
Hen
ce
,
th
is
is
wh
er
e
au
to
en
co
d
er
m
o
d
els
ar
e
ef
f
ec
tiv
e
at
ca
p
tu
r
in
g
co
m
p
lex
d
ata
d
is
tr
ib
u
tio
n
s
an
d
r
ed
u
cin
g
d
a
ta
d
im
en
s
io
n
s
b
o
th
ess
en
tial
f
o
r
a
n
o
m
aly
d
etec
tio
n
in
h
ig
h
-
d
im
en
s
io
n
al
d
ata
s
p
ac
es.
Desp
ite
m
an
y
r
ec
en
t
wo
r
k
s
,
y
et,
th
e
ap
p
licatio
n
of
au
to
en
co
d
e
r
s
in
I
DS
f
o
r
f
u
tu
r
is
tic
n
etwo
r
k
s
h
as
not
b
ee
n
m
atu
r
ed
a
n
d
is
a
r
elativ
ely
n
ew
ar
ea
of
r
esear
ch
.
T
h
e
d
i
v
er
s
e
r
an
g
e
of
d
e
v
ice
s
an
d
d
ata
in
th
ese
n
etwo
r
k
s
,
alo
n
g
s
id
e
th
e
s
o
p
h
is
ticatio
n
of
m
o
d
er
n
c
y
b
er
th
r
ea
ts
,
n
ec
ess
itates
a
more
r
ef
in
ed
an
d
n
u
an
ce
d
au
to
en
co
d
er
m
o
d
el.
Fu
r
th
er
m
o
r
e,
th
e
r
ea
l
-
tim
e
n
atu
r
e
of
t
h
ese
n
etwo
r
k
s
d
em
an
d
s
m
o
d
els
th
at
ar
e
not
j
u
s
t
ac
cu
r
ate
b
u
t
h
i
g
h
ly
ef
f
icien
t.
T
h
er
ef
o
r
e,
th
e
p
r
o
p
o
s
ed
r
esea
r
ch
wo
r
k
aim
s
to
d
esig
n
a
n
d
d
ev
elo
p
an
ef
f
ec
tiv
e
an
o
m
aly
-
b
ased
I
DS
wh
ich
is
ca
p
ab
le
of
d
etec
tin
g
d
etec
t
ze
r
o
-
d
ay
with
h
ig
h
r
esp
o
n
s
e
m
e
ch
an
is
m
f
o
r
r
a
p
id
ly
ev
o
lv
in
g
d
ig
ital
lan
d
s
ca
p
e.
T
h
is
p
ap
er
p
r
o
p
o
s
es
an
ad
v
an
ce
d
l
o
n
g
s
h
o
r
t
-
te
r
m
m
em
o
r
y
-
a
u
to
en
co
d
er
(
L
STM
-
AE
)
b
as
ed
s
y
s
tem
f
o
r
d
etec
tin
g
an
o
m
alies
in
n
etwo
r
k
tr
af
f
ic,
a
cr
itical
co
m
p
o
n
en
t
f
o
r
s
ec
u
r
in
g
f
u
tu
r
e
n
etwo
r
k
in
f
r
astru
ct
u
r
es.
By
in
teg
r
atin
g
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
L
STM
)
u
n
its
in
t
h
e
au
to
en
co
d
er
'
s
ar
ch
itectu
r
e,
th
e
m
o
d
el
ca
p
tu
r
es
lo
n
g
-
ter
m
d
e
p
en
d
e
n
cies,
m
ak
i
n
g
it
ad
ep
t
at
id
en
tify
in
g
co
m
p
lex
,
p
er
s
is
ten
t
th
r
ea
ts
.
Utilizi
n
g
th
e
NSL
-
KDD
d
ataset
an
d
f
o
cu
s
in
g
on
HT
T
P
tr
af
f
ic
f
ea
tu
r
es
alig
n
s
with
th
e
p
r
e
v
alen
t
u
s
e
of
i
n
ter
n
et
p
r
o
to
co
ls
in
I
o
T
n
etwo
r
k
s
,
th
u
s
en
h
an
cin
g
th
e
s
y
s
tem
'
s
p
r
ac
tical
r
elev
an
c
e
f
o
r
r
ea
l
-
w
o
r
ld
ap
p
licatio
n
s
.
T
h
e
ap
p
r
o
ac
h
is
d
esig
n
ed
to
d
is
ce
r
n
an
d
clas
s
if
y
in
tr
icate
m
alicio
u
s
tr
af
f
ic
p
atter
n
s
ef
f
ec
tiv
ely
,
en
s
u
r
i
n
g
r
o
b
u
s
t
d
ef
en
s
e
m
ec
h
an
is
m
s
in
th
e
e
v
o
lv
in
g
la
n
d
s
ca
p
e
of
n
etwo
r
k
s
ec
u
r
ity
.
T
h
e
m
ain
c
o
n
tr
ib
u
tio
n
s
of
th
e
p
r
o
p
o
s
ed
s
tu
d
y
a
r
e
h
ig
h
lig
h
ted
as
f
o
llo
ws:
–
T
h
e
s
tr
ateg
ic
in
teg
r
atio
n
of
L
STM
u
n
its
with
in
b
o
th
th
e
en
co
d
er
an
d
d
ec
o
d
e
r
s
eg
m
en
ts
to
ca
p
tu
r
e
lo
n
g
-
ter
m
d
ep
e
n
d
en
cies
in
n
et
wo
r
k
tr
af
f
ic
.
–
E
m
p
lo
y
in
g
s
tatis
tical
tech
n
iq
u
es
an
d
s
u
p
er
v
is
ed
class
if
ic
atio
n
f
o
r
p
r
ec
is
e
f
ea
tu
r
e
s
el
ec
tio
n
f
r
o
m
th
e
NSL
-
KDD
d
ataset
an
d
HT
T
P
tr
af
f
ic.
–
Dev
elo
p
m
en
t
of
a
n
o
v
el
ap
p
r
o
ac
h
f
o
r
c
o
m
p
u
tin
g
d
is
tin
ct
th
r
esh
o
ld
s
f
o
r
v
ar
i
o
u
s
attac
k
cla
s
s
es,
en
h
an
cin
g
th
e
m
o
d
el'
s
s
p
ec
if
icity
in
an
o
m
aly
d
etec
tio
n
.
–
Desig
n
in
g
of
a
lig
h
tweig
h
t
a
n
d
in
tellig
en
t
ar
ch
itectu
r
e
th
a
t
is
ad
ap
tab
le
to
t
h
e
co
m
p
lex
ities
of
m
o
d
er
n
,
m
u
lti
-
d
o
m
ain
n
etwo
r
k
e
n
v
ir
o
n
m
en
ts
to
i
d
en
tify
n
o
v
el
th
r
ea
ts
th
at
h
av
e
not
b
ee
n
p
r
e
v
io
u
s
ly
en
co
u
n
ter
ed
,
ad
d
r
ess
in
g
a
cr
itical
n
ee
d
in
c
y
b
er
s
ec
u
r
ity
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
13
,
No
.
4
,
Dec
em
b
er
20
24
:
4
8
9
5
-
4
9
0
5
4898
2.
M
E
T
H
O
D
T
h
e
p
r
o
p
o
s
ed
s
tu
d
y
in
tr
o
d
u
c
es
a
n
o
v
el
L
STM
-
AE
b
ased
an
o
m
aly
d
etec
tio
n
s
y
s
tem
(
ADS)
to
ad
d
r
ess
co
m
p
lex
ities
ass
o
ciat
ed
with
ca
p
tu
r
in
g
tem
p
o
r
al
d
ep
en
d
en
cies
in
th
e
n
etwo
r
k
tr
af
f
ic,
wh
ich
is
a
is
in
h
er
en
tly
s
eq
u
en
tial,
with
d
a
ta
p
o
in
ts
o
cc
u
r
r
in
g
at
s
p
ec
if
ic
tim
estam
p
s
.
L
STM
s
ar
e
s
p
ec
if
ically
d
esig
n
ed
to
h
an
d
le
s
u
ch
tim
e
s
er
ies
d
ata,
ca
p
tu
r
in
g
th
e
tem
p
o
r
al
d
ep
e
n
d
en
cies
an
d
r
elatio
n
s
h
ip
s
b
e
twee
n
co
n
s
ec
u
tiv
e
o
b
s
er
v
atio
n
s
.
By
lear
n
in
g
to
r
ec
o
n
s
tr
u
ct
n
o
r
m
al
tr
af
f
ic
s
eq
u
en
ce
s
,
th
e
L
STM
-
AE
au
t
o
m
atica
lly
ex
tr
ac
ts
r
elev
an
t
f
ea
tu
r
es
th
at
ch
ar
ac
te
r
ize
n
o
r
m
al
n
etwo
r
k
b
e
h
av
io
r
.
T
h
is
f
ea
tu
r
e
ex
tr
ac
tio
n
ca
p
a
b
ilit
y
is
cr
u
cial
f
o
r
an
o
m
aly
d
etec
tio
n
,
as
an
o
m
al
ies
o
f
ten
d
ev
iate
f
r
o
m
th
ese
lear
n
ed
p
atter
n
s
.
T
h
e
p
r
o
p
o
s
e
d
L
STM
-
AE
b
ased
ADS
can
id
en
tif
y
ze
r
o
-
d
a
y
att
ac
k
s
,
d
y
n
am
ic
a
n
d
n
o
v
el
th
r
e
ats
th
at
h
av
e
not
b
ee
n
en
c
o
u
n
ter
ed
b
e
f
o
r
e,
u
n
lik
e
s
ig
n
atu
r
e
-
b
ased
I
DS
th
at
d
ep
e
n
d
on
p
r
ed
e
f
in
ed
p
atter
n
s
.
T
h
is
is
cr
u
cial
in
th
e
d
y
n
am
ic
a
n
d
e
v
o
lv
in
g
th
r
ea
ts
in
th
e
f
u
tu
r
is
tic
n
etwo
r
k
lik
e
I
o
E
ec
o
s
y
s
tem
,
wh
er
e
n
ew
v
u
l
n
er
ab
ilit
ies
an
d
attac
k
m
et
h
o
d
s
em
er
g
e
f
r
eq
u
en
tly
an
d
n
ec
ess
itates
ad
v
an
ce
d
d
etec
tio
n
m
ec
h
an
is
m
s
th
at
ar
e
b
o
th
p
r
ec
is
e
an
d
a
d
ap
tab
le.
