I
AE
S
I
n
t
e
r
n
at
ion
al
Jou
r
n
al
of
Ar
t
if
icial
I
n
t
e
ll
ig
e
n
c
e
(
I
J
-
AI
)
Vol.
14
,
No.
4
,
Augus
t
2025
,
pp.
2945
~
2954
I
S
S
N:
2252
-
8938
,
DO
I
:
10
.
11591/i
jai
.
v
14
.i
4
.
pp
29
45
-
2954
2945
Jou
r
n
al
h
omepage
:
ht
tp:
//
ij
ai
.
iaes
c
or
e
.
c
om
Op
t
imiz
in
g f
ir
e
w
al
l
t
imin
g f
o
r
b
r
u
t
e
f
o
r
c
e
m
itigat
io
n
w
ith
r
an
d
om
f
o
r
e
st
s
Ahm
ad
T
u
r
m
u
d
i
Z
y
1
,
I
s
ar
ian
t
o
1
,
Anggi
M
u
h
am
m
ad
Rif
a'
i
1
,
Ab
d
u
l
Gh
of
ir
2
,
M
u
h
am
m
ad
Naj
am
u
d
d
in
Dw
i
M
ih
ar
j
a
1
,
Anan
t
o
T
r
i
S
as
on
gk
o
1
1
D
e
pa
r
tm
e
nt
of
I
nf
or
ma
ti
c
s
E
ngi
ne
e
r
in
g, F
a
c
ul
ty
of
E
ngi
ne
e
r
in
g,
U
ni
ve
r
s
it
a
s
P
e
li
ta
B
a
ng
s
a
, B
e
ka
s
i,
I
ndone
s
i
a
2
D
e
pa
r
tm
e
nt
of
I
nf
or
ma
ti
c
s
E
ngi
ne
e
r
in
g, F
a
c
ul
ty
of
E
ngi
ne
e
r
in
g, P
r
e
s
id
e
nt
U
ni
ve
r
s
it
y, B
e
ka
s
i,
I
ndone
s
ia
Ar
t
icle
I
n
f
o
AB
S
T
RA
CT
A
r
ti
c
le
h
is
tor
y
:
R
e
c
e
ived
Aug
23
,
2024
R
e
vis
e
d
F
e
b
25
,
2025
Ac
c
e
pted
M
a
r
15
,
2025
Mi
t
i
g
a
t
i
n
g
b
ru
t
e
fo
rce
at
t
ac
k
s
remai
n
s
a
cri
t
i
cal
ch
a
l
l
e
n
g
e
i
n
cy
b
ers
ec
u
ri
t
y
,
req
u
i
ri
n
g
i
n
t
el
l
i
g
en
t
an
d
ad
a
p
t
i
v
e
s
o
l
u
t
i
o
n
s
.
T
h
i
s
res
earch
i
n
t
r
o
d
u
ces
a
n
ap
p
r
o
ach
t
o
o
p
t
i
mi
z
i
n
g
fi
re
w
al
l
d
e
p
l
o
y
me
n
t
t
i
m
i
n
g
fo
r
en
h
a
n
ced
b
ru
t
e
fo
rce
mi
t
i
g
a
t
i
o
n
u
s
i
n
g
p
at
t
ern
reco
g
n
i
t
i
o
n
t
ec
h
n
i
q
u
es
w
i
t
h
t
h
e
ran
d
o
m
fo
re
s
t
al
g
o
ri
t
h
m.
L
ev
erag
i
n
g
t
h
e
U
N
SW
-
N
B1
5
d
a
t
as
et
,
co
mp
re
h
en
s
i
v
e
p
rep
r
o
ces
s
i
n
g
a
n
d
e
x
p
l
o
ra
t
o
r
y
d
at
a
a
n
al
y
s
i
s
(
E
D
A
)
w
ere
p
erf
o
rmed
t
o
en
s
u
re
t
h
e
d
a
t
as
e
t
's
s
u
i
t
ab
i
l
i
t
y
fo
r
mach
i
n
e
l
ea
rn
i
n
g
ap
p
l
i
cat
i
o
n
s
.
T
h
e
s
t
u
d
y
u
t
i
l
i
zed
a
s
t
ru
c
t
u
re
d
w
o
r
k
fl
o
w
,
s
p
l
i
t
t
i
n
g
t
h
e
d
a
t
as
e
t
i
n
t
o
t
rai
n
i
n
g
an
d
t
es
t
i
n
g
s
u
b
s
et
s
t
o
r
i
g
o
ro
u
s
l
y
e
v
al
u
at
e
t
h
e
mo
d
el
's
p
erf
o
rm
an
ce.
T
h
e
p
r
o
p
o
s
e
d
ran
d
o
m
fo
re
s
t
mo
d
el
ac
h
i
e
v
ed
a
h
i
g
h
acc
u
racy
o
f
9
8
.
8
7
%
,
s
u
p
p
o
rt
e
d
b
y
p
reci
s
i
o
n
,
recal
l
,
an
d
F1
-
s
c
o
res
t
h
at
co
n
fi
rm
i
t
s
effect
i
v
en
e
s
s
i
n
d
i
s
t
i
n
g
u
i
s
h
i
n
g
n
o
rmal
an
d
at
t
ack
t
raff
i
c.
T
h
e
co
n
f
u
s
i
o
n
mat
r
i
x
fu
r
t
h
er
v
al
i
d
a
t
ed
t
h
e
m
o
d
e
l
’s
r
o
b
u
s
t
n
e
s
s
,
h
i
g
h
l
i
g
h
t
i
n
g
i
t
s
p
o
t
en
t
i
a
l
i
n
i
mp
r
o
v
i
n
g
t
h
e
effi
ci
e
n
cy
o
f
f
i
rew
a
l
l
d
e
p
l
o
y
m
en
t
.
T
h
e
s
e
fi
n
d
i
n
g
s
d
e
mo
n
s
t
ra
t
e
t
h
e
cr
i
t
i
cal
ro
l
e
o
f
ad
v
a
n
ced
mach
i
n
e
l
earn
i
n
g
t
ech
n
i
q
u
e
s
i
n
en
h
an
ci
n
g
cy
b
er
s
ecu
r
i
t
y
d
efen
s
es
,
p
art
i
cu
l
ar
l
y
i
n
mi
t
i
g
a
t
i
n
g
b
r
u
t
e
fo
rce
at
t
ac
k
s
t
h
ro
u
g
h
o
p
t
i
m
i
zed
,
d
at
a
-
d
ri
v
en
s
t
rat
e
g
i
e
s
.
K
e
y
w
o
r
d
s
:
B
r
ute
f
or
c
e
a
tt
a
c
k
m
it
igation
C
ybe
r
s
e
c
ur
it
y
f
r
a
mew
or
ks
F
ir
e
wa
ll
opti
mi
z
a
ti
on
P
a
tt
e
r
n
r
e
c
ognit
ion
R
a
ndom
f
or
e
s
t
Th
i
s
i
s
a
n
o
p
en
a
c
ces
s
a
r
t
i
c
l
e
u
n
d
e
r
t
h
e
CC
B
Y
-
SA
l
i
ce
n
s
e.
C
or
r
e
s
pon
din
g
A
u
th
or
:
Ahma
d
T
ur
mudi
Z
y
De
pa
r
tm
e
nt
of
I
nf
or
mat
ics
E
nginee
r
ing
,
F
a
c
ult
y
of
E
nginee
r
ing
,
Unive
r
s
it
a
s
P
e
li
ta
B
a
ngs
a
B
e
ka
s
i,
I
ndone
s
ia
E
mail:
tu
r
mudi
@pe
li
taba
ngs
a
.
a
c
.
id
1.
I
NT
RODU
C
T
I
ON
I
n
the
moder
n
digi
tal
e
r
a
,
c
ybe
r
s
e
c
ur
it
y
is
a
p
r
e
s
s
ing
c
onc
e
r
n
f
or
or
ga
niza
ti
ons
due
to
the
g
r
owing
f
r
e
que
nc
y
a
nd
c
ompl
e
xit
y
of
c
ybe
r
a
tt
a
c
ks
[
1
]
.
Or
ga
ni
z
a
ti
ons
mus
t
r
e
main
a
he
a
d
of
de
ve
lopi
ng
r
is
k
s
be
c
a
us
e
c
ybe
r
c
r
im
inals
a
r
e
a
lwa
ys
c
omi
ng
up
with
ne
w
w
a
ys
to
ge
t
a
r
ound
s
e
c
ur
it
y
s
a
f
e
gua
r
ds
.
B
r
ute
f
or
c
e
a
s
s
a
ult
s
ha
ve
gr
own
pa
r
ti
c
ular
ly
pr
e
va
lent
a
mong
the
s
e
ve
r
a
l
types
of
c
ybe
r
a
tt
a
c
ks
[
2
]
.
I
n
or
de
r
t
o
obtain
una
uthor
ize
d
a
c
c
e
s
s
to
s
ys
tems
,
a
tt
a
c
ke
r
s
in
thes
e
a
tt
a
c
ks
methodica
ll
y
tr
y
a
lar
ge
number
o
f
pa
s
s
wor
d
or
e
nc
r
ypti
on
ke
y
c
ombi
na
ti
ons
.
B
r
ute
f
or
c
e
a
tt
a
c
k
s
a
r
e
a
s
igni
f
ica
nt
da
nge
r
due
to
thei
r
s
im
pli
c
it
y
a
nd
the
potential
f
or
s
e
r
ious
outcome
s
[
3]
.
B
r
ute
f
or
c
e
a
tt
a
c
ks
c
a
n
ha
ve
s
e
r
ious
c
on
s
e
que
nc
e
s
.
S
e
ns
it
ive
inf
or
mation
may
be
e
xpos
e
d
a
s
a
r
e
s
ult
of
da
ta
br
e
a
c
he
s
br
ought
on
by
a
t
tac
ke
r
s
onc
e
t
he
y
ge
t
a
c
c
e
s
s
[
4]
.
T
his
c
a
n
s
e
r
ious
ly
ha
r
m
the
tar
ge
ted
or
ga
niza
ti
on's
c
r
e
dibi
li
ty
a
nd
r
e
putation
in
a
ddit
ion
to
c
a
us
ing
f
inanc
ial
los
s
thr
ough
f
r
a
ud
or
t
he
f
t
[
5
]
.
F
ur
ther
mor
e
,
ha
c
ke
d
s
ys
tems
f
r
e
que
ntl
y
ha
ve
their
int
e
gr
i
ty
c
omp
r
omi
s
e
d,
whic
h
incr
e
a
s
e
s
their
s
us
c
e
pti
bil
it
y
to
othe
r
vu
lner
a
bil
it
ies
a
nd
pos
s
ibl
e
dis
r
upti
ons
[
6]
.
S
im
ple
pa
s
s
wor
d
r
e
s
tr
ictions
a
nd
a
c
c
ount
lockout
poli
c
i
e
s
a
r
e
two
e
xa
mpl
e
s
of
typi
c
a
l
s
tr
a
tegie
s
f
or
thwa
r
ti
ng
br
ute
f
or
c
e
a
tt
a
c
ks
;
howe
ve
r
,
they
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14
,
No.
4
,
Augus
t
20
25
:
294
5
-
2954
2946
f
r
e
que
ntl
y
f
a
ll
s
hor
t
.
T
he
s
e
a
ppr
oa
c
he
s
c
a
n't
ke
e
p
up
with
the
c
onti
nuous
ly
c
ha
nging
na
tur
e
of
c
ybe
r
th
r
e
a
ts
a
nd
a
r
e
us
ua
ll
y
r
e
a
c
ti
ve
r
a
the
r
than
p
r
oa
c
ti
ve
[
7
]
.
Give
n
thes
e
c
ha
ll
e
nge
s
,
ther
e
is
a
n
ur
ge
nt
ne
e
d
f
or
mor
e
a
dva
nc
e
d
a
nd
dyna
mi
c
s
olut
ions
to
e
nha
nc
e
the
e
f
f
e
c
ti
ve
ne
s
s
of
br
ute
f
or
c
e
m
it
igation
[
8]
.
Or
ga
niza
ti
ons
mus
t
a
dopt
s
tr
a
tegie
s
that
a
r
e
not
only
c
a
pa
ble
of
de
tec
ti
ng
a
nd
r
e
s
ponding
to
b
r
ute
f
or
c
e
a
tt
a
c
ks
in
r
e
a
l
t
im
e
but
a
ls
o
c
a
pa
ble
of
a
da
pti
ng
to
ne
w
a
nd
e
mer
ging
a
tt
a
c
k
pa
tt
e
r
ns
[
9]
.
T
his
c
a
ll
s
f
or
the
int
e
gr
a
ti
on
of
c
utt
ing
-
e
dge
tec
hnologi
e
s
a
nd
in
nova
ti
ve
a
ppr
oa
c
he
s
to
c
ybe
r
s
e
c
ur
it
y,
e
ns
ur
ing
that
de
f
e
ns
e
s
r
e
main
r
obus
t
a
nd
e
f
f
e
c
ti
ve
in
the
f
a
c
e
of
a
n
e
ve
r
-
c
ha
nging
thr
e
a
t
lands
c
a
pe
[
10]
.
T
o
tac
kle
thes
e
c
ybe
r
s
e
c
ur
it
y
c
ha
ll
e
nge
s
,
a
dva
nc
e
d
mac
hine
lea
r
ning
tec
hniques
ha
ve
pr
ove
n
to
be
a
pr
omi
s
ing
s
olut
ion.
M
a
c
hine
l
e
a
r
ning
e
na
bles
the
a
na
lys
is
of
lar
ge
da
tas
e
t
s
,
a
ll
owing
f
or
the
identif
i
c
a
ti
on
of
pa
tt
e
r
ns
that
c
ould
indi
c
a
te
malicious
a
c
ti
vit
y
[
1
1]
.
Among
the
va
r
ious
mac
hine
lea
r
ning
a
lgo
r
it
hms
,
the
r
a
ndom
f
o
r
e
s
t
a
lgor
i
thm
ha
s
ga
ined
s
igni
f
ica
nt
a
tt
e
nti
on
in
the
r
e
a
lm
of
c
ybe
r
s
e
c
ur
it
y.
T
his
a
lgo
r
it
hm
is
r
e
nowne
d
f
or
it
s
r
obus
tnes
s
a
nd
a
c
c
ur
a
c
y,
making
it
pa
r
ti
c
ular
ly
we
ll
-
s
uit
e
d
f
or
de
tec
ti
ng
a
nd
mi
ti
ga
ti
ng
c
ybe
r
thr
e
a
ts
[
12]
.
I
ts
a
bil
i
ty
to
ha
ndle
lar
ge
da
ta
s
e
ts
with
numer
ous
f
e
a
tur
e
s
a
ll
ows
it
to
a
na
lyze
c
ompl
e
x
da
ta
s
tr
uc
tur
e
s
a
nd
unc
o
ve
r
s
ubtl
e
pa
tt
e
r
ns
that
ma
y
s
igni
f
y
a
b
r
ute
f
or
c
e
a
tt
a
c
k
[
13
]
.
T
he
r
a
ndom
f
or
e
s
t
a
lgor
it
hm
ope
r
a
tes
by
ge
ne
r
a
ti
ng
mul
ti
ple
de
c
is
ion
tr
e
e
s
dur
ing
the
tr
a
ini
ng
pr
oc
e
s
s
a
nd
then
c
ombi
ning
their
output
s
to
f
or
m
a
f
inal
p
r
e
diction
[
14
]
.
E
a
c
h
de
c
is
ion
tr
e
e
is
c
on
s
tr
uc
ted
f
r
om
a
dis
ti
nc
t
s
ubs
e
t
of
the
tr
a
ini
ng
da
ta,
with
th
e
f
inal
c
las
s
if
ica
ti
on
de
ter
mi
ne
d
by
a
major
it
y
vot
e
a
mong
the
tr
e
e
s
.
T
his
e
ns
e
mbl
e
tec
hnique
is
highl
y
e
f
f
e
c
t
ive
in
r
e
duc
ing
ove
r
f
it
ti
ng
,
a
c
om
mon
is
s
ue
whe
r
e
a
model
pe
r
f
or
ms
we
ll
on
tr
a
ini
ng
da
ta
but
s
tr
uggles
with
ne
w,
uns
e
e
n
da
ta
[
15]
.
B
y
a
ve
r
a
ging
the
pr
e
dictio
ns
f
r
om
s
e
ve
r
a
l
tr
e
e
s
,
r
a
ndom
f
o
r
e
s
t
s
im
pr
ove
the
model's
a
bil
it
y
to
ge
ne
r
a
li
z
e
,
making
it
mor
e
r
e
li
a
ble
in
r
e
a
l
-
wor
ld
s
it
ua
ti
ons
whe
r
e
ne
w
a
nd
une
xpe
c
ted
pa
tt
e
r
ns
may
e
mer
ge
[
16
]
.
I
n
a
ddit
ion
t
o
the
inher
e
nt
s
tr
e
ngths
of
the
r
a
ndom
f
or
e
s
t
a
lgor
it
hm
,
the
int
e
g
r
a
ti
on
of
pa
tt
e
r
n
r
e
c
ognit
ion
tec
hniques
f
ur
ther
e
nha
nc
e
s
it
s
e
f
f
e
c
ti
ve
ne
s
s
in
c
ybe
r
s
e
c
ur
it
y
a
ppli
c
a
ti
ons
[
17]
.
P
a
tt
e
r
n
r
e
c
ognit
ion
invo
lves
identif
ying
r
e
gular
i
ti
e
s
a
nd
a
nomalies
in
da
ta,
whic
h
is
c
r
uc
ial
f
o
r
de
tec
ti
ng
c
ompl
e
x
a
tt
a
c
k
pa
tt
e
r
ns
that
tr
a
dit
ional
methods
mi
ght
ove
r
l
ook.
F
o
r
ins
tanc
e
,
br
u
te
f
o
r
c
e
a
tt
a
c
ks
of
ten
e
xhibi
t
s
pe
c
if
ic
be
ha
vior
a
l
pa
tt
e
r
ns
,
s
uc
h
a
s
r
e
pe
a
ted
logi
n
a
tt
e
mp
ts
withi
n
a
s
hor
t
pe
r
iod
.
B
y
leve
r
a
ging
pa
tt
e
r
n
r
e
c
ognit
ion,
mac
hine
lea
r
ning
models
c
a
n
be
tr
a
ined
to
r
e
c
ognize
thes
e
pa
tt
e
r
ns
a
nd
dif
f
e
r
e
nti
a
te
be
twe
e
n
nor
mal
us
e
r
be
ha
vior
a
nd
po
tential
a
tt
a
c
ks
[
18]
.
T
his
c
ombi
na
ti
on
of
r
a
ndom
f
or
e
s
t
s
a
nd
pa
tt
e
r
n
r
e
c
ognit
ion
p
r
ovides
a
powe
r
f
ul
tool
s
e
t
f
o
r
pr
oa
c
ti
ve
ly
identif
ying
a
n
d
m
it
igating
b
r
ute
f
or
c
e
a
tt
a
c
ks
,
ther
e
by
bols
ter
ing
th
e
ove
r
a
ll
s
e
c
ur
it
y
pos
tur
e
of
a
n
or
ga
niza
ti
on
[
19]
.
T
he
int
e
gr
a
ti
on
of
r
a
ndom
f
or
e
s
t
s
a
nd
pa
tt
e
r
n
r
e
c
ognit
ion
tec
hniques
int
o
c
ybe
r
s
e
c
ur
it
y
s
tr
a
tegie
s
pr
e
s
e
nts
a
f
or
mi
da
ble
a
ppr
oa
c
h
to
e
nha
nc
ing
b
r
ut
e
f
or
c
e
a
tt
a
c
k
mi
ti
ga
ti
on
[
20
]
.
R
a
ndom
f
or
e
s
t
s
,
w
it
h
their
e
ns
e
mbl
e
lea
r
ning
c
a
pa
bil
it
ies
,
c
a
n
e
f
f
e
c
ti
ve
ly
m
a
na
ge
a
nd
a
na
lyze
va
s
t
a
mount
s
of
ne
twor
k
t
r
a
f
f
ic
da
ta,
identif
ying
pa
tt
e
r
ns
that
may
indi
c
a
te
a
n
ongoing
br
ute
f
or
c
e
a
tt
a
c
k
[
21]
.
B
y
c
ombi
ning
thes
e
c
a
p
a
bil
it
ies
with
pa
tt
e
r
n
r
e
c
ognit
ion
tec
hniques
,
it
be
c
omes
pos
s
ibl
e
to
de
tec
t
e
ve
n
the
mos
t
s
ubtl
e
a
nd
s
ophis
ti
c
a
ted
a
tt
a
c
k
pa
tt
e
r
ns
.
T
his
s
yne
r
gy
a
ll
ows
f
or
mor
e
a
c
c
ur
a
te
a
n
d
t
im
e
ly
identi
f
ica
ti
on
o
f
b
r
ute
f
or
c
e
a
tt
a
c
k
s
,
whic
h
is
c
r
it
ica
l
f
o
r
de
ployi
ng
c
ounter
mea
s
ur
e
s
pr
ompt
ly
a
nd
e
f
f
e
c
ti
ve
ly
[
22]
.
B
y
leve
r
a
ging
thes
e
a
dva
n
c
e
d
tec
hnologi
e
s
,
c
ybe
r
s
e
c
ur
it
y
s
tr
a
tegie
s
c
a
n
move
be
yond
tr
a
dit
ional
r
e
a
c
ti
ve
a
ppr
o
a
c
he
s
.
I
ns
tea
d
of
r
e
s
ponding
to
a
tt
a
c
ks
a
f
ter
they
ha
ve
oc
c
ur
r
e
d,
or
ga
niza
ti
ons
c
a
n
i
mpl
e
ment
pr
oa
c
ti
ve
mea
s
ur
e
s
that
a
nti
c
ipate
a
nd
pr
e
ve
nt
potential
thr
e
a
ts
[
23]
.
T
he
a
bil
it
y
to
a
na
lyze
a
nd
identif
y
a
tt
a
c
k
pa
tt
e
r
ns
in
r
e
a
l
-
ti
me
s
igni
f
ica
ntl
y
e
nha
nc
e
s
the
s
pe
e
d
a
nd
a
c
c
ur
a
c
y
of
the
r
e
s
pons
e
,
r
e
d
uc
ing
the
window
of
oppor
tuni
ty
f
o
r
a
tt
a
c
ke
r
s
[
24]
.
T
his
r
e
s
e
a
r
c
h
s
pe
c
if
ica
ll
y
f
oc
us
e
s
on
opti
mi
z
ing
the
ti
mi
ng
of
f
ir
e
wa
ll
de
ploym
e
nts
us
ing
r
a
ndom
f
or
e
s
t
s
a
nd
pa
tt
e
r
n
r
e
c
ognit
ion
.
B
y
de
ter
mi
ning
the
mos
t
e
f
f
e
c
ti
ve
ti
mes
to
a
c
ti
va
te
f
ir
e
wa
ll
s
,
it
is
pos
s
ibl
e
to
p
r
e
e
mpt
ivel
y
block
malicious
tr
a
f
f
ic,
ther
e
by
mi
nim
izing
the
r
is
k
o
f
s
uc
c
e
s
s
f
ul
br
ute
f
or
c
e
a
tt
a
c
ks
[
25]
.
T
he
UN
S
W
-
NB
15
da
tas
e
t
f
r
om
[
26]
,
whic
h
c
onta
ins
c
ompr
e
he
ns
ive
ne
twor
k
t
r
a
f
f
ic
da
ta,
s
e
r
ve
s
a
s
the
f
ounda
ti
on
f
or
tr
a
ini
ng
a
nd
e
va
luating
the
r
a
ndom
f
or
e
s
t
model.
T
his
da
tas
e
t
is
pa
r
ti
c
ular
ly
va
luable
be
c
a
us
e
it
include
s
a
wide
va
r
iety
of
nor
mal
a
nd
m
a
li
c
ious
tr
a
f
f
ic
pa
tt
e
r
ns
,
p
r
ovidi
ng
a
r
obus
t
t
r
a
ini
ng
gr
ound
f
or
the
model
.
T
r
a
ini
ng
the
r
a
ndom
f
or
e
s
t
model
o
n
thi
s
da
tas
e
t
e
na
bles
it
to
e
f
f
e
c
ti
ve
ly
di
f
f
e
r
e
nti
a
te
be
twe
e
n
be
nign
a
nd
malicious
tr
a
f
f
ic
with
a
high
de
gr
e
e
of
a
c
c
ur
a
c
y.
T
he
s
uc
c
e
s
s
f
ul
a
ppli
c
a
ti
on
of
thi
s
te
c
hnology
c
a
n
s
igni
f
ica
ntl
y
im
pr
ove
de
f
e
ns
e
mec
ha
nis
m
s
a
g
a
ins
t
br
ute
f
or
c
e
a
tt
a
c
ks
,
pr
ovidi
ng
a
pr
oa
c
ti
ve
r
a
t
he
r
than
r
e
a
c
ti
ve
a
ppr
oa
c
h
to
c
ybe
r
s
e
c
ur
it
y
[
27]
.
T
he
pur
pos
e
of
thi
s
r
e
s
e
a
r
c
h
is
to
int
e
g
r
a
te
f
i
ndings
int
o
pr
a
c
ti
c
a
l
c
ybe
r
s
e
c
ur
it
y
f
r
a
mew
or
ks
,
e
nha
nc
ing
the
e
f
f
icie
nc
y
a
nd
e
f
f
e
c
ti
ve
ne
s
s
of
ne
twor
k
de
f
e
ns
e
mec
ha
nis
ms
us
ing
the
r
a
ndom
f
or
e
s
t
a
lgor
it
hm.
B
r
ute
f
or
c
e
a
tt
a
c
ks
,
a
s
pe
r
s
is
tent
a
n
d
e
volvi
ng
thr
e
a
ts
,
de
mand
pr
oa
c
ti
ve
s
olut
ions
be
yond
tr
a
dit
ional
r
e
a
c
ti
ve
methods
[
28]
.
T
his
s
tudy
de
mons
tr
a
tes
how
a
dva
nc
e
d
mac
hine
lea
r
ning
tec
hniques
,
pa
r
ti
c
ular
ly
r
a
ndom
f
o
r
e
s
t
,
c
a
n
a
ddr
e
s
s
c
r
it
ica
l
c
ybe
r
s
e
c
ur
it
y
c
ha
ll
e
nge
s
by
br
idgi
ng
the
ga
p
be
twe
e
n
theor
e
ti
c
a
l
models
a
nd
pr
a
c
ti
c
a
l
a
ppli
c
a
ti
ons
[
29]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
Optimiz
ing
fi
r
e
w
all
ti
ming
for
br
ute
for
c
e
miti
gati
on
w
it
h
r
andom
for
e
s
ts
(
A
hmad
T
ur
mudi
Z
y
)
2947
T
his
r
e
s
e
a
r
c
h
c
ontr
ibut
e
s
by
int
r
oduc
ing
a
nove
l
methodology
f
or
opt
im
izing
f
i
r
e
wa
ll
de
ploym
e
nt
ti
mi
ng
th
r
ough
pa
tt
e
r
n
r
e
c
ognit
ion
a
nd
r
a
ndom
f
o
r
e
s
t
s
,
e
na
b
li
ng
p
r
oa
c
ti
ve
de
tec
ti
on
a
nd
mi
ti
ga
ti
on
of
b
r
ute
f
or
c
e
a
tt
a
c
ks
.
I
t
a
ls
o
va
li
da
tes
the
int
e
gr
a
ti
on
o
f
m
a
c
hine
lea
r
ning
with
dyna
mi
c
f
i
r
e
wa
ll
s
tr
a
tegie
s
to
im
pr
ove
r
e
a
l
-
ti
me
r
e
s
pons
e
c
a
pa
bil
it
ies
a
ga
in
s
t
c
ybe
r
thr
e
a
ts
.
Additi
ona
ll
y,
it
e
mphas
ize
s
the
s
igni
f
i
c
a
nc
e
of
leve
r
a
ging
high
-
qua
li
ty
da
tas
e
ts
,
s
uc
h
a
s
UN
S
W
-
NB
15,
to
a
c
c
ur
a
tely
identif
y
a
tt
a
c
k
pa
tt
e
r
n
s
while
mi
nim
izing
f
a
ls
e
pos
it
ives
a
nd
ne
ga
ti
ve
s
.
Dr
iven
by
the
incr
e
a
s
ing
s
ophis
ti
c
a
ti
on
of
br
ute
f
or
c
e
a
tt
a
c
ks
,
thi
s
s
tudy
of
f
e
r
s
a
s
c
a
lable
a
nd
a
da
ptable
s
olut
ion
that
int
e
gr
a
tes
s
e
a
ml
e
s
s
ly
int
o
e
xis
ti
ng
s
e
c
ur
it
y
f
r
a
mew
or
ks
.
Optim
izing
f
ir
e
wa
ll
de
ploym
e
nt
ti
mi
ng
e
ns
ur
e
s
e
f
f
icie
nt
r
e
s
our
c
e
uti
li
z
a
ti
on,
a
voids
dis
r
upti
ons
to
legiti
mate
tr
a
f
f
ic,
a
nd
pr
e
e
mpt
ively
blocks
malicious
a
c
ti
vit
ies
.
T
his
a
ppr
oa
c
h
s
tr
e
n
gthens
im
media
te
de
f
e
ns
e
s
a
nd
pr
ovides
a
f
ounda
ti
on
f
o
r
f
utur
e
a
dva
nc
e
ments
in
int
e
ll
igent
a
nd
a
utom
a
ted
c
ybe
r
s
e
c
ur
it
y
s
olut
ions
.
B
y
a
ddr
e
s
s
ing
thes
e
objec
ti
ve
s
,
thi
s
r
e
s
e
a
r
c
h
c
ontr
ibu
tes
to
the
de
ve
lo
pment
of
a
da
pti
ve
a
nd
r
e
s
il
ient
c
ybe
r
s
e
c
ur
it
y
tec
hnolog
ies
,
e
quippi
ng
or
ga
niza
ti
ons
with
e
f
f
e
c
ti
ve
tool
s
to
c
ounter
e
mer
ging
thr
e
a
ts
a
nd
p
r
otec
t
c
r
it
ica
l
ne
twor
k
inf
r
a
s
tr
uc
tur
e
s
.