T
h
e
p
r
o
p
o
s
ed
L
STM
-
AE
b
ased
ADS
is
d
esig
n
ed
to
lear
n
an
d
id
en
tify
n
o
r
m
al
tr
af
f
ic
p
atter
n
s
,
th
er
e
b
y
d
etec
tin
g
d
ev
iatio
n
s
th
at
m
ay
in
d
icate
s
ec
u
r
ity
b
r
e
ac
h
es.
Fu
r
th
er
m
o
r
e,
th
is
s
tu
d
y
en
h
a
n
ce
s
th
e
ADS
f
r
am
ew
o
r
k
by
en
a
b
lin
g
th
e
m
o
d
el
to
c
o
m
p
u
te
d
is
tin
ct
th
r
esh
o
ld
s
f
o
r
v
ar
io
u
s
attac
k
cl
ass
es.
T
h
er
ef
o
r
e,
th
e
p
r
o
p
o
s
e
d
L
STM
-
AE
b
ased
ADS
f
u
n
ctio
n
s
s
im
ilar
ly
to
an
I
DS
but
with
e
n
h
an
ce
d
ca
p
ab
ilit
ies.
I
ts
p
r
im
ar
y
f
u
n
ctio
n
is
to
d
etec
t
a
n
o
m
alies
by
r
ec
o
g
n
izin
g
d
ev
iatio
n
s
f
r
o
m
lear
n
ed
n
o
r
m
al
p
atter
n
s
.
M
o
r
eo
v
e
r
,
it
s
u
r
p
ass
es
tr
ad
itio
n
al
I
DS
by
not
o
n
ly
id
en
tify
in
g
an
o
m
alies
b
u
t
also
class
if
y
in
g
th
em
b
ased
on
th
e
s
p
ec
if
ic
ty
p
e
of
attac
k
,
m
ak
in
g
it
a
m
o
r
e
ef
f
ec
tiv
e
an
d
s
o
p
h
is
ticated
ap
p
r
o
ac
h
f
o
r
n
etwo
r
k
s
ec
u
r
ity
.
F
ig
u
r
e
1
p
r
esen
ts
th
e
s
ch
em
atic
ar
ch
itectu
r
e
of
th
e
p
r
o
p
o
s
ed
L
STM
-
AE
b
ased
A
DS
f
o
r
f
u
tu
r
is
tic
n
etwo
r
k
.
Fig
u
r
e
1.
Flo
w
d
ia
g
r
am
of
th
e
p
r
o
p
o
s
ed
L
STM
-
AE
b
ased
A
DS
T
he
p
r
o
p
o
s
ed
L
STM
-
AE
b
ased
f
r
am
ewo
r
k
co
n
s
is
ts
of
m
u
ltip
le
co
m
p
u
tatio
n
al
b
lo
c
k
s
each
in
p
r
o
g
r
ess
iv
e
an
d
s
eq
u
e
n
tial
m
an
n
er
a
s
s
h
o
wn
in
Fig
u
r
e
1
.
Fir
s
tly
,
th
e
ad
o
p
ted
d
ataset
is
ex
p
lo
r
ed
an
d
an
al
y
ze
d
to
ex
tr
ac
t
s
ig
n
if
ican
t
in
s
ig
h
ts
on
th
e
d
ataset
s
am
p
les.
Seco
n
d
ly
,
t
h
e
p
r
ep
r
o
ce
s
s
in
g
o
p
er
ati
o
n
is
ca
r
r
ied
f
o
r
d
ata
n
o
r
m
aliza
tio
n
a
n
d
f
ea
tu
r
e
v
e
cto
r
izatio
n
.
T
h
e
n
o
p
tim
al
f
ea
tu
r
es
ar
e
s
elec
ted
an
d
f
ed
to
tr
ain
L
STM
-
AE
to
lear
n
th
e
n
o
r
m
al
tr
af
f
ic
p
atter
n
.
T
h
e
f
r
a
m
ewo
r
k
u
s
es
a
d
is
tan
ce
f
u
n
ctio
n
-
b
ased
to
co
m
p
u
tes
r
ec
o
n
s
tr
u
ctio
n
er
r
o
r
a
n
d
u
s
in
g
s
tatis
tical
an
aly
s
is
a
s
u
itab
le
th
r
esh
o
ld
is
d
et
er
m
in
ed
f
o
r
a
n
o
m
aly
d
etec
tio
n
.
T
h
e
p
er
f
o
r
m
an
ce
of
m
o
d
el
is
v
alid
ated
in
ter
m
s
of
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all
r
ate
an
d
F1
-
s
co
r
e.
2
.
1
.
Da
t
a
s
et
T
h
e
p
r
o
p
o
s
ed
s
tu
d
y
h
as
ad
o
p
ted
NSL
-
KDD
d
ataset,
cr
ea
ted
b
ased
on
s
im
u
latio
n
of
a
v
ar
iety
of
in
tr
u
s
io
n
s
with
in
a
m
ilit
ar
y
n
etwo
r
k
en
v
ir
o
n
m
en
t.
T
h
e
s
tatis
tics
of
th
e
NSL
-
KDD
d
ata
s
et
is
i
llu
s
tr
ated
in
T
ab
le
1.
It
in
h
er
en
tly
p
r
o
v
id
es
p
r
e
-
s
p
lit
tr
ain
in
g
a
n
d
test
in
g
s
ets,
with
th
e
class
if
icatio
n
task
ca
n
ter
ed
ar
o
u
n
d
a
r
esp
o
n
s
e
v
a
r
iab
le.
A
u
n
iq
u
e
c
h
ar
ac
ter
is
tic
of
th
e
NSL
-
KDD
d
ataset
is
th
e
in
clu
s
io
n
of
a
more
attac
k
s
am
p
les
in
th
e
test
s
et
co
m
p
a
r
ed
to
t
h
e
tr
ain
in
g
s
et,
r
ef
lectin
g
r
ea
l
-
wo
r
ld
s
ce
n
ar
io
s
wh
e
r
e
n
ew
attac
k
ty
p
es
o
f
ten
em
er
g
es
af
ter
d
e
v
elo
p
m
e
n
t
of
th
e
m
o
d
el.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
A
n
o
ve
l d
ee
p
a
n
o
ma
ly
d
etec
ti
o
n
a
p
p
r
o
a
ch
fo
r
in
tr
u
s
io
n
d
ete
ctio
n
in
…
(
S
a
i Kr
is
h
n
a
La
ksh
min
a
r
a
ya
n
a
)
4899
T
ab
le
1.
Statis
tics
of
da
taset
with
r
esp
ec
t
to
n
u
m
b
er
of
class
es
Tr
a
f
f
i
c
c
l
a
sses
Tr
a
i
n
i
n
g
se
t
Te
st
i
n
g
s
e
t
To
t
a
l
sam
p
l
e
To
t
a
l
sam
p
l
e
N
o
r
mal
6
7
3
4
3
9
7
1
1
D
o
S
4
5
9
2
7
5
7
4
1
P
r
o
b
e
1
1
6
5
6
1
1
0
6
R
2
L
9
9
5
5
9
4
9
U
2
R
52
37
2
.
2
.
P
re
pro
ce
s
s
ing
Pre
p
r
o
ce
s
s
in
g
is
one
of
th
e
c
r
itical
s
tep
s
r
eq
u
ir
ed
in
an
y
d
ata
-
d
r
iv
en
task
to
ac
h
iev
e
ac
c
u
r
ate,
an
d
un
-
b
iased
class
if
icatio
n
r
esu
lts
.
T
h
e
p
r
o
p
o
s
ed
p
r
ep
r
o
ce
s
s
in
g
o
p
er
atio
n
s
ad
o
p
te
d
in
th
e
p
r
o
p
o
s
ed
L
STM
-
AE
b
ased
ADS
f
r
am
ewo
r
k
in
v
o
l
v
es
clea
n
in
g
th
e
d
ataset
to
r
em
o
v
e
n
o
is
e,
o
u
tlier
s
,
an
d
ir
r
e
lev
an
t
in
f
o
r
m
atio
n
.
Fu
r
th
er
d
ata
tr
an
s
f
o
r
m
atio
n
t
ec
h
n
iq
u
es
s
u
ch
as
n
o
r
m
aliza
t
io
n
an
d
en
c
o
d
in
g
a
r
e
ap
p
lied
to
en
s
u
r
e
th
at
th
e
L
STM
-
AE
r
ec
eiv
es
in
p
u
t
th
at
ac
cu
r
ately
r
ef
lects
th
e
u
n
d
er
l
y
in
g
p
atter
n
s
of
n
etwo
r
k
b
e
h
av
i
o
r
.
2
.
2
.
1
.
Da
t
a
clea
nin
g
T
h
e
in
itial
d
ata
clea
n
in
g
o
p
e
r
atio
n
f
o
cu
s
s
es
on
id
en
tify
in
g
m
is
s
in
g
in
s
tan
ce
s
,
s
u
b
s
titu
tin
g
it
with
NaN
v
alu
es
an
d
th
en
r
em
o
v
al
of
th
ese
NaN
to
ac
h
iev
e
co
m
p
leten
ess
an
d
en
s
u
r
es
th
e
in
teg
r
ity
of
th
e
d
ataset,
s
u
ch
th
at
=
{
∈
|
is
n
otN
A
N
}
,
wh
er
e
r
ep
r
esen
ts
th
e
o
r
ig
i
n
al
d
ataset
an
d
r
ep
r
esen
ts
th
e
d
ataset
af
ter
r
em
o
v
al
of
Na
N
v
alu
es.
T
h
e
n
ex
t
c
o
n
s
id
er
ed
in
th
is
d
ata
clea
n
in
g
o
p
er
atio
n
is
r
em
o
v
al
of
o
u
tlier
s
u
s
in
g
m
ed
ian
ab
s
o
lu
te
d
ev
iatio
n
esti
m
ato
r
(
MA
DE
)
m
eth
o
d
d
ef
in
ed
as
f
o
llo
ws:
=
(
|
−
(
)
|
)
(
1
)
|
−
(
)
|
>
×
(
2
)
W
h
er
e
in
(
1
)
,
(
∙
)
d
en
o
tes
a
f
u
n
ctio
n
of
m
ed
ian
o
p
e
r
atio
n
a
n
d
in
(
2
)
,
r
ep
r
esen
ts
a
co
n
s
ta
n
t,
s
et
to
1
.
3
6
9
3
f
o
r
n
o
r
m
ally
d
is
tr
ib
u
ted
d
ata
to
ac
h
iev
e
ap
p
r
o
x
i
m
ately
a
68%
co
n
s
is
ten
cy
with
th
e
s
tan
d
ar
d
d
e
v
iatio
n
.
2
.
2
.
2
.
No
rma
liza
t
io
n
a
nd
s
ca
lin
g
Data
n
o
r
m
aliza
tio
n
is
a
cr
u
cia
l
p
r
o
ce
s
s
in
d
ata
p
r
ep
r
o
ce
s
s
in
g
o
p
er
atio
n
to
en
s
u
r
e
th
at
ea
c
h
f
ea
tu
r
e
of
th
e
d
ataset
co
n
tr
ib
u
tes
eq
u
ally
to
th
e
p
r
ed
ictiv
e
m
o
d
ellin
g
.
T
h
e
p
r
o
p
o
s
ed
s
tu
d
y
im
p
lem
e
n
ts
m
in
-
m
ax
s
ca
lin
g
to
tr
an
s
f
o
r
m
th
e
f
ea
tu
r
e
v
alu
es
to
a
s
tan
d
ar
d
r
an
g
e
of
[
0
,
1
]
.
T
h
e
s
ca
led
v
alu
e
′
is
co
m
p
u
ted
u
s
in
g
(
3
)
:
′
=
−
m
i
n
(
)
m
ax
(
)
−
m
i
n
(
)
(
3
)
W
h
er
e
is
th
e
o
r
i
g
in
al
d
ata,
min
(
)
is
th
e
m
in
im
u
m
v
al
u
e,
an
d
m
a
x
(
)
is
th
e
m
ax
im
u
m
v
al
u
e
of
th
e
ℎ
f
ea
tu
r
e
ac
r
o
s
s
en
tire
d
ataset
in
s
tan
ce
s
.
T
h
is
s
ca
lin
g
tech
n
iq
u
e
en
h
an
ce
s
th
e
ab
ilit
y
of
th
e
L
STM
-
AE
m
o
d
el
to
lear
n
an
d
d
etec
t
an
o
m
alies
ef
f
ec
tiv
ely
.
2
.
2
.
3
.
E
nco
din
g
C
ateg
o
r
ical
v
ar
iab
les
ar
e
co
n
v
er
ted
to
a
n
u
m
er
ical
f
o
r
m
at
u
s
in
g
one
-
hot
e
n
co
d
in
g
.
Fo
r
a
c
ateg
o
r
ical
v
ar
iab
le
with
u
n
iq
u
e
v
al
u
es,
one
-
hot
en
co
d
in
g
cr
ea
t
es
n
b
in
a
r
y
{
1
,
2
,
⋯
,
}
v
ar
iab
les
ea
c
h
r
ep
r
esen
tin
g
one
of
th
e
v
alu
es
in
.
T
h
e
en
c
o
d
in
g
is
p
er
f
o
r
m
ed
as
(
4
)
:
{
1
,
if
=
j
0
,
othe
r
w
ise
(
4
)
In
(
4
)
,
is
th
e
en
co
d
ed
b
in
ar
y
v
ar
iab
le
f
o
r
th
e
i
th
in
s
tan
ce
an
d
j
th
th
e
ca
teg
o
r
y
.
T
h
is
p
r
o
ce
s
s
is
ess
en
tial
f
o
r
co
n
v
er
tin
g
n
o
n
-
n
u
m
er
ic
d
ata
in
to
a
m
ac
h
in
e
-
r
ea
d
a
b
le
f
o
r
m
at
an
d
to
e
n
s
u
r
e
th
at
th
e
L
STM
-
AE
m
o
d
el
ca
n
f
au
ltles
s
ly
p
r
o
ce
s
s
ca
teg
o
r
ical
d
ata,
p
r
eser
v
in
g
th
e
in
f
o
r
m
ati
o
n
al
co
n
te
n
t
f
o
r
an
o
m
al
y
d
ete
ctio
n
.
2
.
3
.
F
e
a
t
ure
s
elec
t
io
n
T
h
e
f
ea
tu
r
e
s
elec
tio
n
p
lay
s
an
im
p
o
r
tan
t
r
o
le
in
th
e
p
r
ed
ictiv
e
m
o
d
ellin
g
as
it
en
h
an
ce
s
th
e
g
en
er
aliza
tio
n
ca
p
ab
ilit
y
of
t
h
e
lear
n
in
g
m
o
d
el
by
r
ed
u
ci
n
g
d
im
en
s
io
n
ality
an
d
elim
in
atin
g
r
ed
u
n
d
an
t
or
ir
r
elev
an
t
d
ata.
T
h
e
p
r
o
p
o
s
e
d
s
tu
d
y
em
p
lo
y
s
d
u
al
ap
p
r
o
ac
h
of
f
ea
tu
r
e
s
elec
tio
n
c
o
n
s
id
er
in
g
p
ea
r
s
o
n
co
r
r
elatio
n
c
o
ef
f
icien
t
(
PC
C
)
an
d
RF
f
o
r
f
ea
t
u
r
e
r
a
n
k
in
g
.
T
h
e
PC
C
is
a
m
ea
s
u
r
e
th
at
r
an
g
es
f
r
o
m
-
1
to
+1
,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
13
,
No
.
4
,
Dec
em
b
er
20
24
:
4
8
9
5
-
4
9
0
5
4900
wh
er
e
+1
in
d
icate
s
a
p
er
f
ec
t
p
o
s
itiv
e
lin
ea
r
r
elatio
n
s
h
ip
,
-
1
in
d
icate
s
a
p
er
f
ec
t
n
eg
ativ
e
lin
ea
r
r
elatio
n
s
h
ip
,
an
d
0
s
ig
n
if
ies
no
lin
ea
r
c
o
r
r
el
atio
n
,
n
u
m
er
ically
g
iv
e
n
as
(
5
)
:
(
,
)
=
∑
(
−
)
(
−
)
√
∑
(
−
)
2
∑
(
−
)
2
(
5
)
W
h
er
e,
(
in
p
u
t
p
r
ed
icto
r
s
)
,
(
tar
g
et
r
esp
o
n
s
e)
,
a
n
d
is
m
ea
n
v
alu
e
of
an
d
,
r
esp
ec
tiv
ely
f
e
atu
r
es
with
a
PC
C
clo
s
e
to
±1
is
co
n
s
id
er
ed
h
ig
h
ly
co
r
r
elate
d
with
th
e
o
u
tp
u
t
class
.
Su
b
s
eq
u
e
n
tly
,
t
h
e
s
tu
d
y
im
p
le
m
en
ts
RF
cla
s
s
if
ier
to
r
an
k
th
e
f
ea
tu
r
es
b
ased
on
th
eir
im
p
o
r
tan
ce
,
wh
ich
is
d
er
iv
ed
f
r
o
m
how
we
ll
th
ey
im
p
r
o
v
e
t
h
e
p
u
r
ity
of
t
h
e
n
o
d
e,
co
n
s
id
er
i
n
g
th
e
d
ec
r
ea
s
e
in
im
p
u
r
ity
ac
r
o
s
s
all
tr
ee
s
in
th
e
f
o
r
est.
Fig
u
r
e
2
illu
s
tr
ates
th
e
im
p
o
r
tan
ce
of
f
ea
tu
r
e
b
ased
on
th
eir
r
a
n
k
in
g
d
eter
m
in
e
d
by
th
e
RF
class
if
ier
.
Fig
u
r
e
2.
I
ll
u
s
tr
atio
n
of
t
o
p
f
e
atu
r
es
r
an
k
ed
ac
co
r
d
in
g
to
th
e
ir
im
p
o
r
tan
ce
Fig
u
r
e
2
illu
s
tr
ates
th
e
f
ea
tu
r
e
r
an
k
i
n
g
with
t
h
eir
r
elativ
e
i
m
p
o
r
tan
ce
of
each
f
ea
tu
r
e
in
d
escen
d
in
g
o
r
d
er
.
B
ased
on
th
e
in
s
ig
h
ts
g
ain
ed
f
r
o
m
PC
C
an
aly
s
is
an
d
f
ea
tu
r
e
r
an
k
in
g
th
e
s
tu
d
y
s
elec
ted
th
e
to
p
30
f
ea
tu
r
es
f
o
cu
s
in
g
on
th
o
s
e
th
at
co
n
s
is
ten
tly
s
h
o
w
a
h
ig
h
co
r
r
elatio
n
or
im
p
o
r
tan
ce
ac
r
o
s
s
b
o
th
m
eth
o
d
s
.
Af
ter
f
ea
tu
r
e
s
elec
tio
n
,
to
p
r
e
p
ar
e
t
h
e
in
p
u
t
d
ata
f
o
r
th
e
L
STM
-
AE
,
th
e
s
tu
d
y
p
er
f
o
r
m
s
r
esh
ap
in
g
th
e
d
ata
to
r
e
f
lect
th
e
s
eq
u
en
tial
n
atu
r
e
r
eq
u
ir
ed
f
o
r
L
STM
p
r
o
ce
s
s
in
g
.
T
h
e
d
ataset,
co
n
s
is
tin
g
of
a
s
er
ies
of
tim
e
-
s
eq
u
en
ce
d
o
b
s
er
v
atio
n
s
[
1
,
2
,
3
,
⋯
,
]
,
is
o
r
g
an
ized
in
to
s
eq
u
en
ce
s
wh
er
e
each
s
e
q
u
en
ce
h
as
a
f
ix
ed
len
g
th
.
E
ac
h
s
eq
u
en
ce
is
a
win
d
o
w
of
co
n
s
ec
u
tiv
e
tim
e
s
tep
s
[
1
,
2
,
3
]
,
an
d
each
tim
e
s
tep
is
a
v
ec
to
r
in
r
ep
r
esen
tin
g
f
ea
tu
r
es
at
tim
e
.
T
h
e
r
esh
ap
in
g
p
r
o
ce
s
s
in
v
o
lv
es
tr
an
s
f
o
r
m
in
g
th
e
d
ata
in
to
2
-
D
ar
r
ay
s
wh
er
e
th
e
f
ir
s
t
d
im
en
s
io
n
is
t
h
e
n
u
m
b
e
r
of
s
am
p
les
(
s
eq
u
en
ce
s
)
,
an
d
th
e
s
ec
o
n
d
d
im
e
n
s
io
n
is
th
e
n
u
m
b
er
of
tim
e
s
tep
s
,
with
ea
c
h
tim
e
s
tep
co
n
tain
in
g
th
e
m
f
ea
tu
r
es.
T
h
is
r
esu
lts
in
an
ar
r
a
y
s
h
ap
e
of
[
s
am
p
les,
tim
e
s
tep
s
×
f
ea
tu
r
es],
wh
ich
is
s
u
itab
le
f
o
r
in
p
u
t
in
to
th
e
L
STM
lay
e
r
of
th
e
au
to
en
c
o
d
er
.