2.
RE
L
AT
E
D
WORKS
T
he
e
volut
ion
o
f
r
e
s
e
a
r
c
h
in
opt
im
izing
f
ir
e
wa
ll
de
ploym
e
nt
ti
m
ing
f
or
e
nha
nc
e
d
b
r
ute
f
or
c
e
mi
ti
ga
ti
on
ha
s
ga
ined
s
igni
f
ica
nt
tr
a
c
ti
on
in
r
e
c
e
nt
ye
a
r
s
.
T
r
a
dit
ional
a
ppr
oa
c
he
s
to
m
it
igating
b
r
ute
f
or
c
e
a
tt
a
c
ks
pr
im
a
r
il
y
r
e
li
e
d
on
s
tatic
r
u
les
a
nd
s
ign
a
tur
e
-
ba
s
e
d
de
tec
ti
on
methods
,
whic
h
of
ten
f
a
ll
s
hor
t
in
a
da
pti
ng
to
the
dyna
mi
c
na
tur
e
o
f
c
ybe
r
th
r
e
a
ts
.
R
e
c
e
nt
s
tudi
e
s
ha
ve
highl
ight
e
d
the
li
mi
tations
of
thes
e
c
onve
nti
ona
l
methods
a
nd
the
ne
e
d
f
or
mo
r
e
a
da
pti
ve
a
nd
int
e
ll
igent
s
olut
ions
.
F
o
r
ins
tanc
e
,
dyna
mi
c
r
ule
a
djus
tm
e
nt
a
nd
r
e
a
l
-
ti
me
tr
a
f
f
ic
a
na
lys
is
ha
ve
be
e
n
pr
opos
e
d
a
s
mor
e
e
f
f
e
c
ti
ve
s
tr
a
tegie
s
f
or
de
te
c
t
ing
a
nd
r
e
s
ponding
to
br
ute
f
or
c
e
a
tt
a
c
ks
[
30]
,
[
31]
.
T
he
s
e
a
dva
nc
e
ments
unde
r
s
c
or
e
the
ne
c
e
s
s
it
y
of
opt
im
izing
f
ir
e
wa
ll
de
ploym
e
nt
ti
mi
ng
to
e
nha
nc
e
the
r
e
s
pons
ivene
s
s
a
nd
e
f
f
icie
nc
y
of
mi
t
igation
e
f
f
or
ts
.
T
he
int
e
gr
a
ti
on
o
f
mac
hine
lea
r
ning
int
o
c
ybe
r
s
e
c
ur
it
y
f
r
a
mew
or
ks
ha
s
r
e
volut
ioni
z
e
d
the
a
ppr
oa
c
h
to
thr
e
a
t
de
tec
ti
on
a
nd
mi
ti
ga
ti
on
[
18]
.
M
a
c
hine
le
a
r
ning
a
lgor
it
hms
,
pa
r
ti
c
ular
ly
s
upe
r
vis
e
d
lea
r
ning
models
,
ha
ve
be
e
n
e
xtens
iv
e
ly
us
e
d
to
a
na
lyz
e
ne
twor
k
tr
a
f
f
ic
a
nd
identif
y
malicious
pa
tt
e
r
ns
.
S
tudi
e
s
ha
ve
de
mons
tr
a
ted
that
mac
hine
lea
r
ning
c
a
n
s
igni
f
ica
ntl
y
im
p
r
ove
the
a
c
c
ur
a
c
y
a
nd
s
pe
e
d
of
de
tec
ti
ng
va
r
ious
types
of
c
ybe
r
a
tt
a
c
ks
,
including
br
ute
f
or
c
e
a
tt
a
c
ks
[
32]
.
T
he
a
bil
it
y
of
mac
hine
lea
r
ning
models
to
lea
r
n
f
r
om
his
tor
ica
l
da
ta
a
nd
r
e
c
ognize
c
ompl
e
x
a
t
tac
k
pa
tt
e
r
ns
make
s
them
invalua
ble
in
the
r
e
a
lm
of
c
ybe
r
s
e
c
ur
it
y.
R
e
s
e
a
r
c
h
e
f
f
or
ts
ha
ve
f
oc
us
e
d
on
de
ve
lopi
ng
models
that
c
a
n
a
da
pt
to
e
volvi
ng
thr
e
a
ts
,
ther
e
by
e
nha
nc
ing
the
ove
r
a
ll
r
obus
tnes
s
of
ne
twor
k
de
f
e
ns
e
mec
ha
nis
ms
.
Among
va
r
ious
mac
hine
lea
r
ning
a
lgor
i
thm
s
,
t
he
r
a
ndom
f
or
e
s
t
a
lgo
r
it
hm
s
tands
out
f
o
r
it
s
r
obus
tnes
s
a
nd
e
f
f
icie
nc
y
in
ha
ndli
ng
lar
ge
da
t
a
s
e
ts
with
numer
ous
f
e
a
tur
e
s
.
As
a
n
e
ns
e
mbl
e
lea
r
ning
method,
r
a
ndom
f
o
r
e
s
t
buil
ds
mul
ti
ple
de
c
is
ion
tr
e
e
s
dur
ing
the
tr
a
ini
ng
pha
s
e
a
nd
a
ggr
e
ga
tes
their
output
s
to
e
nha
nc
e
c
las
s
i
f
ica
ti
on
a
c
c
ur
a
c
y
a
nd
mi
ti
ga
te
ove
r
f
it
ti
ng
[
33
]
.
T
his
method
ha
s
s
hown
pa
r
ti
c
ular
e
f
f
e
c
ti
ve
ne
s
s
in
c
ybe
r
s
e
c
ur
it
y,
whe
r
e
ge
ne
r
a
li
z
ing
f
r
om
diver
s
e
,
high
-
dim
e
ns
ional
da
ta
is
e
s
s
e
nti
a
l.
S
tudi
e
s
on
r
a
ndom
f
or
e
s
t
ha
ve
highl
ight
e
d
it
s
s
upe
r
ior
pe
r
f
o
r
manc
e
i
n
de
tec
ti
ng
a
nd
mi
ti
ga
ti
ng
br
ute
f
o
r
c
e
a
tt
a
c
ks
,
ma
king
it
a
f
a
vor
e
d
opti
on
in
thi
s
f
ield
[
12]
,
[
34
]
.
S
e
ve
r
a
l
s
tudi
e
s
ha
ve
e
xplor
e
d
s
im
il
a
r
theme
s
in
op
ti
mi
z
ing
f
i
r
e
wa
ll
de
ploym
e
nt
a
nd
e
nha
nc
ing
b
r
ute
f
or
c
e
a
tt
a
c
k
mi
ti
ga
ti
on
us
ing
a
dva
nc
e
d
tec
hniques
.
F
or
ins
tanc
e
,
r
e
s
e
a
r
c
h
on
a
da
pti
ve
f
i
r
e
wa
ll
poli
c
ies
that
leve
r
a
ge
r
e
a
l
-
ti
me
d
a
ta
a
na
lys
is
a
nd
mac
hine
lea
r
ning
f
or
dyna
mi
c
r
ule
a
djus
tm
e
nts
ha
s
s
hown
p
r
omi
s
ing
r
e
s
ult
s
[
35]
.
Add
it
ionally
,
s
tudi
e
s
e
mpl
oying
va
r
i
ous
e
ns
e
mbl
e
lea
r
ning
methods
,
including
r
a
ndo
m
f
or
e
s
t
,
ha
ve
highl
ight
e
d
their
e
f
f
e
c
ti
ve
ne
s
s
in
im
p
r
oving
de
tec
ti
on
a
c
c
ur
a
c
y
a
nd
r
e
s
pons
e
ti
mes
[
36]
.
T
he
s
e
wor
ks
c
oll
e
c
ti
ve
ly
e
mphas
ize
the
im
por
tanc
e
of
c
onti
nuo
us
innovation
a
nd
the
int
e
g
r
a
ti
on
o
f
in
telli
ge
nt
a
lg
or
it
hms
in
de
ve
lopi
ng
mo
r
e
r
e
s
il
ient
a
nd
pr
oa
c
ti
ve
c
ybe
r
s
e
c
ur
it
y
s
tr
a
tegie
s
.
T
he
f
indi
ngs
f
r
o
m
thes
e
s
tudi
e
s
p
r
ovide
a
s
oli
d
f
ounda
ti
on
f
or
f
ur
ther
r
e
s
e
a
r
c
h
a
nd
de
ve
lopm
e
nt
in
opti
mi
z
ing
f
i
r
e
wa
ll
de
ploym
e
nt
ti
mi
ng
f
o
r
e
nha
nc
e
d
br
ute
f
or
c
e
mi
ti
ga
ti
on
.
B
a
s
e
d
on
the
r
e
view
e
d
li
ter
a
tur
e
,
it
is
c
lea
r
that
opti
mi
z
ing
f
ir
e
wa
ll
de
ploym
e
nt
ti
mi
ng
plays
a
c
r
uc
ial
r
ole
in
im
pr
oving
b
r
ute
f
or
c
e
mi
ti
ga
t
ion.
T
r
a
dit
ional
methods
r
e
lyi
ng
on
s
tatic
r
ules
a
nd
s
ignatur
e
-
ba
s
e
d
de
tec
ti
on
is
ins
uf
f
icie
nt
in
a
ddr
e
s
s
ing
the
dyna
mi
c
na
tur
e
of
c
ybe
r
thr
e
a
ts
.
M
a
c
hine
l
e
a
r
ning,
pa
r
ti
c
ular
ly
the
r
a
ndom
f
o
r
e
s
t
a
lgor
it
hm
,
ha
s
de
mons
tr
a
ted
it
s
e
f
f
e
c
ti
ve
ne
s
s
in
a
na
lyzing
high
-
dim
e
ns
ional
da
ta
a
nd
identi
f
ying
c
ompl
e
x
a
tt
a
c
k
pa
tt
e
r
ns
,
of
f
e
r
ing
s
igni
f
ica
nt
im
pr
ove
ments
in
de
tec
ti
on
a
c
c
u
r
a
c
y
a
nd
a
da
ptabili
ty.
Th
is
s
tudy
a
im
s
to
opti
mi
z
e
f
ir
e
wa
l
l
de
ploym
e
nt
ti
mi
ng
f
o
r
e
nha
nc
e
d
b
r
ute
f
or
c
e
m
it
igation
us
ing
pa
tt
e
r
n
r
e
c
ognit
ion
with
the
r
a
ndom
f
or
e
s
t
a
lgor
it
hm.
B
y
leve
r
a
ging
it
s
a
bil
it
y
to
pr
oc
e
s
s
diver
s
e
da
ta
a
nd
ge
ne
r
a
li
z
e
e
f
f
e
c
ti
ve
ly
,
thi
s
r
e
s
e
a
r
c
h
s
e
e
ks
to
p
r
ovid
e
a
mor
e
a
c
c
ur
a
te,
e
f
f
icie
nt,
a
nd
a
da
pti
ve
s
ol
uti
on
to
c
ounter
e
volvi
ng
c
ybe
r
thr
e
a
ts
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14
,
No.
4
,
Augus
t
20
25
:
294
5
-
2954
2948
3.
M
E
T
HO
D
T
he
meth
odol
ogy
f
o
r
opt
im
iz
ing
f
i
r
e
wa
ll
de
pl
oym
e
nt
t
im
i
ng
to
im
p
r
ove
b
r
ute
f
o
r
c
e
a
tt
a
c
k
m
it
i
ga
ti
on
invol
ve
s
f
i
ve
c
r
uc
ial
s
t
a
ge
s
,
a
s
de
picte
d
in
F
ig
ur
e
1
.
F
ir
s
t
,
r
e
lev
a
nt
da
ta
f
r
o
m
the
UN
S
W
-
NB
15
da
tas
e
t
is
s
ys
tema
ti
c
a
ll
y
c
ol
lec
te
d
to
e
ns
u
r
e
a
c
o
mp
r
e
he
ns
i
ve
r
e
p
r
e
s
e
nta
ti
o
n
o
f
bo
th
n
or
mal
a
nd
a
tt
a
c
k
tr
a
f
f
ic
.
T
his
da
ta
unde
r
goe
s
pr
e
pr
oc
e
s
s
i
ng
to
e
ns
u
r
e
qua
l
it
y
a
nd
c
ons
is
t
e
nc
y
,
a
dd
r
e
s
s
ing
mi
s
s
i
ng
va
lues
a
nd
e
nc
odin
g
ca
tego
r
ica
l
f
e
a
t
ur
e
s
.
Ne
xt
,
a
p
pr
o
pr
iate
mode
ls
,
pa
r
ti
c
ula
r
ly
the
r
a
ndo
m
f
or
e
s
t
a
l
gor
it
h
m,
a
r
e
s
e
lec
ted
f
o
r
tr
a
ini
ng
a
nd
tes
ti
ng
to
dis
ti
ng
uis
h
be
twe
e
n
n
or
mal
a
nd
a
tt
a
c
k
tr
a
f
f
ic.
An
a
bla
ti
o
n
s
tu
dy
f
o
ll
ows
to
a
s
s
e
s
s
the
i
mpac
t
of
dif
f
e
r
e
nt
c
omp
one
n
ts
on
the
m
o
de
l's
pe
r
f
o
r
ma
nc
e
,
e
s
pe
c
iall
y
t
he
c
o
nt
r
ib
uti
o
n
o
f
pa
tt
e
r
n
r
e
c
ogn
it
i
on
te
c
hniques
to
de
tec
ti
on
a
c
c
ur
a
c
y.
F
inal
ly
,
t
he
r
e
s
ult
s
a
r
e
tho
r
o
ughly
a
na
l
yz
e
d
to
e
va
lua
te
t
he
p
r
op
os
e
d
metho
d's
e
f
f
icie
nc
y
a
nd
e
f
f
e
c
t
ivene
s
s
i
n
op
ti
m
izin
g
f
ir
e
w
a
ll
de
plo
yme
nt
ti
mi
n
g
to
mi
ti
ga
te
b
r
ute
f
o
r
c
e
a
t
tac
ks
.
F
igu
r
e
1
v
is
ua
l
ly
r
e
p
r
e
s
e
nts
th
is
me
tho
dolo
gy
,
il
lus
t
r
a
ti
ng
the
p
r
og
r
e
s
s
ion
thr
ough
thes
e
f
i
ve
s
tage
s
.
F
igur
e
1.
M
e
thodol
ogy
e
nha
nc
e
d
br
ute
f
or
c
e
m
it
ig
a
ti
on
p
ur
pos
e
3.
1.
Dat
a
c
oll
e
c
t
ion
T
he
ini
ti
a
l
pha
s
e
invol
ve
s
s
ys
tema
ti
c
a
ll
y
c
oll
e
c
ti
ng
r
e
leva
nt
da
ta
f
r
om
the
UN
S
W
-
NB
15
da
tas
e
t,
a
c
ompr
e
he
ns
ive
r
e
s
our
c
e
de
s
igned
f
or
c
ybe
r
s
e
c
ur
it
y
r
e
s
e
a
r
c
h,
pa
r
ti
c
ular
ly
in
int
r
us
ion
de
tec
ti
on
a
nd
pr
e
ve
nti
on
s
ys
tems
[
26]
.