2
.
4
.
L
o
ng
s
ho
rt
-
t
er
m
m
emo
ry
-
a
uto
enco
der
m
o
del
f
o
r
ne
t
wo
rk
a
no
m
a
l
y
det
ec
t
io
n
T
h
e
p
r
o
p
o
s
ed
L
STM
-
AE
m
o
d
el
in
teg
r
ates
th
e
s
tr
en
g
th
s
of
L
STM
n
etwo
r
k
s
an
d
a
u
to
e
n
co
d
er
s
to
an
aly
ze
h
ig
h
-
d
im
en
s
io
n
al
tim
e
-
s
er
ies
d
ata
ef
f
icien
tly
.
In
th
is
ar
ch
itectu
r
e,
th
e
L
STM
en
co
d
er
co
m
p
r
ess
es
th
e
in
p
u
t
s
eq
u
e
n
ce
in
to
a
lo
w
-
d
im
en
s
io
n
al
laten
t
v
ec
t
o
r
,
p
r
ese
r
v
in
g
tem
p
o
r
al
d
ep
e
n
d
en
cies
th
r
o
u
g
h
its
m
em
o
r
y
ce
lls
.
T
h
is
p
r
o
ce
s
s
tr
an
s
f
o
r
m
s
th
e
co
m
p
lex
in
p
u
t
d
ata
in
to
a
s
im
p
ler
f
o
r
m
th
at
ca
p
tu
r
es
its
ess
en
tial
p
atter
n
s
.
Su
b
s
eq
u
en
tly
,
th
e
L
STM
d
ec
o
d
er
r
ec
o
n
s
tr
u
cts
th
e
o
r
ig
in
al
s
eq
u
en
ce
f
r
o
m
th
is
laten
t
r
ep
r
esen
tatio
n
,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
A
n
o
ve
l d
ee
p
a
n
o
ma
ly
d
etec
ti
o
n
a
p
p
r
o
a
ch
fo
r
in
tr
u
s
io
n
d
ete
ctio
n
in
…
(
S
a
i Kr
is
h
n
a
La
ksh
min
a
r
a
ya
n
a
)
4901
em
p
lo
y
in
g
r
ec
o
n
s
tr
u
ctio
n
er
r
o
r
to
d
eter
m
in
e
an
o
m
aly
d
e
tectio
n
th
r
esh
o
ld
s
.
Fig
u
r
e
3
d
em
o
n
s
tr
ates
th
e
ar
ch
itectu
r
e,
h
ig
h
lig
h
tin
g
th
e
en
co
d
er
-
d
ec
o
d
er
m
ec
h
an
is
m
a
n
d
th
e
cr
itical
r
o
le
of
L
STM
u
n
its
in
m
ain
tain
i
n
g
s
eq
u
en
ce
in
teg
r
ity
wh
ile
f
ac
ilit
atin
g
d
im
en
s
io
n
ality
r
e
d
u
ctio
n
.
T
h
e
L
STM
u
n
it
is
an
ad
v
a
n
ce
d
r
ec
u
r
r
en
t
n
e
u
r
al
n
etwo
r
k
(
R
NN)
m
o
d
el
th
at
ad
d
r
ess
es
th
e
is
s
u
es
of
v
an
is
h
in
g
g
r
ad
ien
t
by
e
f
f
ec
tiv
ely
ca
p
tu
r
in
g
lo
n
g
-
ter
m
d
e
p
en
d
en
cies
in
s
eq
u
e
n
tial
d
ata.
As
r
ep
r
esen
ted
in
Fig
u
r
e
3
(
a)
,
th
e
L
STM
-
u
n
it
is
co
m
p
o
s
ed
of
co
m
p
lex
m
ec
h
a
n
is
m
s
in
v
o
lv
in
g
m
u
ltip
le
g
ates
th
at
r
eg
u
late
th
e
f
lo
w
of
in
f
o
r
m
atio
n
.
T
h
e
f
ir
s
t
is
th
e
f
o
r
g
et
-
g
ate,
wh
ich
d
ete
r
m
in
es
wh
at
in
f
o
r
m
atio
n
is
r
ej
ec
ted
f
r
o
m
t
h
e
ce
ll
s
tate.
It
lo
o
k
s
at
th
e
p
r
ev
io
u
s
h
id
d
en
s
tate
−
1
an
d
th
e
cu
r
r
e
n
t
i
n
p
u
t
an
d
ap
p
lies
a
s
ig
m
o
id
f
u
n
ctio
n
to
d
ec
id
e
wh
ich
v
alu
es
ar
e
allo
wed
to
p
ass
,
r
an
g
in
g
f
r
o
m
0
(
co
m
p
letely
f
o
r
g
o
tten
)
to
1
(
c
o
m
p
letely
r
etain
ed
)
.
At
th
e
s
am
e
tim
e,
th
e
in
p
u
t
g
ate
d
eter
m
in
es
wh
at
n
ew
in
f
o
r
m
a
tio
n
is
ad
d
e
d
to
th
e
u
n
it
s
tate.
It
u
s
es
a
s
ig
m
o
id
lay
er
to
u
p
d
ate
th
e
ce
ll
s
tate,
an
d
tan
h
to
b
u
ild
a
v
ec
t
o
r
of
n
ew
ca
n
d
id
ate
v
al
u
es
to
ad
d
to
th
e
s
tate.
T
h
e
ce
ll
-
s
tate
is
u
p
d
ated
by
m
u
ltip
ly
in
g
th
e
o
l
d
s
tate
−
1
by
t
h
e
o
u
tp
u
t
of
t
h
e
f
o
r
g
et
g
ate
to
d
is
ca
r
d
u
n
n
ec
ess
ar
y
in
f
o
r
m
atio
n
an
d
a
d
d
th
e
o
u
tp
u
t
of
t
h
e
in
p
u
t
g
ate
as
n
ew
in
f
o
r
m
atio
n
.
T
h
e
f
in
al
g
ate
in
th
e
L
STM
u
n
it
d
eter
m
i
n
es
th
e
n
e
x
t
h
id
d
en
s
tate
,
wh
ich
co
n
tain
s
in
f
o
r
m
atio
n
ab
o
u
t
th
e
p
r
e
v
io
u
s
in
p
u
ts
.
It
is
co
m
p
u
ted
b
ased
on
th
e
ce
ll
s
tate
p
ass
ed
th
r
o
u
g
h
a
tan
h
-
lay
e
r
(
to
n
o
r
m
alize
th
e
v
alu
es)
an
d
th
en
m
u
ltip
lied
by
th
e
o
u
tp
u
t
of
th
e
s
ig
m
o
id
g
ate,
en
s
u
r
in
g
th
at
o
n
ly
th
e
n
ec
ess
ar
y
in
f
o
r
m
atio
n
is
p
ass
ed
o
n
.
As
d
ep
icted
in
Fig
u
r
e
3
(
b
)
,
th
e
ar
ch
itectu
r
e
of
an
a
u
to
en
c
o
d
er
co
n
s
is
ts
of
two
m
ain
p
ar
ts
s
u
ch
as
th
e
en
co
d
er
a
n
d
th
e
d
ec
o
d
e
r
.
T
h
e
en
co
d
er
c
o
m
p
o
n
en
t
of
th
e
n
e
two
r
k
co
m
p
r
ess
es
th
e
in
p
u
t
i
n
to
a
laten
t
-
s
p
ac
e
r
ep
r
esen
tatio
n
ℎ
.
T
h
e
in
p
u
t
lay
er
,
co
n
s
is
tin
g
of
n
eu
r
o
n
s
1
,
2
,
3
,
⋯
,
,
is
f
u
lly
co
n
n
ec
te
d
to
th
e
late
n
t
s
p
ac
e,
wh
er
e
each
n
eu
r
o
n
hi
r
ep
r
esen
ts
an
en
c
o
d
ed
f
ea
tu
r
e.
T
h
e
en
co
d
er
lear
n
s
to
p
r
eser
v
e
as
m
u
ch
r
elev
an
t
in
f
o
r
m
atio
n
as
p
o
s
s
ib
le
in
th
i
s
r
ed
u
ce
d
r
ep
r
esen
tatio
n
.
T
h
e
laten
t
s
p
ac
e
is
th
e
ce
n
tr
al
la
y
er
of
th
e
n
etwo
r
k
r
ep
r
esen
ts
th
e
c
o
m
p
r
ess
ed
k
n
o
wled
g
e
of
th
e
in
p
u
ts
an
d
h
o
l
d
s
th
e
m
o
s
t
s
ig
n
if
ica
n
t
f
ea
tu
r
es
lear
n
ed
f
r
o
m
th
e
in
p
u
t
d
ata,
wh
ic
h
ar
e
th
e
cr
iti
ca
l
co
m
p
o
n
en
ts
f
o
r
r
ec
o
n
s
tr
u
ctio
n
by
th
e
d
ec
o
d
er
.
In
th
is
p
h
ase,
th
e
n
etwo
r
k
attem
p
ts
to
r
ec
o
n
s
tr
u
ct
th
e
in
p
u
t
d
ata
f
r
o
m
th
e
laten
t
r
ep
r
esen
tatio
n
.
T
h
e
o
u
tp
u
t
lay
er
m
im
ics
th
e
in
p
u
t
lay
er
in
s
ize,
with
n
eu
r
o
n
s
1
′
,
3
′
,
3
′
,
⋯
,
n
′
r
ep
r
esen
tin
g
t
h
e
r
ec
o
n
s
tr
u
cte
d
in
p
u
t
d
ata.
T
h
e
g
o
al
of
th
e
d
ec
o
d
er
is
to
o
u
t
p
u
t
′
th
at
is
as
clo
s
e
as
p
o
s
s
ib
le
to
th
e
o
r
ig
in
al
in
p
u
t
.
In
th
is
m
an
n
er
,
t
h
e
au
to
en
c
o
d
e
r
lear
n
s
to
ca
p
tu
r
e
t
h
e
m
o
s
t
im
p
o
r
tan
t
asp
ec
ts
of
th
e
i
n
p
u
t
d
ata
with
in
th
e
laten
t
s
p
ac
e.
(
a)
(
b
)
Fig
u
r
e
3
.
Neu
r
al
n
etwo
r
k
s
ar
c
h
itectu
r
es
:
(
a)
L
STM
an
d
(
b
)
a
u
to
en
co
d
er
2.
5
.
L
o
ng
s
ho
rt
-
t
er
m
m
emo
ry
-
a
uto
enco
der
m
o
del
t
ra
ini
ng
T
h
e
en
co
d
e
r
in
p
r
o
p
o
s
ed
L
ST
M
-
AE
m
o
d
el
co
m
p
o
s
ed
of
s
t
ac
k
ed
L
STM
lay
er
s
,
lear
n
s
to
co
m
p
r
ess
th
e
h
ig
h
-
d
im
en
s
io
n
al
in
p
u
t
s
e
q
u
en
ce
s
in
to
a
lo
wer
-
d
im
e
n
s
io
n
al
laten
t
r
ep
r
esen
tatio
n
,
ca
p
tu
r
in
g
th
e
ess
en
tial
f
ea
tu
r
es
ess
en
tial
to
r
ec
r
ea
te
th
e
in
p
u
t.
T
h
e
m
o
d
el
r
ec
ei
v
es
in
p
u
t
with
a
s
h
a
p
e
co
r
r
esp
o
n
d
in
g
to
th
e
ch
o
s
e
n
win
d
o
w
len
g
th
a
n
d
n
u
m
b
er
of
f
ea
tu
r
es
(
e.
g
.
,
[
s
am
p
les,
1
0
,
30]
f
o
r
10
tim
estep
s
an
d
30
f
ea
tu
r
es).
T
h
e
f
i
r
s
t
L
STM
lay
er
h
as
128
n
eu
r
o
n
s
an
d
r
etu
r
n
s
s
eq
u
en
ce
s
,
p
ass
in
g
its
o
u
t
p
u
t
to
th
e
s
ec
o
n
d
L
STM
lay
er
with
64
n
eu
r
o
n
s
,
also
r
etu
r
n
in
g
s
eq
u
e
n
ce
s
to
m
ain
tain
th
e
tem
p
o
r
a
l
s
tr
u
ctu
r
e.
T
h
e
th
ir
d
L
STM
lay
er
r
ed
u
ce
s
th
e
d
im
en
s
io
n
ality
f
u
r
th
er
with
32
n
eu
r
o
n
s
an
d
p
r
e
p
ar
in
g
th
e
d
ata
f
o
r
th
e
laten
t
s
p
ac
e
r
ep
r
es
en
tatio
n
.
T
h
is
lay
er
r
ep
licates
th
e
32
-
d
im
en
s
io
n
al
laten
t
r
ep
r
esen
tatio
n
ac
r
o
s
s
th
e
s
eq
u
en
ce
le
n
g
th
to
f
ac
i
litate
th
e
d
ec
o
d
in
g
p
r
o
ce
s
s
.
Af
ter
war
d
s
a
R
ep
ea
tVec
to
r
lay
er
is
u
s
ed
to
d
u
p
licate
th
e
en
co
d
er
'
s
o
u
tp
u
t
a
cr
o
s
s
th
e
tim
estep
s
r
eq
u
ir
ed
by
t
h
e
d
ec
o
d
er
.
T
h
e
d
ec
o
d
in
g
p
r
o
ce
s
s
b
e
g
in
s
with
an
L
STM
la
y
er
with
32
n
eu
r
o
n
s
th
at
r
ec
o
n
s
tr
u
cts
th
e
s
eq
u
en
ce
f
r
o
m
th
e
laten
t
s
p
ac
e
r
ep
r
esen
tatio
n
.
Su
b
s
e
q
u
en
tly
,
a
s
ec
o
n
d
L
STM
lay
er
with
64
n
eu
r
o
n
s
co
n
tin
u
es
th
e
r
ec
o
n
s
tr
u
ctio
n
p
r
o
ce
s
s
,
s
till
r
etu
r
n
in
g
s
eq
u
en
ce
s
.
T
h
e
th
ir
d
an
d
f
in
al
L
STM
d
ec
o
d
er
lay
er
h
as
128
n
eu
r
o
n
s
,
an
d
r
ec
o
n
s
tr
u
c
ts
th
e
s
eq
u
en
ce
at
th
e
o
r
ig
in
al
f
ea
tu
r
e
d
im
en
s
io
n
ality
.
A
T
im
eDistrib
u
ted
wr
ap
p
er
is
ap
p
lied
to
a
d
en
s
e
lay
er
with
a
n
eu
r
alu
n
it
eq
u
al
to
th
e
n
u
m
b
er
of
s
elec
ted
f
ea
tu
r
es
(
i.e
.
,
30
n
eu
r
al
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
13
,
No
.
4
,
Dec
em
b
er
20
24
:
4
8
9
5
-
4
9
0
5
4902
u
n
its
in
p
r
o
p
o
s
ed
co
n
te
x
t)
,
en
s
u
r
in
g
th
e
o
u
tp
u
t
h
as
th
e
s
am
e
s
tr
u
ctu
r
e
as
th
e
in
p
u
t
d
ata.
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
is
co
m
p
iled
with
an
Ad
am
o
p
tim
izer
an
d
(
m
ea
n
s
q
u
ar
e
er
r
o
r
)
MSE
as
th
e
lo
s
s
f
u
n
ctio
n
.
Fu
r
th
er
m
o
r
e,
th
e
tr
ain
in
g
p
r
o
ce
s
s
in
v
o
lv
es
f
ee
d
i
n
g
th
e
m
o
d
el
b
atch
es
of
n
o
r
m
al
tr
af
f
ic
d
ata
an
d
u
s
in
g
b
ac
k
p
r
o
p
ag
atio
n
th
r
o
u
g
h
tim
e
to
u
p
d
ate
th
e
m
o
d
el
wei
g
h
ts
.
T
h
e
m
o
d
el
is
tr
ain
ed
o
v
er
50
ep
o
c
h
s
u
n
til
th
e
r
ec
o
n
s
t
r
u
ctio
n
er
r
o
r
on
th
e
tr
ain
in
g
s
et
s
tab
ilizes.
Alg
o
r
ith
m
1
d
escr
ib
es
th
e
p
r
o
ce
s
s
of
tr
ain
in
g
L
STM
-
AE
f
o
r
ADS.
Alg
o
r
ith
m
1
: T
r
ain
in
g
L
STM
-
AE
f
o
r
ADS
I
n
p
u
t:
∈
ℝ
×
×
a
ten
s
o
r
o
f
n
o
r
m
al
n
etw
o
r
k
tr
af
f
ic
d
ata,
wh
e
r
e
n
is
th
e
n
u
m
b
er
o
f
s
am
p
les,
t
is
tim
estep
s
p
er
s
am
p
le,
an
d
f
is
th
e
n
u
m
b
er
o
f
f
ea
tu
r
es (
3
0
af
ter
f
ea
t
u
r
e
s
elec
tio
n
)
,
an
d
E
(
e
p
o
ch
s
)
,
B
S (
b
a
tch
s
ize)
Ou
tp
u
t:
(
R
ec
o
n
s
tr
u
ctio
n
er
r
o
r
)
,
T
h
(
t
h
r
esh
o
ld
f
o
r
n
o
r
m
al
tr
a
f
f
ic)
,
(
T
r
ain
ed
m
o
d
el
p
ar
am
et
er
s
)
Star
t
1.
I
n
itialize
th
e
L
STM
-
Au
to
en
co
d
er
m
o
d
el
with
a
s
p
ec
if
ied
ar
c
h
itectu
r
e
2.
Fo
r
ea
ch
lay
er
in
th
e
e
n
co
d
er
:
Ad
d
L
STM
lay
er
with
s
p
ec
if
ie
d
n
u
m
b
er
o
f
n
eu
r
o
n
s
ℎ
,
∈
{
1
,
2
,
⋯
,
}
wh
er
e
is
th
e
n
u
m
b
er
o
f
L
ST
M
lay
er
s
in
th
e
en
c
o
d
er
.
3.
Ad
d
R
ep
ea
tVec
to
r
lay
er
to
r
ep
licate
th
e
laten
t r
ep
r
esen
tatio
n
.
4.
Fo
r
ea
ch
lay
er
in
th
e
d
ec
o
d
er
:
Ad
d
L
STM
lay
er
m
ir
r
o
r
in
g
th
e
en
co
d
er
'
s
s
tr
u
ctu
r
e
in
r
e
v
er
s
e
o
r
d
er
.
5.
Ad
d
T
im
eDistrib
u
ted
Den
s
e
l
ay
er
with
f
n
eu
r
o
n
s
to
r
ec
o
n
s
tr
u
ct
th
e
o
r
ig
i
n
al
f
ea
tu
r
e
d
im
en
s
io
n
6.
C
o
m
p
ile
th
e
m
o
d
el
with
th
e
Ad
am
o
p
tim
izer
a
n
d
MSE
lo
s
s
f
u
n
ctio
n
.
7.
R
esh
ap
e
in
p
u
t d
ata
X
to
3
D
f
o
r
m
at
s
u
itab
le
f
o
r
L
STM
lay
e
r
s
:
→
ℎ
∈
ℝ
×
×
8.
T
r
ain
th
e
m
o
d
el
f
o
r
th
e
s
p
ec
if
i
ed
n
u
m
b
er
o
f
E
an
d
B
S.
9.
Af
ter
tr
ain
in
g
,
u
s
e
th
e
tr
ain
ed
m
o
d
el
to
r
ec
o
n
s
tr
u
ct
th
e
i
n
p
u
t
d
ata:
′
=
(
ℎ
)
10.
C
alcu
late
th
e
r
ec
o
n
s
tr
u
ctio
n
e
r
r
o
r
f
o
r
ea
ch
s
am
p
le
u
s
in
g
th
e
E
u
clid
ea
n
d
is
tan
ce
=
√
∑
(
−
′
)
2
=
1
,
f
o
r
=
1
to
n
11.
Ag
g
r
eg
ate
th
e
to
d
eter
m
in
e
a
th
r
esh
o
ld
ℎ
f
o
r
n
o
r
m
al
tr
af
f
ic
u
s
in
g
s
tatis
tical
an
aly
s
is
ℎ
=
(
)
+
×
(
)
wh
er
e
k
is
a
s
ca
lin
g
f
ac
to
r
ch
o
s
en
u
s
in
g
em
p
ir
ical
an
al
y
s
is
.
E
n
d
Up
o
n
th
e
co
m
p
letio
n
of
Alg
o
r
ith
m
1,
a
tr
ain
ed
L
STM
-
AE
m
o
d
el
is
ca
p
ab
le
of
r
ec
o
n
s
tr
u
ctin
g
n
o
r
m
al
n
etwo
r
k
tr
af
f
ic
with
m
in
im
al
er
r
o
r
.