T
he
da
tas
e
t
e
nc
ompas
s
e
s
a
diver
s
e
s
e
t
of
f
e
a
tur
e
s
that
c
ha
r
a
c
ter
ize
ne
twor
k
tr
a
f
f
ic,
including
c
onne
c
ti
on
dur
a
ti
on,
the
pr
otocol
e
mpl
oye
d,
the
s
tate
of
the
c
onne
c
ti
on,
the
number
of
pa
c
ke
ts
tr
a
ns
mi
tt
e
d
a
nd
r
e
c
e
ived,
a
nd
the
tot
a
l
bytes
e
xc
h
a
nge
d
s
how
in
T
a
ble
1.
Additi
ona
ll
y,
it
c
ontains
c
omput
e
d
f
e
a
tur
e
s
s
uc
h
a
s
the
s
our
c
e
-
to
-
de
s
ti
na
ti
on
ti
me
-
to
-
li
ve
(
T
T
L
)
va
lue,
the
bi
t
r
a
te
be
twe
e
n
s
our
c
e
a
nd
de
s
ti
na
ti
on,
int
e
r
-
pa
c
ke
t
a
r
r
ival
t
im
e
s
,
a
nd
j
it
ter
.
T
he
da
tas
e
t
c
ons
is
ts
of
both
numer
ic
a
nd
c
a
tegor
i
c
a
l
f
e
a
tur
e
s
.
Nume
r
ic
f
e
a
tur
e
s
include
c
onne
c
ti
on
dur
a
ti
on,
pa
c
ke
t
c
ounts
,
byte
c
ounts
,
a
nd
da
ta
t
r
a
ns
f
e
r
r
a
tes
,
whi
le
c
a
tegor
ica
l
f
e
a
tur
e
s
e
nc
ompas
s
pr
otocol
type,
s
e
r
vice
type,
a
nd
c
onne
c
ti
on
s
tate
.
T
he
tar
ge
t
va
r
iable
,
labe
l
,
ind
ica
tes
whe
ther
the
tr
a
f
f
ic
is
no
r
mal
(
0
)
or
a
n
a
tt
a
c
k
(
1)
,
with
a
n
a
d
dit
ional
a
tt
a
c
k_c
a
t
c
olu
mn
s
pe
c
if
ying
the
a
tt
a
c
k
c
a
tegor
y
.
T
o
p
r
e
pa
r
e
the
da
tas
e
t
f
o
r
mac
hine
lea
r
ning
,
c
a
tegor
ica
l
f
e
a
tur
e
s
a
r
e
c
onve
r
ted
int
o
numer
ic
f
or
mat
us
ing
tec
hniques
s
uc
h
a
s
labe
l
e
nc
oding.
Any
mi
s
s
ing
va
lues
a
r
e
a
ddr
e
s
s
e
d
by
e
it
he
r
f
il
l
ing
th
e
m
in
or
r
e
movi
ng
the
a
f
f
e
c
ted
r
ows
.
T
he
da
tas
e
t
is
then
di
vided
int
o
f
e
a
tur
e
s
(
X
)
a
nd
the
tar
ge
t
va
r
iable
(
y
)
,
whe
r
e
X
include
s
a
ll
c
olum
ns
e
xc
e
pt
labe
l
a
nd
a
tt
a
c
k_c
a
t
,
a
nd
y
r
e
pr
e
s
e
nts
the
labe
l
.
T
he
f
e
a
tur
e
s
a
r
e
s
tanda
r
dize
d
to
br
ing
them
to
a
c
ompar
a
ble
s
c
a
le,
wi
th
a
mea
n
of
z
e
r
o
a
nd
a
s
tanda
r
d
de
viation
of
one
.
T
his
p
r
e
pr
oc
e
s
s
ing
s
tep
e
ns
ur
e
s
the
da
ta
is
we
ll
-
pr
e
pa
r
e
d
f
o
r
tr
a
ini
ng
mac
hine
lea
r
ning
models
,
f
a
c
il
it
a
ti
ng
p
r
e
c
is
e
a
nd
e
f
f
icie
nt
de
tec
ti
on
a
nd
mi
ti
ga
ti
on
of
b
r
ute
f
o
r
c
e
a
tt
a
c
ks
.
3.
2.
Dat
a
p
r
e
p
r
oc
e
s
s
in
g
Da
ta
pr
e
pr
oc
e
s
s
ing
is
a
c
r
it
ica
l
s
tep
in
the
m
a
c
hine
lea
r
ning
p
ipeline
to
e
ns
ur
e
the
qua
li
ty
,
c
ons
is
tenc
y,
a
nd
r
e
a
dines
s
of
the
da
tas
e
t
f
or
tr
a
ini
ng
the
model.
I
t
p
lays
a
vit
a
l
r
ole
in
im
pr
o
ving
the
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
Optimiz
ing
fi
r
e
w
all
ti
ming
for
br
ute
for
c
e
miti
gati
on
w
it
h
r
andom
for
e
s
ts
(
A
hmad
T
ur
mudi
Z
y
)
2949
pe
r
f
or
manc
e
a
nd
a
c
c
ur
a
c
y
o
f
mac
hine
lea
r
ning
a
lgor
it
hms
by
e
li
m
inating
nois
e
a
nd
ha
ndli
ng
m
is
s
ing
or
incons
is
tent
va
lue
s
.
T
his
s
tep
invol
ve
s
s
e
ve
r
a
l
s
ub
-
pr
oc
e
s
s
e
s
,
c
he
c
king
da
ta
type
s
,
e
nc
oding
c
a
t
e
gor
ica
l
f
e
a
tur
e
s
,
h
a
ndli
ng
mi
s
s
ing
va
lues
,
s
tanda
r
dizing
f
e
a
tur
e
va
r
iable
s
,
a
nd
s
pli
tt
ing
da
ta
,
whic
h
a
r
e
e
s
s
e
nti
a
l
to
pr
e
pa
r
e
the
da
tas
e
t
e
f
f
e
c
ti
ve
ly
be
f
or
e
model
tr
a
ini
n
g.
‒
C
he
c
king
da
ta
types
:
th
e
ini
ti
a
l
pha
s
e
of
p
r
e
pr
oc
e
s
s
ing
invol
ve
s
ins
pe
c
ti
ng
the
da
ta
types
of
e
a
c
h
c
ol
umn
in
the
da
tas
e
t.
T
h
is
s
tep
is
c
r
uc
ial
to
identif
y
wh
ich
f
e
a
tur
e
s
a
r
e
c
a
tegor
ica
l
a
nd
whic
h
a
r
e
numer
ica
l.
C
a
tegor
ica
l
f
e
a
tur
e
s
ne
e
d
to
be
e
nc
ode
d
int
o
a
numer
ic
f
o
r
mat,
while
numer
ica
l
f
e
a
tu
r
e
s
ne
e
d
t
o
be
s
tanda
r
dize
d
to
e
ns
ur
e
unif
or
m
s
c
a
li
ng.
‒
E
nc
oding
c
a
tegor
ica
l
f
e
a
tur
e
s
:
in
the
da
tas
e
t,
c
a
tegor
ica
l
f
e
a
tur
e
s
s
uc
h
a
s
'p
r
oto'
(
pr
otocol
type)
,
's
e
r
vice
'
(
ne
twor
k
s
e
r
vice
on
the
de
s
ti
na
ti
on)
,
a
nd
's
tate
'
(
s
tate
a
nd
c
ondit
ion
of
the
pr
otocol)
we
r
e
identi
f
ied.
T
he
s
e
f
e
a
tur
e
s
c
a
nnot
be
dir
e
c
tl
y
us
e
d
by
mos
t
m
a
c
hine
lea
r
ning
a
lgor
it
hms
that
r
e
qui
r
e
numer
ic
i
nput.
T
he
r
e
f
or
e
,
they
we
r
e
e
nc
ode
d
us
ing
the
L
a
be
lE
nc
ode
r
.
L
a
be
l
e
nc
oding
c
onve
r
ts
c
a
tegor
ica
l
va
lues
int
o
a
numer
ic
f
o
r
mat
whe
r
e
e
a
c
h
unique
c
a
tegor
y
is
a
s
s
igned
a
dis
ti
nc
t
int
e
ge
r
va
lue.
T
his
tr
a
ns
f
or
m
a
ti
on
e
na
bles
the
r
a
ndom
f
or
e
s
t
model
to
pr
oc
e
s
s
a
nd
lea
r
n
f
r
om
thes
e
f
e
a
tur
e
s
e
f
f
e
c
ti
ve
ly.
‒
Ha
ndli
ng
mi
s
s
ing
va
lues
:
the
da
tas
e
t
wa
s
then
c
h
e
c
ke
d
f
or
mi
s
s
ing
va
lues
,
whic
h
c
a
n
ne
ga
ti
ve
ly
i
mpac
t
the
pe
r
f
or
manc
e
of
the
mac
hine
lea
r
ning
model
if
not
ha
ndled
pr
ope
r
ly.
M
is
s
ing
va
lues
c
a
n
a
r
is
e
d
ue
to
va
r
ious
r
e
a
s
ons
,
s
uc
h
a
s
incomplete
da
ta
c
oll
e
c
ti
on
or
da
ta
c
or
r
upti
on
.
I
n
th
is
r
e
s
e
a
r
c
h,
mi
s
s
ing
v
a
lues
we
r
e
a
ddr
e
s
s
e
d
by
e
it
he
r
dr
opping
r
ows
with
mi
s
s
ing
da
ta
or
f
il
li
ng
them
us
ing
a
pp
r
opr
iate
methods
li
ke
f
or
wa
r
d
f
il
l
or
mea
n
im
putation
.
I
n
thi
s
ins
tanc
e
,
r
ows
with
mi
s
s
ing
va
lues
we
r
e
dr
oppe
d
to
maintai
n
the
da
tas
e
t's
int
e
gr
it
y
a
nd
e
ns
ur
e
that
the
model
is
t
r
a
i
ne
d
on
c
ompl
e
te
da
ta.
‒
S
tanda
r
dizing
f
e
a
tur
e
va
r
iable
s
:
the
las
t
pha
s
e
of
da
ta
p
r
e
pr
oc
e
s
s
ing
invol
ve
s
s
tanda
r
dizing
the
f
e
a
tur
e
va
r
iabl
e
s
,
a
c
r
it
ica
l
s
tep
to
e
ns
ur
e
that
e
a
c
h
f
e
a
tur
e
e
qua
ll
y
inf
luenc
e
s
the
model's
lea
r
ning
pr
o
c
e
s
s
.
S
tanda
r
diza
ti
on
a
djus
ts
the
da
ta
s
o
that
it
ha
s
a
mea
n
of
z
e
r
o
a
nd
a
s
tanda
r
d
de
viation
o
f
one
.
T
his
pr
oc
e
s
s
is
pa
r
ti
c
ular
ly
vit
a
l
f
or
a
lgor
it
hms
that
r
e
ly
on
dis
tanc
e
mea
s
ur
e
s
,
li
ke
r
a
ndom
f
or
e
s
t
s
,
a
s
it
pr
e
ve
nts
f
e
a
tur
e
s
with
lar
ge
r
s
c
a
les
f
r
om
ove
r
whe
lm
ing
the
lea
r
ning
pr
oc
e
s
s
.
T
he
S
tanda
r
dS
c
a
ler
f
r
o
m
s
c
iki
t
-
lea
r
n
wa
s
uti
li
z
e
d
to
s
tanda
r
dize
the
numer
ica
l
f
e
a
tur
e
s
,
ther
e
by
im
pr
oving
the
model's
pe
r
f
or
manc
e
a
n
d
a
c
c
e
ler
a
ti
ng
c
onve
r
ge
nc
e
.
‒
S
pli
tt
ing
da
ta:
pos
t
-
pr
e
pr
oc
e
s
s
ing
,
the
da
tas
e
t
is
s
pli
t
int
o
f
e
a
tur
e
s
(
X)
a
nd
tar
ge
t
va
r
iable
(
y)
,
with
80
%
of
the
da
ta
a
ll
oc
a
ted
f
o
r
t
r
a
ini
ng
a
nd
20%
f
or
tes
ti
n
g.
T
he
f
e
a
tur
e
s
a
r
e
a
ll
c
olu
mns
e
xc
e
pt
'labe
l'
a
nd
a
ny
non
-
numer
ic
c
olum
ns
li
ke
'a
tt
a
c
k_c
a
t',
while
'labe
l'
r
e
pr
e
s
e
nts
the
tar
ge
t
va
r
iable
indi
c
a
ti
ng
whe
ther
the
tr
a
f
f
ic
is
nor
mal
o
r
a
n
a
tt
a
c
k.
T
his
s
pli
tt
ing
pr
e
pa
r
e
s
the
da
ta
f
or
the
s
ubs
e
que
nt
tr
a
ini
ng
a
nd
t
e
s
ti
ng
pha
s
e
s
.
T
a
ble
1.
He
a
d
s
a
mpl
e
da
tas
e
t
F
e
a
tu
re
s
a
mpl
e
_1
s
a
mpl
e
_
2
s
a
mpl
e
_
3
s
a
mpl
e
_
4
s
a
mpl
e
_
5
id
0
1
2
3
4
dur
1
2
3
4
5
pr
ot
o
tc
p
tc
p
tc
p
tc
p
tc
p
S
e
r
vi
c
e
-
-
-
f
tp
-
s
ta
te
F
I
N
F
I
N
F
I
N
F
I
N
F
I
N
s
pkt
s
6
14
8
12
10
dpkt
s
4
38
16
12
6
s
byt
e
s
258
734
364
628
534
dbyt
e
s
172
42014
13186
770
268
r
a
te
74.08749
78.473372
14.170161
13.677108
33.373826
S
por
t_
lt
m
1
1
1
1
1
ds
t_
s
r
c
_l
tm
1
2
3
3
40
f
tp
_l
ogi
n
0
0
0
1
0
f
tp
_c
md
0
0
0
1
0
ht
tp
_mt
hd
0
0
0
0
0
s
r
c
_l
tm
1
1
2
2
2
s
r
v_ds
t
1
6
6
1
39
ip
s
_por
ts
0
0
0
0
0
a
tt
a
c
k_c
a
t
nor
ma
l
nor
ma
l
nor
ma
l
nor
ma
l
nor
ma
l
la
be
l
0
0
0
0
0
3.
3.
M
od
e
l
s
e
lec
t
ion
Dur
ing
the
model
s
e
lec
ti
on
pha
s
e
,
the
r
a
ndom
f
or
e
s
t
a
lgor
it
hm
wa
s
s
e
lec
ted
f
or
it
s
r
obus
tnes
s
,
a
c
c
ur
a
c
y,
a
nd
c
a
pa
bil
it
y
to
mana
ge
lar
ge
da
tas
e
ts
with
numer
ous
f
e
a
tur
e
s
.
F
i
r
s
t
int
r
oduc
e
d
by
B
r
e
im
a
n
in
2001,
r
a
ndom
f
or
e
s
t
s
a
r
e
a
n
e
ns
e
mbl
e
lea
r
ning
tec
hnique
that
buil
ds
mul
ti
ple
de
c
is
ion
t
r
e
e
s
a
nd
a
ggr
e
ga
tes
their
output
s
to
e
nha
nc
e
c
las
s
if
ica
ti
on
a
c
c
ur
a
c
y
a
nd
mi
nim
ize
ove
r
f
it
ti
ng
.