T
h
e
r
ec
o
n
s
tr
u
ctio
n
er
r
o
r
ε
ca
lcu
lated
f
o
r
e
ac
h
s
am
p
le
in
th
e
n
o
r
m
al
tr
a
f
f
ic
d
ataset,
s
er
v
es
as
a
b
aselin
e
f
o
r
d
etec
tin
g
d
e
v
i
atio
n
s
in
d
icativ
e
of
d
if
f
er
en
t
p
o
ten
tial
an
o
m
alies.
T
h
er
ef
o
r
e,
th
e
p
r
o
p
o
s
ed
s
tu
d
y
co
n
s
id
er
s
a
s
et
of
k
n
o
wn
an
o
m
alies,
=
{
1
,
2
,
…
,
}
,
wh
er
e
each
r
ep
r
esen
ts
a
d
if
f
e
r
en
t
class
of
attac
k
an
d
m
is
th
e
to
tal
n
u
m
b
er
of
attac
k
ty
p
es,
t
h
e
s
y
s
tem
f
ir
s
t
co
m
p
u
tes
th
e
r
ec
o
n
s
tr
u
ctio
n
e
r
r
o
r
f
o
r
each
a
n
o
m
aly
in
s
tan
ce
u
s
in
g
d
is
tan
c
e
f
u
n
ctio
n
(
DF)
f
o
r
m
u
la
as
(
6
)
:
=
√
∑
(
,
−
,
′
)
2
=
1
(
6
)
W
h
er
e,
,
is
th
e
o
r
ig
in
al
d
ata
f
o
r
attac
k
ty
p
e
at
f
ea
tu
r
e
,
an
d
,
′
is
th
e
r
ec
o
n
s
tr
u
cted
d
ata.
F
o
r
ea
c
h
attac
k
ty
p
e
th
e
p
r
o
p
o
s
ed
s
y
s
tem
p
er
f
o
r
m
s
er
r
o
r
d
is
tr
ib
u
tio
n
an
aly
s
is
by
co
n
s
id
er
in
g
th
e
d
is
tr
ib
u
tio
n
of
to
u
n
d
er
s
tan
d
h
o
w
it
d
if
f
er
s
f
r
o
m
th
e
n
o
r
m
al
tr
af
f
ic
er
r
o
r
d
i
s
tr
ib
u
tio
n
.
T
h
e
th
r
esh
o
l
d
ℎ
f
o
r
d
etec
tin
g
each
attac
k
ty
p
e
is
d
eter
m
in
e
d
b
ase
d
on
th
e
s
tatis
tical
an
aly
s
is
of
its
r
ec
o
n
s
tr
u
ctio
n
er
r
o
r
s
,
as
(
7
)
:
ℎ
=
+
∙
(
7
)
W
h
er
e
an
d
ar
e
th
e
m
ea
n
an
d
s
tan
d
ar
d
d
ev
iatio
n
of
th
e
r
ec
o
n
s
tr
u
ctio
n
er
r
o
r
s
f
o
r
attac
k
ty
p
e
an
d
is
a
co
n
s
tan
t
th
at
ad
ju
s
ts
th
e
s
en
s
itiv
ity
of
th
e
th
r
esh
o
ld
.
Fig
u
r
e
4
p
r
esen
ts
th
e
d
is
tr
i
b
u
tio
n
of
id
en
tifie
d
th
r
esh
o
ld
v
al
u
es
f
o
r
d
if
f
er
e
n
t
n
etwo
r
k
tr
af
f
ic
class
es.
T
h
e
Fig
u
r
e
4
s
h
o
ws
d
is
tr
ib
u
tio
n
of
th
r
esh
o
ld
v
alu
es
f
o
r
d
if
f
er
en
t
n
etwo
r
k
tr
a
f
f
ic
class
es.
Fig
u
r
e
4
(
a
)
p
r
esen
ts
v
io
lin
p
lo
t
d
is
p
lay
in
g
th
e
m
ed
ian
of
r
ec
o
n
s
tr
u
ctio
n
er
r
o
r
s
f
o
r
n
o
r
m
al
tr
af
f
ic
an
d
e
ac
h
attac
k
ty
p
e.
A
h
o
r
izo
n
tal
lin
e
r
ep
r
esen
ts
th
e
s
et
th
r
esh
o
ld
v
al
u
e,
wh
ich
d
e
lin
ea
tes
n
o
r
m
al
b
eh
av
io
r
f
r
o
m
p
o
ten
tial
attac
k
s
.
Fig
u
r
e
4
(
b
)
tab
u
lates
th
e
th
r
es
h
o
ld
v
al
u
es
ℎ
co
r
r
esp
o
n
d
i
n
g
to
each
tr
af
f
ic
class
,
d
er
iv
ed
f
r
o
m
s
tatis
tical
an
aly
s
is
of
r
ec
o
n
s
tr
u
ctio
n
er
r
o
r
s
s
er
v
in
g
th
e
cr
iter
ia
f
o
r
d
y
n
a
m
ic
an
o
m
aly
class
if
icatio
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
A
n
o
ve
l d
ee
p
a
n
o
ma
ly
d
etec
ti
o
n
a
p
p
r
o
a
ch
fo
r
in
tr
u
s
io
n
d
ete
ctio
n
in
…
(
S
a
i Kr
is
h
n
a
La
ksh
min
a
r
a
ya
n
a
)
4903
(
a)
(
b
)
Fig
u
r
e
4.
T
h
r
esh
o
ld
v
alu
es
f
o
r
n
etwo
r
k
tr
a
f
f
ic
class
es
:
(
a)
v
is
u
al
d
ep
ictio
n
o
f
id
en
tifie
d
ℎ
v
alu
es
an
d
(
b
)
n
u
m
er
ical
d
e
p
ictio
n
o
f
ℎ
v
alu
es
3.
RE
SU
L
T
S
AND
D
I
SCU
SS
I
O
N
T
h
e
ex
p
er
im
e
n
ts
wer
e
co
n
d
u
c
ted
o
n
test
-
s
et
o
f
NSL
-
KDD
d
atasets
,
an
d
th
e
r
esu
lts
wer
e
co
m
p
ar
ed
with
th
e
s
im
ilar
ex
is
tin
g
wo
r
k
s
.
Fig
u
r
e
5
p
r
o
v
id
es
o
u
tco
m
e
an
aly
s
is
o
f
th
e
p
r
o
p
o
s
ed
L
ST
M
-
AE
s
y
s
tem
with
r
esp
ec
t
to
two
s
u
b
-
f
ig
u
r
es:
i
)
a
co
n
f
u
s
io
n
m
atr
ix
an
d
ii
)
a
r
ec
eiv
er
o
p
e
r
atin
g
c
h
ar
ac
ter
is
tic
(
R
OC
)
cu
r
v
e.
I
n
Fig
u
r
e
5
(
a)
th
e
co
n
f
u
s
io
n
m
atr
ix
s
h
o
ws
th
e
n
u
m
b
er
o
f
c
o
r
r
ec
t
an
d
in
c
o
r
r
ec
t
class
if
icatio
n
s
m
ad
e
b
y
th
e
L
S
T
M
-
A
E
m
o
d
e
l
.
B
as
e
d
o
n
t
h
e
c
l
o
s
e
r
a
n
al
y
s
is
i
n
t
h
e
c
a
s
e
o
f
t
r
u
e
n
e
g
at
i
v
e
(
n
o
r
m
a
l
,
n
o
r
m
a
l
)
w
it
h
8
0
5
3
p
r
e
d
i
c
t
e
d
s
a
m
p
le
s
i
n
d
ic
a
t
i
n
g
t
h
e
c
o
u
n
t
o
f
n
o
r
m
a
l
i
n
s
t
a
n
c
es
a
p
p
r
o
p
r
i
a
t
e
l
y
r
e
c
o
g
n
is
e
d
a
s
n
o
r
m
a
l
,
w
h
i
l
e
f
a
ls
e
p
o
s
i
t
i
v
es
(
n
o
r
m
a
l
,
a
t
t
a
c
k
)
s
h
o
w
1
6
5
7
i
n
s
t
a
n
c
e
s
i
n
c
o
r
r
e
ct
l
y
i
d
e
n
t
i
f
i
e
d
as
a
t
t
a
c
k
s
.
I
n
c
a
s
e
o
f
f
a
l
s
e
n
e
g
a
t
i
v
es
(
a
t
t
a
c
k
,
n
o
r
m
a
l
)
t
h
e
m
o
d
e
l
p
r
ed
i
c
t
e
d
5
2
4
i
n
s
t
a
n
c
es
r
e
p
r
e
s
e
n
t
in
g
t
h
e
n
u
m
b
e
r
o
f
a
t
t
a
c
k
s
t
h
at
w
e
r
e
n
o
t
d
e
te
c
t
e
d
b
y
t
h
e
m
o
d
e
l
,
w
h
i
l
e
t
r
u
e
p
o
s
it
i
v
e
(
a
t
t
a
c
k
,
a
tt
a
c
k
)
w
it
h
1
2
3
0
9
e
x
h
i
b
i
t
i
n
g
t
h
e
n
u
m
b
e
r
o
f
a
t
t
a
c
k
s
c
o
r
r
e
c
t
l
y
i
d
e
n
t
i
f
i
e
d
.
I
n
F
i
g
u
r
e
5
(
b
)
t
h
e
a
n
a
l
y
s
i
s
o
f
R
O
C
c
u
r
v
e
s
h
o
w
s
d
i
a
g
n
o
s
t
i
c
a
b
i
l
i
t
y
o
f
t
h
e
c
l
a
s
s
i
f
i
e
r
c
o
n
c
e
r
n
i
n
g
t
r
u
e
p
o
s
i
t
i
v
e
r
a
t
e
(
T
P
R
)
V
s
f
a
l
s
e
p
o
s
i
t
i
v
e
r
a
t
e
(
F
P
R
)
.
T
h
e
g
r
a
p
h
t
r
e
n
d
i
n
d
i
c
a
t
e
s
a
h
i
g
h
T
P
R
a
n
d
a
l
o
w
F
P
R
a
s
a
r
e
a
u
n
d
e
r
t
h
e
c
u
r
v
e
(
A
U
C
)
w
i
t
h
0
.
9
5
9
s
u
g
g
e
s
t
s
e
x
c
e
l
l
e
n
t
m
o
d
e
l
p
e
r
f
o
r
m
a
n
c
e
,
w
h
e
r
e
a
v
a
l
u
e
o
f
1
.
0
r
e
p
r
e
s
e
n
t
s
a
p
e
r
f
e
c
t
m
o
d
e
l
a
n
d
0
.
5
r
e
p
r
e
s
e
n
t
s
a
n
o
-
s
k
i
l
l
c
l
a
s
s
i
f
i
e
r
.