T
his
a
lgor
it
hm
is
pa
r
ti
c
ular
ly
we
ll
-
s
uit
e
d
f
or
dif
f
e
r
e
nt
iating
be
twe
e
n
nor
mal
a
nd
a
tt
a
c
k
tr
a
f
f
ic
in
ne
twor
k
da
ta.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14
,
No.
4
,
Augus
t
20
25
:
294
5
-
2954
2950
O
nc
e
t
he
da
ta
is
s
p
li
t
,
th
e
r
a
n
do
m
f
o
r
e
s
t
mo
de
l
is
i
n
it
ia
l
ize
d
a
nd
t
r
a
ine
d
us
i
ng
th
e
t
r
a
i
ni
ng
s
e
t
.
Ke
y
p
a
r
a
met
e
r
s
,
s
uc
h
a
s
the
n
um
be
r
o
f
t
r
e
e
s
in
t
he
f
o
r
e
s
t
(
n_
e
s
t
im
a
t
o
r
s
)
,
a
r
e
s
e
t
t
o
10
0
,
mea
n
in
g
the
mo
de
l
wi
ll
c
ons
t
r
u
c
t
10
0
de
c
is
io
n
t
r
e
e
s
.
T
he
a
lg
o
r
i
t
hm
wo
r
ks
by
b
oo
ts
t
r
a
p
s
a
m
pl
i
ng
,
whe
r
e
r
a
n
do
m
s
a
m
pl
e
s
o
f
t
he
t
r
a
in
in
g
d
a
t
a
a
r
e
d
r
a
wn
f
o
r
e
a
c
h
t
r
e
e
.
E
a
c
h
t
r
e
e
is
bu
il
t
us
ing
a
s
ubs
e
t
of
f
e
a
t
u
r
e
s
a
t
e
a
c
h
n
ode
,
wh
ich
in
c
r
e
a
s
e
s
d
iv
e
r
s
i
ty
a
nd
r
e
du
c
e
s
ove
r
f
it
t
in
g
.
No
de
s
pl
i
tt
in
g
i
s
ba
s
e
d
on
c
r
i
te
r
ia
li
ke
Gi
ni
im
pu
r
i
ty
,
c
a
l
c
u
la
ted
a
s
i
n
(
1
)
[
1
5
]
.
(
)
=
1
−
∑
2
=
1
(
1)
W
he
r
e
is
the
p
r
oba
bil
it
y
o
f
c
las
s
i
in
the
da
tas
e
t
D
(
1)
.
Af
ter
a
l
l
tr
e
e
s
a
r
e
c
ons
tr
uc
ted,
they
make
in
de
pe
nde
nt
pr
e
dictions
.
T
he
f
inal
p
r
e
diction
is
de
ter
mi
ne
d
thr
ough
major
it
y
voti
ng:
e
a
c
h
t
r
e
e
c
a
s
ts
a
vote
f
or
a
c
las
s
,
a
nd
the
c
las
s
that
r
e
c
e
ives
the
mos
t
votes
is
s
e
lec
ted.
T
his
e
ns
e
mbl
e
a
pp
r
oa
c
h
e
nha
nc
e
s
the
model's
a
c
c
ur
a
c
y
a
nd
s
tabili
ty,
r
e
duc
ing
va
r
i
a
nc
e
a
nd
mi
ti
ga
ti
ng
ove
r
f
it
ti
ng
.
T
he
tr
a
in
ing
pha
s
e
invol
ve
s
f
e
e
ding
the
tr
a
ini
ng
da
ta
int
o
the
model
to
lea
r
n
pa
tt
e
r
ns
dis
ti
nguis
hing
nor
mal
f
r
om
a
tt
a
c
k
tr
a
f
f
ic,
e
ns
ur
ing
r
obus
t
a
nd
r
e
li
a
ble
c
las
s
if
ica
ti
on
pe
r
f
or
manc
e
.
3.
4.
Abl
a
t
ion
s
t
u
d
y
An
a
blation
s
tudy
wa
s
c
a
r
r
ied
out
to
e
va
luate
how
va
r
ious
c
omponents
inf
luenc
e
d
the
model's
pe
r
f
or
manc
e
.
T
his
pr
oc
e
s
s
invol
ve
d
s
ys
tema
ti
c
a
ll
y
modi
f
ying
or
r
e
movi
ng
s
pe
c
if
ic
c
omponents
of
th
e
model
to
a
s
s
e
s
s
thei
r
inf
luenc
e
on
ove
r
a
ll
a
c
c
ur
a
c
y
a
n
d
r
obus
tnes
s
.
Ke
y
a
s
pe
c
ts
s
uc
h
a
s
the
number
of
t
r
e
e
s
(
n_e
s
ti
mator
s
)
,
the
maximum
t
r
e
e
de
pth
(
max_de
pth)
,
a
nd
pr
e
pr
oc
e
s
s
ing
tec
hniques
li
ke
f
e
a
tur
e
s
c
a
li
ng
a
nd
e
nc
oding
we
r
e
tho
r
oughly
e
xa
mi
ne
d.
T
he
f
indi
n
gs
f
r
om
thi
s
s
tud
y
we
r
e
ins
tr
umenta
l
in
f
ine
-
tuni
ng
the
model,
e
nha
nc
ing
it
s
c
a
pa
bil
it
y
to
de
tec
t
br
ute
f
o
r
c
e
a
tt
a
c
ks
e
f
f
e
c
ti
ve
ly.
T
his
a
ppr
oa
c
h
e
ns
ur
e
d
that
th
e
model
wa
s
opti
mi
z
e
d
f
or
pe
a
k
pe
r
f
or
manc
e
in
pr
a
c
ti
c
a
l
a
ppli
c
a
ti
ons
.
3.
5.
Re
s
u
lt
a
n
alys
is
T
he
r
e
s
u
lt
s
we
r
e
a
n
a
l
y
z
e
d
c
om
pr
e
h
e
n
s
iv
e
l
y
t
o
e
v
a
l
ua
te
th
e
e
f
f
i
c
i
e
nc
y
a
nd
e
f
f
e
c
ti
v
e
n
e
s
s
of
t
h
e
p
r
o
po
s
e
d
a
p
pr
oa
c
h
in
op
ti
m
iz
in
g
f
i
r
e
w
a
ll
de
p
lo
ym
e
n
t
t
im
i
ng
f
or
e
n
ha
n
c
e
d
br
ut
e
f
or
c
e
mi
ti
ga
ti
on.
K
e
y
p
e
r
f
or
m
a
nc
e
m
e
t
r
i
c
s
s
u
c
h
a
s
a
c
c
u
r
a
c
y,
pr
e
c
i
s
i
on,
r
e
c
a
ll
,
a
n
d
F
1
-
s
c
or
e
w
e
r
e
c
a
lc
ul
a
t
e
d
us
i
ng
th
e
f
oll
o
win
g
f
or
m
ul
a
s
[
1
1]
.
T
he
TP
r
e
pr
e
s
e
n
ts
tr
ue
po
s
i
ti
ve
s
,
TN
tr
ue
n
e
g
a
t
iv
e
s
,
FP
f
a
l
s
e
p
o
s
i
ti
v
e
s
,
a
n
d
FN
f
a
l
s
e
n
e
ga
ti
ve
s
.
‒
Ac
c
ur
a
c
y
mea
s
ur
e
s
the
ove
r
a
ll
c
or
r
e
c
tnes
s
of
th
e
model
by
e
va
luating
the
pr
opor
ti
on
o
f
tot
a
l
c
or
r
e
c
t
pr
e
dictions
out
of
a
ll
p
r
e
dictions
made
:
=
+
+
+
+
(
2)
‒
P
r
e
c
is
ion
a
s
s
e
s
s
e
s
the
model's
a
bil
it
y
to
identi
f
y
o
nly
the
r
e
leva
nt
ins
tanc
e
s
of
a
tt
a
c
ks
by
c
a
lcula
ti
n
g
the
r
a
ti
o
of
c
or
r
e
c
tl
y
pr
e
dicte
d
a
tt
a
c
k
ins
tanc
e
s
to
the
t
otal
pr
e
dicte
d
a
tt
a
c
k
ins
tanc
e
s
:
=
+
(
3)
‒
R
e
c
a
ll
a
s
s
e
s
s
e
s
the
model's
a
bil
it
y
to
identif
y
a
l
l
a
c
tual
a
tt
a
c
k
ins
tanc
e
s
by
de
ter
mi
ning
the
r
a
ti
o
of
c
or
r
e
c
tl
y
pr
e
dicte
d
a
tt
a
c
k
ins
tanc
e
s
to
the
tot
a
l
nu
mber
of
a
c
tual
a
tt
a
c
k
ins
tanc
e
s
.
=
+
(
4)
‒
F1
-
s
c
or
e
of
f
e
r
s
a
ba
lanc
e
d
mea
s
ur
e
be
twe
e
n
pr
e
c
i
s
ion
a
nd
r
e
c
a
ll
,
making
it
pa
r
ti
c
ular
ly
va
luable
in
c
a
s
e
s
of
im
ba
lanc
e
d
c
las
s
dis
tr
ibut
ion.
I
t
is
c
a
lcula
ted
a
s
the
ha
r
moni
c
mea
n
o
f
p
r
e
c
is
ion
a
nd
r
e
c
a
ll
.
1
−
=
2
.
.
+
(
5)
4.
RE
S
UL
T
S
AN
D
DI
S
CU
S
S
I
ON
T
he
r
e
s
u
lt
s
of
th
is
s
tu
dy
d
e
m
on
s
tr
a
t
e
th
e
e
f
f
e
c
ti
v
e
ne
s
s
of
u
s
i
ng
t
h
e
r
a
nd
om
f
or
e
s
t
a
lg
or
i
thm
f
or
o
pt
im
iz
in
g
f
ir
e
wa
ll
d
e
pl
oym
e
nt
t
im
in
g
to
mi
t
ig
a
te
br
u
t
e
f
o
r
c
e
a
tt
a
c
k
s
.
T
he
mo
de
l
a
c
hi
e
v
e
d
hi
gh
a
c
c
ur
a
c
y
i
n
di
s
t
in
gu
i
s
h
in
g
b
e
t
w
e
e
n
nor
ma
l
a
n
d
a
t
ta
c
k
tr
a
f
f
ic,
a
s
e
vi
de
nc
e
d
b
y
ke
y
pe
r
f
or
ma
n
c
e
m
e
tr
ic
s
.
T
h
e
c
a
lc
ul
a
t
e
d
a
c
c
ur
a
c
y
wa
s
98.
87
%
,
in
di
c
a
ti
ng
a
h
ig
h
o
ve
r
a
ll
c
or
r
e
c
t
n
e
s
s
in
th
e
m
od
e
l’
s
pr
e
d
i
c
ti
on
s
.
P
r
e
c
i
s
io
n
w
a
s
f
ou
nd
to
b
e
9
8.
9
9%
,
r
e
f
l
e
c
t
in
g
t
he
m
od
e
l
’
s
a
bi
li
t
y
to
c
o
r
r
e
c
t
ly
id
e
nti
f
y
a
t
ta
c
k
i
n
s
t
a
nc
e
s
wi
th
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
Optimiz
ing
fi
r
e
w
all
ti
ming
for
br
ute
for
c
e
miti
gati
on
w
it
h
r
andom
for
e
s
ts
(
A
hmad
T
ur
mudi
Z
y
)
2951
m
in
im
a
l
f
a
l
s
e
po
s
it
i
v
e
s
.
T
he
r
e
c
a
ll
r
a
te
s
t
oo
d
a
t
9
9.
7
7%
,
s
h
ow
c
a
s
i
ng
t
he
m
od
e
l’
s
c
a
p
a
bi
li
ty
t
o
c
a
p
tu
r
e
a
lm
o
s
t
a
ll
a
c
tu
a
l
a
t
ta
c
k
in
s
t
a
nc
e
s
,
a
nd
t
h
e
F
1
-
s
c
or
e
w
a
s
9
9.
3
8%
,
in
di
c
a
ti
ng
a
r
o
bu
s
t
ba
la
nc
e
b
e
tw
e
e
n
p
r
e
c
i
s
io
n
a
nd
r
e
c
a
ll
.
T
he
c
onf
us
ion
matr
ix
f
ur
ther
e
lucida
ted
the
model's
pe
r
f
or
manc
e
.
I
t
s
howe
d
in
T
a
ble
2
that
out
o
f
35,
069
tot
a
l
ins
tanc
e
s
,
10
,
827
no
r
mal
ins
tanc
e
s
we
r
e
c
or
r
e
c
tl
y
c
las
s
if
ied,
while
342
we
r
e
mi
s
c
la
s
s
if
ied
a
s
a
tt
a
c
ks
.
C
onve
r
s
e
ly,
23,
845
a
tt
a
c
k
ins
tanc
e
s
we
r
e
c
or
r
e
c
tl
y
identif
ied,
with
only
55
be
ing
wr
ongly
c
las
s
if
ied
a
s
nor
mal
tr
a
f
f
ic.
T
his
high
t
r
ue
pos
it
ive
r
a
te
,
c
oupled
with
mi
nim
a
l
f
a
ls
e
ne
ga
ti
ve
s
a
nd
f
a
ls
e
pos
it
ives
,
unde
r
s
c
or
e
s
the
mode
l's
r
e
li
a
bil
it
y
in
r
e
a
l
-
wor
ld
s
c
e
na
r
ios
.
T
a
ble
2.
C
onf
us
ion
m
a
tr
ix
P
r
e
di
c
te
d
A
c
tu
a
l
\
P
r
e
di
c
te
d
0
1
T
ot
a
l
A
c
tu
a
l
0
10827
342
11169
1
55
23845
23900
T
ot
a
l
10882
24187
35069
An
a
blation
s
tudy
f
u
r
ther
dis
s
e
c
ted
the
c
ontr
ibut
i
ons
of
dif
f
e
r
e
nt
c
omponents
withi
n
the
model
.
B
y
s
ys
tema
ti
c
a
ll
y
r
e
movi
ng
or
modi
f
ying
pa
r
ts
o
f
th
e
model,
it
wa
s
obs
e
r
ve
d
that
c
e
r
tain
f
e
a
tur
e
s
s
igni
f
ica
ntl
y
e
nha
nc
e
d
the
model's
pe
r
f
or
manc
e
,
ther
e
by
f
ine
-
t
uning
it
s
e
f
f
e
c
ti
ve
ne
s
s
in
de
tec
ti
ng
br
ute
f
or
c
e
a
tt
a
c
ks
.
T
he
ins
ight
s
ga
ined
f
r
om
the
a
blation
s
tudy
he
lped
in
r
e
f
in
ing
the
model
,
e
ns
ur
ing
that
the
mos
t
c
r
it
ica
l
c
omponents
we
r
e
opti
mi
z
e
d
f
or
be
tt
e
r
pe
r
f
o
r
manc
e
.
T
he
c
onf
us
ion
mat
r
ix
p
r
ovided
a
ddit
ional
ins
ight
s
int
o
the
model's
a
bil
it
y
to
c
las
s
if
y
nor
mal
a
nd
a
tt
a
c
k
tr
a
f
f
ic
a
c
c
ur
a
tely.
I
t
r
e
ve
a
led
a
high
tr
ue
po
s
it
ive
r
a
te
with
mi
nim
a
l
f
a
ls
e
ne
ga
ti
ve
s
a
nd
f
a
ls
e
pos
it
ives
,
unde
r
s
c
or
ing
the
model's
r
e
li
a
bil
it
y
in
r
e
a
l
-
wor
ld
s
c
e
na
r
ios
.
T
his
c
ompr
e
he
ns
ive
a
na
lys
is
c
onf
ir
ms
the
pr
opos
e
d
a
ppr
oa
c
h's
e
f
f
ica
c
y
in
e
nha
nc
i
ng
c
ybe
r
s
e
c
ur
it
y
mea
s
ur
e
s
thr
ough
opti
mi
z
e
d
f
ir
e
wa
ll
de
ploym
e
nt
ti
mi
ng,
ult
im
a
tely
c
ontr
ibut
ing
to
mo
r
e
r
obus
t
n
e
twor
k
de
f
e
ns
e
mec
ha
nis
ms
a
ga
ins
t
e
volvi
ng
br
ute
f
or
c
e
a
tt
a
c
ks
.
T
his
r
e
s
e
a
r
c
h
highl
ight
s
the
potential
of
mac
hine
lea
r
ning,
pa
r
ti
c
ular
ly
the
r
a
ndom
f
or
e
s
t
a
lgor
it
hm,
in
a
ddr
e
s
s
ing
c
r
it
ica
l
c
ybe
r
s
e
c
ur
it
y
c
ha
ll
e
nge
s
.
C
onti
nuous
innovation
a
nd
r
igor
ous
e
va
luation,
a
s
de
mons
tr
a
ted
in
thi
s
s
tudy,
a
r
e
e
s
s
e
nti
a
l
f
or
de
ve
lopi
n
g
e
f
f
e
c
ti
ve
s
olut
ions
to
c
ounter
a
c
t
the
e
ve
r
-
e
volvi
ng
lands
c
a
pe
of
c
ybe
r
thr
e
a
ts
.
5.
CONC
L
USI
ON
T
his
s
tudy
int
r
oduc
e
s
a
r
obus
t
a
pp
r
oa
c
h
to
opti
mi
z
ing
f
i
r
e
wa
ll
de
ploym
e
nt
ti
mi
ng
f
or
e
nha
nc
e
d
br
ute
f
or
c
e
a
tt
a
c
k
mi
ti
ga
ti
on
by
leve
r
a
ging
pa
tt
e
r
n
r
e
c
ognit
ion
wit
h
the
r
a
ndom
f
or
e
s
t
a
lgor
it
h
m.
T
he
e
xpe
r
im
e
ntal
r
e
s
ult
s
de
mons
tr
a
te
the
model's
e
xc
e
pti
ona
l
pe
r
f
or
manc
e
,
a
c
hieving
a
n
a
c
c
ur
a
c
y
of
98.
87%
,
pr
e
c
is
ion
of
98
.
99%
,
r
e
c
a
ll
of
99
.
77%
,
a
nd
a
n
F
1
-
s
c
or
e
of
99.
38%
.
T
he
s
e
metr
ics
highl
ight
the
model's
a
bil
it
y
to
a
c
c
ur
a
tely
identif
y
a
tt
a
c
k
tr
a
f
f
ic
whi
le
maintaining
a
low
r
a
te
o
f
f
a
ls
e
pos
it
ives
a
nd
e
f
f
e
c
ti
ve
ly
de
tec
ti
ng
ne
a
r
ly
a
ll
a
c
tual
a
tt
a
c
k
ins
tanc
e
s
.
T
he
a
blation
s
tudy
f
u
r
ther
pr
ovided
c
r
it
ica
l
ins
ight
s
int
o
the
c
ontr
ibut
ions
of
s
pe
c
if
ic
model
c
omponents
,
e
na
bli
ng
f
ine
-
tuni
ng
to
e
nha
nc
e
the
a
lgor
it
hm’
s
ove
r
a
ll
e
f
f
e
c
ti
ve
ne
s
s
.
T
he
f
indi
ngs
c
onf
ir
m
the
e
f
f
ic
a
c
y
of
int
e
gr
a
ti
ng
mac
hine
lea
r
ning
tec
hniqu
e
s
int
o
c
ybe
r
s
e
c
ur
it
y
f
r
a
mew
or
ks
to
s
tr
e
ngthen
ne
twor
k
de
f
e
ns
e
mec
ha
nis
m
s
.
B
y
opti
mi
z
ing
f
ir
e
wa
ll
de
ploym
e
nt
ti
mi
ng,
thi
s
r
e
s
e
a
r
c
h
c
ontr
ibut
e
s
to
de
ve
lopi
ng
pr
oa
c
ti
ve
a
nd
a
da
pti
ve
s
tr
a
tegie
s
f
or
mi
ti
ga
ti
ng
br
u
te
f
or
c
e
a
tt
a
c
ks
.
F
utur
e
wor
k
c
ould
e
xtend
thi
s
methodolog
y
to
a
ddr
e
s
s
other
c
ybe
r
th
r
e
a
ts
,
r
e
f
ine
the
model
t
o
ha
ndle
e
mer
ging
a
tt
a
c
k
pa
tt
e
r
ns
,
a
nd
e
xplo
r
e
r
e
a
l
-
ti
me
de
ploym
e
nt
s
c
e
na
r
ios
.
T
he
s
e
a
dva
nc
e
ments
will
play
a
pivot
a
l
r
ole
in
f
os
ter
ing
mor
e
r
e
s
il
ient
a
nd
s
e
c
ur
e
ne
twor
k
in
f
r
a
s
tr
uc
tur
e
s
to
mee
t
the
de
man
ds
of
a
n
e
ve
r
-
e
volvi
ng
c
ybe
r
s
e
c
ur
it
y
lands
c
a
pe
.
F
UN
DI
NG
I
NF
ORM
AT
I
ON
T
his
wor
k
wa
s
s
uppor
ted
by
a
r
e
s
e
a
r
c
h
G
r
a
nt
f
r
om
De
pa
r
tm
e
nt
of
R
e
s
e
a
r
c
h
a
nd
C
omm
unit
y
S
e
r
vice
,
Unive
r
s
it
a
s
P
e
li
ta
B
a
ngs
a
,
f
or
s
uppor
ti
ng
a
nd
f
unding
thi
s
r
e
s
e
a
r
c
h.
T
he
a
uthor
de
c
lar
e
s
no
c
onf
li
c
t
of
int
e
r
e
s
t.
T
his
r
e
s
e
a
r
c
h
r
e
c
e
ived
no
s
pe
c
if
ic
gr
a
nt
f
r
om
a
ny
f
unding
a
ge
nc
y
in
the
publi
c
,
c
omm
e
r
c
ial,
or
not
-
f
or
-
pr
of
it
s
e
c
tor
s
.
AU
T
HO
R
CONT
RI
B
U
T
I
ONS
S
T
AT
E
M
E
N
T
T
his
jour
na
l
us
e
s
the
C
ontr
ibut
o
r
R
oles
T
a
xo
nomy
(
C
R
e
diT
)
to
r
e
c
ognize
indi
vidual
a
uthor
c
ontr
ibut
ions
,
r
e
duc
e
a
utho
r
s
hip
dis
putes
,
a
nd
f
a
c
il
it
a
te
c
oll
a
bor
a
ti
on.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14
,
No.
4
,
Augus
t
20
25
:
294
5
-
2954
2952
Nam
e
of
Au
t
h
or
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
Ahma
d
T
ur
mudi
Z
y
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
I
s
a
r
ianto
✓
✓
✓
✓
✓
A
n
g
g
i
M
u
h
a
m
m
a
d
R
i
f
a
'
i
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
Abdul
Ghof
ir
✓
✓
✓
✓
M
uha
mm
a
d
Na
jamuddin
Dw
i
M
ihar
ja
✓
✓
✓
✓
✓
✓
✓
✓
Ana
nto
T
r
i
S
a
s
ongko
✓
✓
✓
✓
✓
C
:
C
onc
e
pt
ua
li
z
a
ti
on
M
:
M
e
th
odol
ogy
So
:
So
f
twa
r
e
Va
:
Va
li
da
ti
on
Fo
:
Fo
r
ma
l
a
na
ly
s
is
I
:
I
nve
s
ti
ga
ti
on
R
:
R
e
s
our
c
e
s
D
:
D
a
ta
C
ur
a
ti
on
O
:
W
r
it
in
g
-
O
r
ig
in
a
l
D
r
a
f
t
E
:
W
r
it
in
g
-
R
e
vi
e
w
&
E
di
ti
ng
Vi
:
Vi
s
ua
li
z
a
ti
on
Su
:
Su
pe
r
vi
s
io
n
P
:
P
r
oj
e
c
t
a
dmi
ni
s
tr
a
ti
on
Fu
:
Fu
ndi
ng a
c
qui
s
it
io
n
CONF
L
I
CT
OF
I
NT
E
RE
S
T
S
T
AT
E
M
E
N
T
T
he
a
uthor
s
de
c
lar
e
that
ha
ve
no
known
c
ompetin
g
f
inanc
ial
int
e
r
e
s
ts
or
pe
r
s
ona
l
r
e
lations
hips
that
c
ould
ha
ve
a
ppe
a
r
e
d
to
in
f
luenc
e
the
wor
k
r
e
por
te
d
in
thi
s
pa
pe
r
.
DA
T
A
AV
AI
L
A
B
I
L
I
T
Y
T
he
da
ta
that
s
uppor
t
the
f
ind
ings
of
thi
s
s
tudy
a
r
e
ope
nly
a
va
il
a
ble
in
UN
S
W
-
NB
15
a
t
htt
p:/
/doi
.
or
g
/10.
1109/M
il
C
I
S
.
2015.
7348942
,
r
e
f
e
r
e
nc
e
number
26.
RE
F
E
RE
NC
E
S
[
1]
A
. D
je
nna
, S
. H
a
r
ous
, a
nd D
. E
. S
a
id
ouni
, “
I
nt
e
r
ne
t
of
th
in
gs
m
e
e
t
in
te
r
ne
t
of
t
hr
e
a
ts
:
ne
w
c
onc
e
r
n c
ybe
r
s
e
c
ur
it
y i
s
s
u
e
s
of
c
r
it
ic
a
l
c
ybe
r
i
nf
r
a
s
tr
uc
tu
r
e
,”
A
ppl
ie
d Sc
ie
nc
e
s
, vol
. 11, no. 10, M
a
y 20
21, doi:
10.3390/app111045
80.
[
2]
J
.
L
uxe
mbur
k,
K
.
H
yne
k,
a
nd
T
.
C
e
jk
a
,
“
D
e
te
c
ti
on
of
H
T
T
P
S
br
ut
e
-
f
or
c
e
a
tt
a
c
ks
w
it
h
pa
c
ke
t
-
le
ve
l
f
e
a
tu
r
e
s
e
t,
”
in
2021
I
E
E
E
11t
h
A
nnual
C
om
put
in
g
and
C
om
m
uni
c
at
io
n
W
or
k
s
hop
and
C
onf
e
r
e
nc
e
(
C
C
W
C
)
,
J
a
n.
2021,
pp.
0114
–
0122
,
doi
:
10.1109/C
C
W
C
51732.2021.9375998.
[
3]
M
.
Z
.
H
us
s
a
in
,
Z
.
M
.
H
a
na
pi
,
A
.
A
bdul
la
h, M
.
H
u
s
s
in
, a
nd
M
.
I
.
H
.
N
in
gga
l,
“
A
n
e
f
f
ic
ie
nt
s
e
c
ur
e
a
nd e
ne
r
gy
r
e
s
il
ie
nt
tr
us
t
-
ba
s
e
d
s
ys
te
m
f
or
de
te
c
ti
on
a
nd
mi
ti
ga
ti
on
of
s
ybi
l
a
tt
a
c
k
d
e
te
c
ti
on
(
S
A
N
)
,”
P
e
e
r
J
C
om
put
e
r
S
c
ie
nc
e
,
vol
.
10,
A
ug.
20
24,
doi
:
10.7
717/
pe
e
r
j
-
c
s
.2231.
[
4]
A
.
F
.
O
to
om,
W
.
E
le
is
a
h,
a
nd
E
.
E
.
A
bda
ll
a
h,
“
D
e
e
p
le
a
r
ni
ng
f
or
a
c
c
ur
a
te
de
te
c
ti
on
of
br
ut
e
f
or
c
e
a
tt
a
c
ks
o
n
I
o
T
n
e
twor
ks
,”
P
r
oc
e
di
a C
om
put
e
r
S
c
ie
nc
e
, vol
. 220, pp. 291
–
298, 2023, doi:
10.1016/j
.pr
oc
s
.2023.03.038.
[
5]
S
.
S
.
N
a
le
ga
e
v
a
nd
N
.
V
.
P
e
tr
ov,
“
S
im
pl
e
c
r
it
e
r
ia
to
de
te
r
mi
n
e
th
e
s
e
t
of
ke
y
pa
r
a
me
te
r
s
of
th
e
D
R
P
E
me
th
od
by
a
br
ut
e
-
f
or
c
e
a
tt
a
c
k
,”
P
hy
s
ic
s
P
r
oc
e
di
a
, vol
. 73, pp. 281
–
286, 2015, doi:
10.
1016/j
.phpr
o.2015.09.137.
[
6]
A
.
S
.
E
du,
M
.
A
goyi
,
a
nd
D
.
A
go
z
ie
,
“
D
ig
it
a
l
s
e
c
ur
it
y
vul
ne
r
a
bi
li
ti
e
s
a
nd
th
r
e
a
t
s
im
pl
ic
a
ti
ons
f
or
f
in
a
nc
ia
l
in
s
ti
tu
ti
ons
de
p
lo
yi
ng
di
gi
ta
l
te
c
hnol
ogy
pl
a
tf
or
ms
a
nd
a
ppl
ic
a
ti
on:
F
M
E
A
a
nd
F
T
O
P
S
I
S
a
na
ly
s
is
,”
P
e
e
r
J
C
o
m
put
e
r
Sc
i
e
nc
e
,
vol
.
7,
A
ug. 2021, doi:
1
0.7717/pe
e
r
j
-
c
s
.658.
[
7]
A
.
J
os
hi
,
M
.
W
a
z
id
,
a
nd
R
.
H
.
G
ouda
r
,
“
A
n
e
f
f
ic
ie
nt
c
r
ypt
og
r
a
phi
c
s
c
he
me
f
or
te
xt
me
s
s
a
ge
pr
ot
e
c
ti
on
a
ga
in
s
t
br
ut
e
f
or
c
e
a
nd
c
r
ypt
a
na
ly
ti
c
a
tt
a
c
ks
,”
P
r
o
c
e
di
a C
om
put
e
r
Sc
ie
nc
e
, vol
. 48, pp.
360
–
366, 2015, doi:
10.1016/j
.pr
oc
s
.2015.0
4.194.
[
8]
J.
-
S
.