F
i
g
u
r
e
6
p
r
e
s
e
n
t
s
t
h
e
e
v
a
l
u
a
t
i
o
n
o
f
t
h
e
L
S
T
M
-
A
E
m
o
d
e
l
c
o
n
c
e
r
n
i
n
g
a
n
o
m
a
l
y
s
c
o
r
e
s
o
f
n
e
t
w
o
r
k
t
r
a
f
f
i
c
s
a
m
p
l
e
s
a
n
d
c
o
m
p
a
r
a
t
i
v
e
a
n
a
l
y
s
i
s
w
i
t
h
s
i
m
i
l
a
r
e
x
i
s
t
i
n
g
a
p
p
r
o
a
c
h
e
s
.
(
a)
(
b
)
Fig
u
r
e
5.
Per
f
o
r
m
an
c
e
an
aly
s
i
s
of
th
e
L
STM
-
AE
with
(
a)
co
n
f
u
s
io
n
m
atr
i
x
an
d
(
b
)
R
OC
c
u
r
v
e
B
ased
on
th
e
an
aly
s
is
of
s
ca
tt
er
p
lo
t
in
Fig
u
r
e
6
(
a)
it
can
be
s
ee
n
th
at
th
e
an
o
m
aly
s
co
r
es
of
n
etwo
r
k
tr
af
f
ic
s
am
p
les
wh
er
e
th
e
n
o
r
m
al
tr
af
f
ic
is
m
ar
k
ed
in
g
r
ee
n
,
an
d
an
o
m
alies
ar
e
in
d
icate
d
in
r
ed
.
A
d
ash
ed
b
lu
e
lin
e
r
ep
r
esen
ts
th
e
an
o
m
aly
d
etec
tio
n
th
r
esh
o
ld
;
an
d
s
co
r
es
ab
o
v
e
th
is
lin
e
ar
e
clas
s
if
ied
as
an
o
m
alies.
T
h
e
ch
o
s
en
th
r
esh
o
l
d
is
ef
f
ec
tiv
e
in
s
ep
ar
atin
g
a
n
o
m
alies
f
r
o
m
n
o
r
m
al
s
am
p
les.
Fig
u
r
e
6
(
b
)
p
r
o
v
id
es
a
co
m
p
ar
ativ
e
an
aly
s
is
of
th
e
p
r
o
p
o
s
ed
L
STM
-
AE
m
o
d
el
a
g
ain
s
t
s
ev
er
al
s
im
ilar
k
in
d
of
t
h
e
ex
is
tin
g
r
esear
ch
wo
r
k
s
s
u
ch
as
n
o
n
-
s
y
m
m
etr
ic
d
ee
p
a
u
to
-
en
c
o
d
er
(
NDAE
)
p
r
o
p
o
s
ed
in
[
6
]
,
R
NN
b
ased
I
DS
in
[
3
0
]
,
s
elf
-
tau
g
h
t
lear
n
in
g
(
STL
)
in
tr
o
d
u
ce
d
by
[
3
1
]
,
s
eq
u
en
tial
lear
n
in
g
ac
ce
ler
ato
r
(
SLA)
s
ch
em
e
f
o
r
attac
k
d
etec
tio
n
in
[
3
2
]
,
r
estricte
d
B
o
ltzm
an
n
m
ac
h
in
e
f
o
r
D
o
S
att
ac
k
by
[
3
3
]
,
L
STM
an
d
AE
b
ased
I
DS
p
r
esen
ted
by
[
3
4
]
.
B
ased
on
th
e
g
r
ap
h
tr
en
d
,
t
h
e
p
r
o
p
o
s
ed
an
o
m
aly
-
b
ased
I
DS
o
u
tp
er
f
o
r
m
e
d
e
x
is
tin
g
m
o
d
el
with
s
ig
n
if
ican
t
im
p
r
o
v
em
en
t
in
ac
cu
r
ac
y
s
co
r
e
.
T
h
e
r
ea
s
o
n
is
th
at
th
e
p
r
o
p
o
s
ed
s
tu
d
y
,
em
p
lo
y
in
g
s
u
itab
le
p
r
ep
r
o
ce
s
s
in
g
o
p
er
atio
n
s
an
d
l
ev
er
ag
in
g
d
ee
p
L
STM
ca
p
ab
ilit
ies
wi
th
in
th
e
au
to
en
co
d
er
f
r
am
ewo
r
k
,
is
b
etter
ab
le
to
ca
p
tu
r
e
th
e
u
n
d
e
r
ly
in
g
an
d
co
m
p
lex
p
atter
n
s
of
th
e
i
n
p
u
t
d
ataset
an
d
r
ec
o
n
s
tr
u
cts
n
o
r
m
al
i
n
p
u
t
s
ig
n
al
with
v
er
y
less
r
ec
o
n
s
tr
u
ctio
n
e
r
r
o
r
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
13
,
No
.
4
,
Dec
em
b
er
20
24
:
4
8
9
5
-
4
9
0
5
4904
(
a)
(
b
)
Fig
u
r
e
6.
E
v
alu
atio
n
of
th
e
p
r
o
p
o
s
ed
L
STM
-
AE
with
(
a
)
an
o
m
aly
s
co
r
e
f
o
r
n
etwo
r
k
tr
a
f
f
i
c
an
d
(
b)
co
m
p
ar
ativ
e
a
n
aly
s
is
4.
CO
NCLU
SI
O
N
T
h
is
p
ap
er
h
as
p
r
esen
ted
an
ad
v
an
ce
d
an
d
r
elia
b
le
n
e
two
r
k
ADS
by
in
teg
r
atin
g
L
STM
in
au
to
en
co
d
er
ar
ch
itectu
r
e
f
o
r
b
o
th
u
n
i
-
class
an
d
m
u
lti
-
class
an
o
m
aly
d
etec
tio
n
f
o
r
f
u
tu
r
is
tic
n
etwo
r
k
.
T
h
e
p
r
o
p
o
s
ed
L
STM
-
AE
m
o
d
el
d
y
n
am
ically
ca
p
tu
r
es
tem
p
o
r
al
d
ep
e
n
d
en
cies
in
th
e
d
ata
t
h
r
o
u
g
h
its
L
STM
u
n
it
an
d
ef
f
icien
tly
co
m
p
r
ess
es
an
d
r
ec
o
n
s
tr
u
cts
th
e
in
p
u
t
d
ata,
en
ab
lin
g
t
h
e
d
etec
tio
n
of
b
o
th
k
n
o
wn
an
d
n
o
v
el
attac
k
p
atter
n
s
.
T
h
e
u
tili
za
tio
n
of
a
s
tatis
t
ically
-
d
r
iv
en
f
ea
tu
r
e
s
elec
tio
n
p
r
o
ce
s
s
en
s
u
r
es
th
at
th
e
m
o
d
el
is
tr
ain
ed
on
th
e
m
o
s
t
r
elev
a
n
t
f
ea
tu
r
es,
en
h
an
ci
n
g
its
p
r
ed
icti
v
e
ac
cu
r
ac
y
.
By
ex
clu
s
iv
ely
t
r
ain
in
g
on
n
o
r
m
al
tr
af
f
ic
d
ata,
th
e
L
STM
-
AE
m
o
d
el
av
o
id
s
th
e
co
m
m
o
n
p
itfa
lls
ass
o
ciate
d
wi
th
class
im
b
alan
ce
an
d
u
n
r
ea
lis
tic
d
ata
r
ep
r
esen
tatio
n
f
o
u
n
d
in
m
an
y
ex
is
tin
g
d
atasets
.
Mo
r
eo
v
er
,
th
e
in
tr
o
d
u
ctio
n
of
d
is
tin
ct
th
r
esh
o
ld
s
f
o
r
d
if
f
er
en
t
class
es
of
n
etwo
r
k
in
tr
u
s
io
n
s
r
ep
r
esen
ts
a
k
e
y
a
d
v
a
n
ce
m
en
t
in
th
e
f
ield
of
cy
b
er
s
ec
u
r
ity
.
T
h
e
f
u
tu
r
e
wo
r
k
will
ex
p
lo
r
e
f
u
r
th
er
en
h
an
ce
m
en
ts
of
th
e
p
r
o
p
o
s
ed
s
y
s
tem
u
s
in
g
r
ein
f
o
r
ce
m
en
t
le
ar
n
in
g
tech
n
iq
u
e
to
m
ax
im
ize
th
e
ef
f
ec
tiv
e
n
ess
of
th
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
.
ACK
NO
WL
E
DG
E
M
E
NT
T
h
e
au
t
h
o
r
th
an
k
s
R
E
VA
Un
iv
er
s
ity
B
an
g
alo
r
e
f
o
r
th
e
m
o
r
al
s
u
p
p
o
r
t,
h
i
g
h
lig
h
tin
g
th
at
th
e
r
esear
ch
s
tu
d
y
was
co
n
d
u
cte
d
with
o
u
t
ex
ter
n
al
f
in
a
n
cial
s
u
p
p
o
r
t.
RE
F
E
R
E
NC
E
S
[
1
]
Y
.
A
b
u
a
d
l
l
a
,
G
.
K
v
a
s
c
e
v
,
S
.
G
a
j
i
n
,
a
n
d
Z.
J
o
v
a
n
o
v
i
c
,
“
F
l
o
w
-
b
a
s
e
d
a
n
o
m
a
l
y
i
n
t
r
u
s
i
o
n
d
e
t
e
c
t
i
o
n
s
y
st
e
m
u
s
i
n
g
t
w
o
n
e
u
r
a
l
n
e
t
w
o
r
k
st
a
g
e
s,
”
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
I
n
f
o
r
m
a
t
i
o
n
S
y
s
t
e
m
s
,
v
o
l
.
1
1
,
n
o
.
2
,
p
p
.
6
0
1
–
6
2
2
,
2
0
1
4
,
d
o
i
:
1
0
.
2
2
9
8
/
C
S
I
S
1
3
0
4
1
5
0
3
5
A
.
[
2
]
M
.
Y
.
A
z
a
r
,
V
.
V
a
r
a
d
h
a
r
a
j
a
n
,
L.
H
a
mey
,
a
n
d
U
.
Tu
p
a
k
u
l
a
,
“
A
u
t
o
e
n
c
o
d
e
r
-
b
a
se
d
f
e
a
t
u
r
e
l
e
a
r
n
i
n
g
f
o
r
c
y
b
e
r
se
c
u
r
i
t
y
a
p
p
l
i
c
a
t
i
o
n
s
,
”
i
n
2
0
1
7
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
i
n
t
C
o
n
f
e
re
n
c
e
o
n
N
e
u
r
a
l
N
e
t
w
o
r
k
s (I
J
C
N
N
)
,
2
0
1
7
,
p
p
.
3
8
5
4
–
3
8
6
1
,
d
o
i
:
1
0
.
1
1
0
9
/
I
JC
N
N
.
2
0
1
7
.
7
9
6
6
3
4
2
.
[
3
]
B
.
Z
h
a
n
g
,
Y
.
Y
u
,
a
n
d
J
.
L
i
,
“
N
e
t
w
o
r
k
i
n
t
r
u
s
i
o
n
d
e
t
e
c
t
i
o
n
b
a
s
e
d
o
n
st
a
c
k
e
d
sp
a
r
se
a
u
t
o
e
n
c
o
d
e
r
a
n
d
b
i
n
a
r
y
t
r
e
e
e
n
semb
l
e
m
e
t
h
o
d
,
”
i
n
2
0
1
8
I
EEE
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
C
o
m
m
u
n
i
c
a
t
i
o
n
s
Wo
r
k
s
h
o
p
s
,
M
a
y
2
0
1
8
,
p
p
.
1
–
6
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
C
W
.
2
0
1
8
.
8
4
0
3
7
5
9
.
[
4
]
B
.
A
b
o
l
h
a
s
a
n
z
a
d
e
h
,
“
N
o
n
l
i
n
e
a
r
d
i
me
n
si
o
n
a
l
i
t
y
r
e
d
u
c
t
i
o
n
f
o
r
i
n
t
r
u
si
o
n
d
e
t
e
c
t
i
o
n
u
s
i
n
g
a
u
t
o
-
e
n
c
o
d
e
r
b
o
t
t
l
e
n
e
c
k
f
e
a
t
u
r
e
s
,
”
i
n
2
0
1
5
7
t
h
C
o
n
f
e
re
n
c
e
o
n
I
n
f
o
rm
a
t
i
o
n
a
n
d
K
n
o
w
l
e
d
g
e
T
e
c
h
n
o
l
o
g
y
(
I
K
T
)
,
M
a
y
2
0
1
5
,
p
p
.
1
–
5
,
d
o
i
:
1
0
.
1
1
0
9
/
I
K
T
.
2
0
1
5
.
7
2
8
8
7
9
9
.
[
5
]
S
.
N
.
M
i
g
h
a
n
a
n
d
M
.
K
a
h
a
n
i
,
“
D
e
e
p
l
e
a
r
n
i
n
g
b
a
s
e
d
l
a
t
e
n
t
f
e
a
t
u
r
e
e
x
t
r
a
c
t
i
o
n
f
o
r
i
n
t
r
u
s
i
o
n
d
e
t
e
c
t
i
o
n
,
”
i
n
E
l
e
c
t
r
i
c
a
l
E
n
g
i
n
e
e
r
i
n
g
(
I
C
EE),
I
r
a
n
i
a
n
C
o
n
f
e
r
e
n
c
e
o
n
,
M
a
y
2
0
1
8
,
p
p
.
1
5
1
1
–
1
5
1
6
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
EE.
2
0
1
8
.
8
4
7
2
4
1
8
.
[
6
]
N
.
S
h
o
n
e
,
T
.
N
.
N
g
o
c
,
V
.
D
.
P
h
a
i
,
a
n
d
Q
.
S
h
i
,
“
A
d
e
e
p
l
e
a
r
n
i
n
g
a
p
p
r
o
a
c
h
t
o
n
e
t
w
o
r
k
i
n
t
r
u
s
i
o
n
d
e
t
e
c
t
i
o
n
,
”
I
EEE
T
ra
n
s
a
c
t
i
o
n
s
o
n
Em
e
rg
i
n
g
T
o
p
i
c
s i
n
C
o
m
p
u
t
a
t
i
o
n
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
2
,
n
o
.
1
,
p
p
.
4
1
–
5
0
,
F
e
b
.
2
0
1
8
,
d
o
i
:
1
0
.
1
1
0
9
/
T
ETC
I
.
2
0
1
7
.
2
7
7
2
7
9
2
.
[
7
]
U
.
Ç
e
k
mez
,
Z.
Er
d
e
m,
A
.
G
.
Y
a
v
u
z
,
O
.
K
.
S
a
h
i
n
g
o
z
,
a
n
d
A
.
B
u
l
d
u
,
“
N
e
t
w
o
r
k
a
n
o
mal
y
d
e
t
e
c
t
i
o
n
w
i
t
h
d
e
e
p
l
e
a
r
n
i
n
g
,
”
i
n
2
0
1
8
2
6
t
h
S
i
g
n
a
l
Pr
o
c
e
ssi
n
g
a
n
d
C
o
m
m
u
n
i
c
a
t
i
o
n
s
A
p
p
l
i
c
a
t
i
o
n
s
C
o
n
f
e
re
n
c
e
,
2
0
1
8
,
p
p
.
1
–
4
,
d
o
i
:
1
0
.
1
1
0
9
/
S
I
U
.
2
0
1
8
.
8
4
0
4
8
1
7
.
[
8
]
R
.
C
.
A
y
g
u
n
a
n
d
A
.
G
.
Y
a
v
u
z
,
“
A
s
t
o
c
h
a
s
t
i
c
d
a
t
a
d
i
s
c
r
i
m
i
n
a
t
i
o
n
b
a
s
e
d
a
u
t
o
e
n
c
o
d
e
r
a
p
p
r
o
a
c
h
f
o
r
n
e
t
w
o
r
k
a
n
o
m
a
l
y
d
e
t
e
c
t
i
o
n
,
”
i
n
2
0
1
7
2
5
t
h
S
i
g
n
a
l
Pr
o
c
e
ssi
n
g
a
n
d
C
o
m
m
u
n
i
c
a
t
i
o
n
s
Ap
p
l
i
c
a
t
i
o
n
s
C
o
n
f
e
r
e
n
c
e
,
2
0
1
7
,
p
p
.
1
–
4
,
d
o
i
:
1
0
.
1
1
0
9
/
S
I
U
.
2
0
1
7
.
7
9
6
0
4
1
0
.
[
9
]
S
.
N
a
s
e
e
r
e
t
a
l
.
,
“
E
n
h
a
n
c
e
d
n
e
t
w
o
r
k
a
n
o
ma
l
y
d
e
t
e
c
t
i
o
n
b
a
se
d
o
n
d
e
e
p
n
e
u
r
a
l
n
e
t
w
o
r
k
s,
”
I
EEE
Ac
c
e
ss
,
v
o
l
.
6
,
p
p
.
4
8
2
3
1
–
4
8
2
4
6
,
2
0
1
8
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
1
8
.
2
8
6
3
0
3
6
.
[
1
0
]
Z.
C
h
e
n
,
C
.
K
.
Y
e
o
,
B
.
S
.
Le
e
,
a
n
d
C
.
T
.
L
a
u
,
“
A
u
t
o
e
n
c
o
d
e
r
-
b
a
sed
n
e
t
w
o
r
k
a
n
o
m
a
l
y
d
e
t
e
c
t
i
o
n
,
”
i
n
2
0
1
8
Wi
r
e
l
e
ss
T
e
l
e
c
o
m
m
u
n
i
c
a
t
i
o
n
s
S
y
m
p
o
si
u
m
(
WT
S
)
,
A
p
r
.
2
0
1
8
,
p
p
.
1
–
5
,
d
o
i
:
1
0
.
1
1
0
9
/
W
TS.2
0
1
8
.
8
3
6
3
9
3
0
.
[
1
1
]
“
K
D
D
c
u
p
1
9
9
9
d
a
t
a
,
”
U
C
I
K
D
D
Arc
h
i
v
e
,
1
9
9
9
.
A
c
c
e
sse
d
:
M
a
r
.
3
0
,
2
0
1
9
.
[
O
n
l
i
n
e
]
.
A
v
a
i
l
a
b
l
e
:
h
t
t
p
:
/
/
k
d
d
.
i
c
s.
u
c
i
.
e
d
u
/
d
a
t
a
b
a
s
e
s/
k
d
d
c
u
p
9
9
/
k
d
d
c
u
p
9
9
.
h
t
m
l
[
1
2
]
F
.
F
a
r
a
h
n
a
k
i
a
n
a
n
d
J
.
H
e
i
k
k
o
n
e
n
,
“
A
d
e
e
p
a
u
t
o
-
e
n
c
o
d
e
r
b
a
s
e
d
a
p
p
r
o
a
c
h
f
o
r
i
n
t
r
u
s
i
o
n
d
e
t
e
c
t
i
o
n
s
y
s
t
e
m
,
”
i
n
2
0
1
8
2
0
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
A
d
v
a
n
c
e
d
C
o
m
m
u
n
i
c
a
t
i
o
n
T
e
c
h
n
o
l
o
g
y
(
I
C
A
C
T
)
,
F
e
b
.
2
0
1
8
,
p
p
.
1
7
8
–
1
8
3
,
d
o
i
:
1
0
.
2
3
9
1
9
/
I
C
A
C
T
.
2
0
1
8
.
8
3
2
3
6
8
8
.
[
1
3
]
“
N
S
L
-
K
D
D
d
a
t
a
se
t
,
”
U
n
i
v
e
rs
i
t
y
o
f
N
e
w
Br
u
n
sw
i
c
k
.
A
c
c
e
s
sed
:
Ja
n
.
3
0
,
2
0
1
9
.
[
O
n
l
i
n
e
]
.
A
v
a
i
l
a
b
l
e
:
h
t
t
p
s
:
/
/
w
w
w
.
u
n
b
.
c
a
/
c
i
c
/
d
a
t
a
se
t
s/
n
s
l
.
h
t
ml
.
[
1
4
]
A
.
O
.
A
l
z
a
h
r
a
n
i
a
n
d
M
.
J
.
F
.
A
l
e
n
a
z
i
,
“
D
e
s
i
g
n
i
n
g
a
n
e
t
w
o
r
k
i
n
t
r
u
s
i
o
n
d
e
t
e
c
t
i
o
n
sy
s
t
e
m
b
a
se
d
o
n
ma
c
h
i
n
e
l
e
a
r
n
i
n
g
f
o
r
so
f
t
w
a
r
e
d
e
f
i
n
e
d
n
e
t
w
o
r
k
s,
”
F
u
t
u
r
e
I
n
t
e
r
n
e
t
,
v
o
l
.
1
3
,
n
o
.
5
,
A
p
r
.
2
0
2
1
,
d
o
i
:
1
0
.
3
3
9
0
/
f
i
1
3
0
5
0
1
1
1
.
Evaluation Warning : The document was created with Spire.PDF for Python.