C
ho,
Y
.
-
S
.
J
e
ong,
a
nd
S
.
O
.
P
a
r
k,
“
C
ons
id
e
r
a
ti
on
on
th
e
br
ut
e
-
f
or
c
e
a
tt
a
c
k
c
os
t
a
nd
r
e
tr
ie
va
l
c
os
t:
A
ha
s
h
-
ba
s
e
d
r
a
di
o
-
f
r
e
que
nc
y
id
e
nt
if
ic
a
ti
on
(
R
F
I
D
)
ta
g
mut
ua
l
a
ut
he
nt
ic
a
ti
on
pr
ot
oc
ol
,”
C
om
put
e
r
s
&
M
at
he
m
at
ic
s
w
it
h
A
ppl
ic
at
io
ns
,
vol
.
69,
no. 1, pp. 58
–
65, J
a
n. 2015, doi:
10.1016/j
.c
a
mwa
.2012.02.025.
[
9]
S
.
J
a
c
ob,
Y
.
Q
ia
o,
Y
.
Y
e
,
a
nd
B
.
L
e
e
,
“
A
noma
lo
us
di
s
tr
ib
ut
e
d
tr
a
f
f
ic
:
d
e
te
c
ti
ng
c
ybe
r
s
e
c
ur
it
y
a
tt
a
c
ks
a
mongs
t
mi
c
r
os
e
r
v
ic
e
s
us
in
g gr
a
ph c
onvolut
io
na
l
ne
twor
ks
,”
C
om
put
e
r
s
& Se
c
ur
it
y
, v
ol
. 118, J
ul
. 2022, doi:
10.1016/j
.c
os
e
.2022.102728.
[
10]
T
a
s
ke
e
n
a
nd
S
.
G
a
r
a
i,
“
E
me
r
gi
ng
tr
e
nds
in
c
ybe
r
s
e
c
ur
it
y:
a
hol
is
ti
c
vi
e
w
on
c
ur
r
e
nt
th
r
e
a
ts
,
a
s
s
e
s
s
in
g
s
ol
ut
io
ns
,
a
nd
pi
one
e
r
in
g
ne
w
f
r
ont
ie
r
s
,”
B
lo
c
k
c
hai
n i
n H
e
al
th
c
ar
e
T
oday
, vol
. 7, no. 1,
A
pr
. 202
4, doi:
10.30953/bht
y.v7.302.
[
11]
A
.
M
.
R
if
a
i,
S
.
R
a
ha
r
jo
,
E
.
U
ta
mi
,
a
nd
D
.
A
r
ia
tm
a
nt
o,
“
A
na
ly
s
is
f
or
di
a
gnos
is
of
pne
umoni
a
s
ympt
oms
us
in
g
c
h
e
s
t
X
-
r
a
y
b
a
s
e
d
on
M
obi
le
N
e
tV2
mode
ls
w
it
h
im
a
ge
e
nha
nc
e
me
nt
us
in
g
w
h
it
e
ba
la
nc
e
a
nd
c
ont
r
a
s
t
li
mi
te
d
a
d
a
pt
iv
e
hi
s
to
gr
a
m
e
qua
li
z
a
t
io
n
(
C
L
A
H
E
)
,”
B
io
m
e
di
c
al
Si
gnal
P
r
oc
e
s
s
in
g and C
ont
r
ol
, vol
. 90
, A
pr
. 2024, doi:
10.1016/j
.bs
pc
.2023.105857.
[
12]
N
.
M
is
hr
a
a
nd
S
.
P
a
ndya
,
“
I
nt
e
r
ne
t
of
T
hi
ngs
a
ppl
ic
a
ti
ons
,
s
e
c
ur
it
y
c
ha
ll
e
nge
s
,
a
tt
a
c
ks
,
in
tr
us
io
n
de
te
c
ti
on,
a
nd
f
ut
ur
e
vi
s
io
n
s
:
a
s
ys
te
ma
ti
c
r
e
vi
e
w
,
”
I
E
E
E
A
c
c
e
s
s
, vol
. 9, pp. 59353
–
59377, 2021, doi:
10.1109/AC
C
E
S
S
.2021.3073408.
[
13]
V
.
G
i
r
a
ddi
,
S
.
G
i
r
a
ddi
,
N
.
D
G
,
A
.
B
id
a
r
a
ga
ddi
,
a
nd
S
.
G
.
K
a
na
ka
r
e
ddi
,
“
M
a
c
hi
n
e
le
a
r
ni
ng
a
ppr
oa
c
h
to
in
tr
us
io
n
de
te
c
t
io
n:
pe
r
f
or
ma
nc
e
e
va
lu
a
ti
on
,”
P
r
oc
e
di
a C
om
put
e
r
Sc
ie
n
c
e
, vol
. 23
5
, pp. 1851
–
1859, 2024, doi:
10.1016/j
.pr
oc
s
.2024.04.176.
[
14]
Q
.
D
u
a
nd
J
.
Z
ha
i,
“
A
ppl
ic
a
ti
on
of
a
r
ti
f
ic
ia
l
in
te
ll
ig
e
nc
e
s
e
ns
or
s
ba
s
e
d
on
r
a
ndom
f
or
e
s
t
a
lg
or
it
hm
in
f
in
a
nc
ia
l
r
e
c
ogni
ti
on
mode
ls
,”
M
e
as
ur
e
m
e
nt
:
Se
ns
o
r
s
, vol
. 33,
J
un. 2024, doi:
10.1016/j
.me
a
s
e
n.2024.101245.
[
15]
M
.
C
he
n
a
nd
Z
.
L
iu
,
“
P
r
e
di
c
ti
ng
pe
r
f
or
ma
nc
e
of
s
tu
de
nt
s
by
opt
im
iz
in
g
tr
e
e
c
ompone
nt
s
of
r
a
ndom
f
or
e
s
t
us
in
g
ge
n
e
ti
c
a
lg
or
it
hm,”
H
e
li
y
on
, vol
. 10, no. 12, J
un. 2024, doi:
10.1016/j
.he
li
yon.2024.e
32570.
[
16]
J
.
Z
ha
ng,
“
I
mpa
c
t
of
a
n
im
pr
ove
d
r
a
ndom
f
or
e
s
t
-
ba
s
e
d
f
in
a
nc
ia
l
ma
na
ge
m
e
nt
mode
l
on
th
e
e
f
f
e
c
ti
ve
ne
s
s
of
c
or
por
a
te
s
us
ta
in
a
bi
li
ty
de
c
is
io
n
s
,”
Sy
s
te
m
s
and Soft
C
om
put
in
g
, vol
. 6, De
c
. 2024, doi:
10.1016/j
.s
a
s
c
.2024.200102.
[
17]
A
.
K
.
D
e
y,
G
.
P
.
G
upt
a
,
a
nd
S
.
P
.
S
a
hu,
“
A
n
e
f
f
ic
ie
nt
c
y
be
r
a
s
s
a
ul
t
de
te
c
ti
on
s
ys
te
m
u
s
in
g
f
e
a
tu
r
e
opt
im
iz
a
ti
on
f
o
r
I
oT
-
ba
s
e
d
c
ybe
r
s
pa
c
e
,
”
P
r
oc
e
di
a C
om
put
e
r
Sc
ie
n
c
e
, vol
. 235, pp. 757
–
76
6, 2024, doi:
10.1016/j
.pr
oc
s
.2024.04.072.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
Optimiz
ing
fi
r
e
w
all
ti
ming
for
br
ute
for
c
e
miti
gati
on
w
it
h
r
andom
for
e
s
ts
(
A
hmad
T
ur
mudi
Z
y
)
2953
[
18]
A
.
M
.
R
if
a
’
i,
E
.
U
ta
mi
,
a
nd
D
.
A
r
ia
tm
a
nt
o,
“
A
na
ly
s
i
s
f
or
di
a
gn
os
is
of
pne
umoni
a
s
ympt
oms
us
in
g
c
he
s
t
x
-
r
a
y
ba
s
e
d
on
r
e
s
ne
t
-
50
mode
ls
w
it
h
di
f
f
e
r
e
nt
e
poc
h
,”
in
2022
6t
h
I
nt
e
r
nat
io
nal
C
onf
e
r
e
nc
e
on
I
nf
or
m
at
io
n
T
e
c
hnol
ogy
,
I
nf
or
m
at
io
n
Sy
s
te
m
s
and
E
le
c
tr
ic
al
E
ngi
ne
e
r
in
g (
I
C
I
T
I
SE
E
)
, D
e
c
. 2022, pp. 471
–
476
, doi
:
10.1109/I
C
I
T
I
S
E
E
57756.2022.10057805.
[
19]
A
.
R
a
z
a
,
K
.
M
uni
r
,
M
.
S
.
A
lm
ut
a
ir
i,
a
nd
R
.
S
e
ha
r
,
“
N
ove
l
c
la
s
s
pr
oba
bi
li
ty
f
e
a
tu
r
e
s
f
or
op
ti
mi
z
in
g
ne
twor
k
a
tt
a
c
k
de
te
c
ti
on
w
it
h
ma
c
hi
ne
l
e
a
r
ni
ng
,”
I
E
E
E
A
c
c
e
s
s
, v
ol
. 11, pp. 98685
–
98694, 2023, doi:
10.1109/AC
C
E
S
S
.2023.3313596.
[
20]
S
.
F
a
c
c
hi
ne
tt
i,
S
.
A
.
O
s
me
tt
i,
a
nd
C
.
T
a
r
a
nt
ol
a
,
“
N
e
twor
k
mode
ls
f
or
c
ybe
r
a
tt
a
c
k
s
e
va
lu
a
ti
on,”
Soc
io
-
E
c
onomic
P
la
nni
ng
Sc
ie
nc
e
s
, vol
. 87,
J
un. 2023, doi:
10.1016/j
.s
e
ps
.2023.101584.
[
21
]
E
.
I
r
s
ha
d
a
nd
A
.
B
.
S
id
di
qui
,
“
C
ont
e
xt
-
a
w
a
r
e
c
ybe
r
-
th
r
e
a
t
a
t
tr
ib
ut
io
n
ba
s
e
d
on
hybr
id
f
e
a
tu
r
e
s
,”
I
C
T
E
x
p
r
e
s
s
,
vol
.
10,
no
.
3,
pp. 553
–
569, J
un. 2024, doi:
10.1016/j
.i
c
te
.2024.04.005.
[
22]
W
.
F
.
U
r
mi
e
t
al
.
,
“
A
s
ta
c
ke
d
e
ns
e
mbl
e
a
ppr
oa
c
h
to
de
t
e
c
t
c
ybe
r
a
tt
a
c
ks
ba
s
e
d
on
f
e
a
tu
r
e
s
e
le
c
ti
on
te
c
hni
que
s
,
”
I
nt
e
r
nat
i
onal
J
our
nal
of
C
ogni
ti
v
e
C
om
put
in
g i
n E
ngi
ne
e
r
in
g
, vol
. 5, pp. 316
–
331, 2024, doi:
10.1016/j
.i
jc
c
e
.2024.07.005.
[
23]
M
.
F
.
S
a
f
it
r
a
,
M
.
L
ub
is
,
a
nd
H
.
F
a
khr
ur
r
oj
a
,
“
C
ount
e
r
a
tt
a
c
ki
ng
c
ybe
r
th
r
e
a
ts
:
a
f
r
a
me
w
or
k
f
or
th
e
f
ut
ur
e
of
c
ybe
r
s
e
c
ur
it
y
,”
Sus
ta
in
abi
li
ty
, vol
. 15, no. 18, S
e
p. 2023, doi:
10.3390/s
u151813369.
[
24]
A
.
N
.
K
ia
,
F
.
M
ur
phy,
B
.
S
he
e
ha
n,
a
nd
D
.
S
ha
nnon,
“
A
c
ybe
r
r
is
k
pr
e
di
c
ti
on
mode
l
us
in
g
c
omm
on
vul
ne
r
a
bi
li
ti
e
s
a
nd
e
xpos
ur
e
s
,”
E
x
pe
r
t
Sy
s
te
m
s
w
it
h A
ppl
ic
at
io
ns
, vol
. 237, M
a
r
. 2
024, doi:
10.1016/j
.e
s
w
a
.2023.121599.
[
25]
C
.
M
ir
one
a
nu,
A
.
A
r
c
hi
p,
C
.
-
M
.
A
ma
r
a
nde
i,
a
nd
M
.
C
r
a
us
,
“
E
xpe
r
im
e
nt
a
l
c
ybe
r
a
tt
a
c
k
de
te
c
ti
on
f
r
a
me
w
or
k
,”
E
le
c
tr
oni
c
s
,
vol
. 10, no. 14, J
ul
. 2021, doi:
10.3390/ele
c
tr
oni
c
s
10141
682.
[
26]
N
.
M
ous
ta
f
a
a
nd
J
.
S
la
y,
“
U
N
S
W
-
N
B
15:
a
c
ompr
e
he
ns
iv
e
da
ta
s
e
t
f
or
ne
twor
k
in
tr
us
io
n
de
te
c
ti
on
s
ys
te
ms
(
U
N
S
W
-
N
B
15
ne
twor
k
da
ta
s
e
t)
,”
in
2015
M
il
it
ar
y
C
om
m
uni
c
at
io
ns
and
I
nf
or
m
at
io
n
Sy
s
te
m
s
C
onf
e
r
e
nc
e
(
M
il
C
I
S)
,
N
ov.
2015,
pp.
1
–
6
,
doi
:
10.1109/M
i
lC
I
S
.2015.7348942.
[
27]
M
.
N
a
de
e
m,
A
.
A
r
s
ha
d,
S
.
R
ia
z
,
S
.
S
.
B
a
nd,
a
nd
A
.
M
os
a
vi
,
“
I
nt
e
r
c
e
pt
th
e
c
lo
ud
ne
twor
k
f
r
om
br
ut
e
f
o
r
c
e
a
nd
D
D
oS
a
tt
a
c
ks
vi
a
in
tr
us
io
n
de
te
c
ti
on
a
nd
pr
e
ve
nt
io
n
s
ys
te
m
,”
I
E
E
E
A
c
c
e
s
s
,
vol
.
9,
pp.
152300
–
152309,
2021,
doi
:
10.1109/AC
C
E
S
S
.2021.3126535.
[
28]
G
.
U
ç
tu
,
M
.
A
lk
a
n,
İ
.
A
.
D
oğr
u,
a
nd
M
.
D
ör
te
r
le
r
,
“
A
s
ugg
e
s
te
d
te
s
tb
e
d
to
e
va
lu
a
te
mul
ti
c
a
s
t
n
e
twor
k
a
nd
th
r
e
a
t
pr
e
ve
n
ti
on
pe
r
f
or
ma
nc
e
of
ne
xt
ge
ne
r
a
ti
on
f
ir
e
w
a
ll
s
,”
F
ut
ur
e
G
e
n
e
r
at
io
n
C
om
put
e
r
S
y
s
te
m
s
,
vol
.
124,
pp.
56
–
67,
N
ov.
20
21,
doi
:
10.1016/j
.f
ut
ur
e
.2021.05.013.
[
29]
E
.
P
e
ll
e
gr
in
o
e
t
al
.
,
“
M
a
c
hi
ne
le
a
r
ni
ng
r
a
ndom
f
or
e
s
t
f
or
pr
e
di
c
ti
ng
onc
os
oma
ti
c
v
a
r
ia
nt
N
G
S
a
na
ly
s
is
,”
Sc
ie
nt
if
ic
R
e
po
r
ts
,
vol
. 11, no. 1, Nov. 202
1, doi:
10.1038/s
41598
-
021
-
01253
-
y.
[
30]
M
.
R
odr
íg
ue
z
,
Á
.
A
le
s
a
n
c
o,
L
.
M
e
ha
vi
ll
a
,
a
nd
J
.
G
a
r
c
ía
,
“
E
va
lu
a
ti
on
of
ma
c
hi
ne
l
e
a
r
ni
ng
te
c
hni
que
s
f
or
tr
a
f
f
ic
f
lo
w
-
ba
s
e
d
in
tr
us
io
n de
te
c
ti
on
,”
Se
ns
o
r
s
, vol
. 22, no. 23, Nov. 20
22, doi:
1
0.3390/s
22239326.
[
31]
S
.
Z
ha
ng,
X
.
X
ie
,
a
nd
Y
. X
u,
“
A
br
ut
e
-
f
or
c
e
bl
a
c
k
-
box
me
th
o
d
to
a
tt
a
c
k
ma
c
hi
ne
le
a
r
ni
ng
-
ba
s
e
d s
ys
te
m
s
in
c
yb
e
r
s
e
c
ur
it
y
,”
I
E
E
E
A
c
c
e
s
s
, vol
. 8, pp. 128250
–
128263, 2020, doi:
10.1109/AC
C
E
S
S
.2020.3008433.
[
32]
I
.
F
.
K
il
in
c
e
r
,
F
.
E
r
ta
m,
a
nd
A
.
S
e
ngur
,
“
M
a
c
hi
ne
le
a
r
ni
ng
me
th
ods
f
or
c
ybe
r
s
e
c
ur
it
y
in
tr
us
io
n
de
te
c
ti
on:
d
a
ta
s
e
ts
a
nd
c
ompa
r
a
ti
ve
s
tu
dy,”
C
om
put
e
r
N
e
t
w
or
k
s
, vol
. 188, Apr
. 2021, doi:
10.1016/j
.c
omne
t.
2021.107840.
[
33]
Y
.
Z
ha
ng,
J
.
L
iu
,
a
nd
W
.
S
he
n,
“
A
r
e
vi
e
w
of
e
ns
e
mbl
e
le
a
r
ni
ng
a
lg
or
it
hms
us
e
d
in
r
e
mot
e
s
e
ns
in
g
a
ppl
ic
a
ti
ons
,”
A
pp
li
e
d
Sc
ie
nc
e
s
, vol
. 12, no. 17,
A
ug. 2022, doi:
10.3390/app121786
5
4.
[
34]
M
.
A
ld
ua
il
ij
,
Q
.
W
.
K
ha
n,
M
.
T
a
hi
r
,
M
.
S
a
r
da
r
a
z
,
M
.
A
ld
ua
il
ij
,
a
nd
F
.
M
a
li
k,
“
M
a
c
hi
ne
-
le
a
r
ni
ng
-
ba
s
e
d
D
D
oS
a
tt
a
c
k
de
te
c
t
io
n
us
in
g
mut
ua
l
in
f
or
ma
ti
on
a
nd
r
a
ndom
f
o
r
e
s
t
f
e
a
tu
r
e
i
mpor
ta
nc
e
me
th
od
,”
Sy
m
m
e
tr
y
,
vol
.
14,
no.
6,
M
a
y
2022,
doi
:
10.3390/s
ym14061095.
[
35]
A
.
H
a
ma
r
s
he
h,
“
A
n
a
da
pt
iv
e
s
e
c
ur
it
y
f
r
a
me
w
or
k
f
o
r
in
te
r
ne
t
o
f
th
in
gs
ne
twor
ks
le
ve
r
a
gi
ng
S
D
N
a
nd
ma
c
hi
ne
le
a
r
ni
ng
,”
A
pp
li
e
d
Sc
ie
nc
e
s
, vol
. 14, no. 11,
M
a
y 2024, doi:
10.3390/app141145
30
.
[
36]
B
.
A
.
T
a
ma
a
nd
S
.
L
im
,
“
E
ns
e
mbl
e
le
a
r
ni
ng
f
or
in
tr
us
io
n
de
te
c
ti
on
s
ys
t
e
ms
:
a
s
ys
t
e
ma
ti
c
ma
ppi
ng
s
tu
dy
a
nd
c
r
os
s
-
be
nc
h
ma
r
k
e
va
lu
a
ti
on,”
C
om
put
e
r
Sc
ie
nc
e
R
e
v
ie
w
, vol
. 39, F
e
b. 2021, doi:
10.1016/j
.c
os
r
e
v.2020.100357.
B
I
OG
RA
P
HI
E
S
OF
AU
T
HO
RS
A
hm
a
d
Tur
m
udi
Z
y
i
s
a
l
ec
t
u
r
er
w
i
t
h
e
x
p
er
t
i
s
e
i
n
s
ecu
ri
t
y
s
y
s
t
em
s
,
s
en
t
i
me
n
t
an
al
y
s
i
s
,
co
mp
u
t
er
n
e
t
w
o
rk
i
n
g
,
an
d
ap
p
l
i
e
d
s
y
s
t
em
s
.
H
e
h
o
l
d
s
a
Mas
t
er’s
d
e
g
ree
fro
m
Pres
i
d
en
t
U
n
i
v
er
s
i
t
y
a
n
d
i
s
cu
rre
n
t
l
y
affi
l
i
a
t
e
d
w
i
t
h
U
n
i
v
ers
o
t
a
s
Pel
i
t
a
Ban
g
s
a
.
H
i
s
res
earc
h
fo
cu
s
es
o
n
en
h
a
n
ci
n
g
t
h
e
s
ecu
r
i
t
y
an
d
effi
ci
e
n
cy
o
f
n
e
t
w
o
r
k
ed
s
y
s
t
ems
an
d
t
h
e
ap
p
l
i
cat
i
o
n
o
f
s
en
t
i
me
n
t
an
a
l
y
s
i
s
i
n
v
a
ri
o
u
s
d
o
ma
i
n
s
.
H
e
can
b
e
co
n
t
act
e
d
at
emai
l
:
t
u
rm
u
d
i
@
p
el
i
t
a
b
an
g
s
a.
ac.
i
d
.
I
s
a
ri
a
nt
o
i
s
a
l
ect
u
rer
i
n
U
n
i
v
ers
i
t
a
s
Pel
i
t
a
Ban
g
s
a
fo
r
In
fo
rma
t
i
o
n
T
ech
n
o
l
o
g
y
fro
m
2
0
2
2
.
Mas
t
er
D
eg
ree
fro
m
Pres
i
d
e
n
t
U
n
i
v
ers
i
t
y
in
Mas
t
er
Sci
en
ce
o
f
I
n
fo
rma
t
i
o
n
T
ech
n
o
l
o
g
y
(M.
Sc.
)
i
n
2
0
1
9
.
H
i
s
res
earc
h
i
n
t
ere
s
t
s
are
in
i
n
fo
rm
at
i
o
n
ret
r
i
ev
a
l
,
art
i
f
i
ci
a
l
i
n
t
el
l
i
g
en
t
,
i
n
f
o
rmat
i
o
n
s
y
s
t
em,
n
e
t
w
o
rk
s
ecu
r
i
t
y
,
an
d
c
o
mp
u
t
er
n
et
w
o
r
k
.
H
e
ca
n
b
e
c
o
n
t
act
e
d
at
emai
l
:
i
s
ari
a
n
t
o
@
p
el
i
t
a
b
an
g
s
a.
ac.
i
d
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14
,
No.
4
,
Augus
t
20
25
:
294
5
-
2954
2954
A
ng
g
i
M
uha
m
m
a
d
R
i
fa
'
i
i
s
a
d
ed
i
cat
e
d
l
ec
t
u
rer
i
n
t
h
e
I
n
fo
rma
t
i
c
s
E
n
g
i
n
eer
i
n
g
p
ro
g
ram
at
U
n
i
v
ers
o
t
a
s
Pel
i
t
a
Ban
g
s
a
,
w
i
t
h
e
x
p
er
t
i
s
e
i
n
a
rt
i
fi
c
i
al
i
n
t
el
l
i
g
en
ce
,
ear
n
ed
a
Mas
t
er's
d
e
g
ree
fro
m
A
MIK
O
M
U
n
i
v
er
s
i
t
y
Y
o
g
y
ak
a
rt
a.
H
i
s
res
earc
h
areas
i
n
c
l
u
d
e
d
at
a
s
ci
e
n
ce,
mach
i
n
e
l
earn
i
n
g
,
d
eep
l
ear
n
i
n
g
,
i
ma
g
e
p
ro
ce
s
s
i
n
g
,
an
d
c
o
mp
u
t
er
v
i
s
i
o
n
,
p
art
i
cu
l
arl
y
i
n
t
h
e
f
i
el
d
o
f
b
i
o
i
n
fo
rma
t
i
c
s
.
H
e
h
a
s
p
u
b
l
i
s
h
e
d
s
e
v
era
l
p
a
p
ers
i
n
i
n
t
ern
a
t
i
o
n
a
l
j
o
u
r
n
al
s
an
d
co
n
fere
n
ces
an
d
i
s
cu
rren
t
l
y
a
res
earch
er
at
U
n
i
v
er
s
o
t
as
Pel
i
t
a
Ban
g
s
a
.
H
e
can
b
e
co
n
t
ac
t
ed
a
t
emai
l
:
an
g
g
i
m
u
h
amma
d
@
p
el
i
t
a
b
an
g
s
a.
ac.
i
d
.
A
bdul
Gho
fi
r
i
s
a
m
as
t
er
an
d
b
ach
e
l
o
r
d
eg
ree
i
n
In
fo
rmat
i
o
n
T
ech
n
o
l
o
g
y
fro
m
Pres
i
d
en
t
U
n
i
v
er
s
i
t
y
,
In
d
o
n
es
i
a
,
i
n
2
0
1
5
an
d
2
0
1
0
.
Cu
rren
t
l
y
,
h
e
i
s
t
each
i
n
g
at
Pres
i
d
e
n
t
U
n
i
v
er
s
i
t
y
fo
r
c
o
mp
u
t
er
n
et
w
o
r
k
,
n
et
w
o
r
k
s
ec
u
ri
t
y
,
a
n
d
s
o
c
k
et
p
ro
g
rammi
n
g
.
T
h
e
res
earc
h
w
as
ab
o
u
t
cach
e
d
i
s
t
ri
b
u
t
i
o
n
i
n
s
q
u
i
d
p
ro
x
y
s
er
v
er,
n
et
w
o
r
k
d
efen
s
e
s
y
s
t
em
mo
n
i
t
o
r
i
n
g
,
d
i
g
i
t
al
i
mag
e
s
ecu
re
d
,
an
d
s
mar
t
el
ec
t
ri
c
i
t
y
me
t
er,
s
i
n
c
e
2
0
1
5
.
H
e
can
b
e
co
n
t
ac
t
ed
a
t
ema
i
l
:
g
eo
ff@
p
re
s
i
d
en
t
.
ac.
i
d
.
M
uha
m
m
a
d
N
a
j
a
m
uddi
n
D
w
i
M
i
h
a
rja
i
s
an
i
n
f
o
r
mat
i
c
s
en
t
h
u
s
i
a
s
t
w
i
t
h
a
fo
cu
s
on
me
d
i
ca
l
i
n
fo
rma
t
i
c
s
.
H
e
b
eg
a
n
h
i
s
acad
em
i
c
j
o
u
r
n
ey
at
U
n
i
v
er
s
i
t
as
I
s
l
am
In
d
o
n
es
i
a
i
n
2
0
1
4
,
s
p
eci
a
l
i
z
i
n
g
i
n
me
d
i
ca
l
i
n
fo
rma
t
i
c
s
.
Cu
rren
t
l
y
,
h
e
i
s
p
u
rs
u
i
n
g
h
i
s
u
n
d
erg
ra
d
u
a
t
e
d
e
g
ree
i
n
In
fo
rma
t
i
c
s
at
In
s
t
i
t
u
t
T
ek
n
o
l
o
g
i
A
d
h
i
T
ama
Su
ra
b
ay
a
(IT
A
T
S).
He
i
s
p
a
s
s
i
o
n
at
e
ab
o
u
t
i
n
t
eg
ra
t
i
n
g
t
ech
n
o
l
o
g
y
i
n
t
o
h
eal
t
h
care
an
d
i
s
al
w
ay
s
eag
er
t
o
ex
p
l
o
re
n
ew
ad
v
a
n
cemen
t
s
i
n
t
h
e
fi
el
d
.
H
e
can
b
e
co
n
t
ac
t
ed
a
t
emai
l
:
n
a
j
amu
d
d
i
n
d
.
d
w
i
@
g
mai
l
.
co
m.
A
na
nto
Tr
i
Sa
s
o
n
g
ko
h
o
l
d
s
a
D
o
c
t
o
r
o
f
Co
m
p
u
t
er
Sci
en
ce
fr
o
m
U
n
i
v
e
rs
i
t
a
s
In
d
o
n
e
s
i
a,
In
d
o
n
e
s
i
a
,
i
n
2
0
2
2
.
H
e
recei
v
ed
h
i
s
M.
Sc.
(In
fo
rmat
i
o
n
Man
ag
eme
n
t
)
fro
m
t
h
e
G
eo
rg
e
W
as
h
i
n
g
t
o
n
U
n
i
v
ers
i
t
y
,
W
as
h
i
n
g
t
o
n
,
D
C,
U
SA
,
i
n
1
9
9
3
.
H
e
al
s
o
g
o
t
a
B.
E
n
g
.
(E
l
ec
t
ri
ca
l
E
n
g
i
n
eer
i
n
g
)
i
n
1
9
9
0
fro
m
U
n
i
v
er
s
i
t
as
G
ad
j
ah
Mad
a,
In
d
o
n
es
i
a.
H
e
h
as
c
u
rren
t
l
y
an
A
s
s
i
s
t
a
n
t
Pr
o
fe
s
s
o
r
at
I
n
fo
rma
t
i
c
s
E
n
g
i
n
eer
i
n
g
at
U
n
i
v
er
s
i
t
as
Pel
i
t
a
Ba
n
g
s
a,
In
d
o
n
e
s
i
a,
s
i
n
ce
2
0
1
5
.
H
i
s
re
s
earch
i
n
cl
u
d
e
s
w
i
rel
e
s
s
a
d
h
o
c
n
et
w
o
r
k
s
,
art
i
fi
c
i
a
l
i
n
t
e
l
l
i
g
e
n
ce,
mach
i
n
e
l
earn
i
n
g
,
d
a
t
a
s
c
i
en
ce,
a
n
d
i
n
fo
rma
t
i
o
n
s
y
s
t
ems
.
H
e
ca
n
b
e
co
n
t
ac
t
ed
a
t
ema
i
l
:
an
an
t
o
@
p
e
l
i
t
ab
a
n
g
s
a.
ac.
i
d
.
Evaluation Warning : The document was created with Spire.PDF for Python.