Indonesi
an
Journa
l
of El
ect
ri
cal Engineer
ing
an
d
Comp
ut
er
Scie
nce
Vo
l.
24
,
No.
1
,
Octo
be
r
2021
,
pp.
59
0
~
59
9
IS
S
N: 25
02
-
4752, DO
I: 10
.11
591/ijeecs
.v
24
.i
1
.
pp
590
-
59
9
590
Journ
al h
om
e
page
:
http:
//
ij
eecs.i
aesc
or
e.c
om
Optimi
zed ma
ch
ine learn
ing alg
orithm
for in
trusion
detecti
on
Royid
a A.
I
br
ah
em
A
lh
ayal
i
1
, Mohamm
ad Alj
anabi
2
,
Ahmed
Hussein
A
li
3
,
Most
afa Abd
u
lghfo
or Mo
hamm
ed
4
, T
ole
Sut
ikn
o
5
1
Depa
rtment of
Com
pute
r
Engi
n
ee
ring
,
Co
ll
eg
e of
Engi
n
ee
rin
g,
Univer
sit
y
of
Di
y
a
la,
Di
y
a
la,
Ir
a
q
2,3
Depa
rtment
of
computer
,
Col
lege
of Educ
at
ion
,
A
l
-
Ira
qi
a
Univ
e
rsit
y
,
B
aghda
d
,
I
raq
2,3
Depa
rtment
of
Com
pute
r
Sci
en
ce
,
A
l
Sa
la
m
Un
ive
rsit
y
Coll
ege,
Baghda
d
,
Ir
a
q
4
Im
am Aadha
m
Univer
sit
y
Co
ll
e
ge,
B
aghda
d
,
Ir
a
q
5
Depa
rtment of
El
e
ct
i
ca
l
Eng
ineeri
ng,
Univer
si
tas
Ahm
ad
Dahla
n
,
Yog
y
ak
art
a
,
In
donesia
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
J
ul
9
,
2021
Re
vised
Sep
2
,
2021
Accepte
d
Se
p
4
,
2021
Intrusion
detec
t
i
on
is
m
ai
nl
y
a
ch
ie
ved
b
y
using
o
pti
m
iz
ation
a
lgo
rit
hm
s.
The
nee
d
for
opt
imiz
at
ion
al
gori
thms
for
int
rusion
de
t
ec
t
ion
is
necess
itate
d
b
y
the
inc
re
asing
num
ber
of
feature
s
in
audi
t
d
ata,
a
s
well
as
the
p
erf
orm
anc
e
fai
lur
e
of the
hu
m
an
-
base
d
sm
art
int
rusion
d
et
e
cti
on
s
y
stem (IDS
) in t
erms
of
the
ir
prolong
ed
tra
ini
ng
ti
m
e
and
cl
assificat
ion
ac
cur
acy
.
Thi
s
art
icle
pre
sents
an
improved
int
rusion
d
et
e
ct
ion
t
ec
hn
iq
ue
for
bina
r
y
class
ifi
cation.
The
proposal
i
s
a
combinat
ion
of
diffe
ren
t
opti
m
iz
ers,
in
cluding
Rao
opti
m
iz
ation
al
g
orit
hm
,
ex
tre
m
e
le
a
rning
m
ac
h
i
ne
(E
LM),
sup
port
ve
ct
or
m
ac
hine
(SV
M),
and
logi
sti
c
reg
ression
(L
R)
(for
feature
sele
c
ti
on
&
weight
ing)
,
as
well
as
a
h
y
brid
R
ao
-
SV
M
al
gorit
hm
with
supervis
ed
m
ac
hine
le
arn
ing
(
ML
)
t
ec
hniqu
es
for
fe
at
ure
subs
et
se
lecti
on
(FS
S).
The
proc
ess
of
sele
c
ti
ng
the
lea
st
num
ber
of
fe
at
ure
s
wi
thout
s
ac
rif
ic
ing
the
FS
S
ac
cur
a
c
y
was
conside
red
a
m
ult
i
-
ob
je
c
tive
opti
m
izati
on
proble
m
.
Th
e
al
gori
thm
-
spec
ific,
par
ame
te
r
-
l
ess c
once
pt of t
he
proposed
Rao
-
SV
M was a
lso e
xplore
d
in
thi
s
stud
y
.
Th
e
KD
DCup
99
a
nd
CICIDS
2017
were
used
as
t
he
intrus
ion
dat
ase
t
for
the
e
xper
iments,
whe
re
significant
improvem
ent
s
were
note
d
wi
th
the
new
Rao
-
SV
M
compare
d
to
the
othe
r
al
go
rit
hm
s.
Rao
-
SVM
pre
sente
d
bet
t
er
result
s
th
an
m
an
y
exi
st
i
ng
works
by
re
ac
hing
100%
a
cc
ura
c
y
for
KD
DCup 99
dataset
and
97%
for
CICIDS
dat
as
et
.
Ke
yw
or
ds:
Extrem
e lea
rn
ing m
achine
Feat
ur
e
subset
sel
ect
ion
In
tr
us
i
on d
et
ec
ti
on
syst
em
Lo
gisti
c regres
sion
Ma
chine
le
a
rn
i
ng
Ra
o op
ti
m
iz
a
tio
n al
gorithm
S
upport
v
ect
or m
achine
This
is an
open
acc
ess arti
cl
e
un
der
the
CC
B
Y
-
SA
l
ic
ense
.
Corres
pond
in
g
Aut
h
or
:
Ah
m
ed
H
us
sei
n Ali
Dep
a
rtm
ent o
f C
om
pu
te
r
Scie
nce
A
l
Sala
m
U
nive
rsity
Colle
ge
Ba
ghda
d
,
9818
2,
Ir
a
q
Em
a
il
:
m
sc.ahme
d.
h.
al
i@
gma
il
.co
m
1.
INTROD
U
CTION
The
nee
d
for
netw
ork
i
nform
at
ion
secu
rity
has
i
ncr
ease
d
recently
due
to
th
e
a
dv
a
nc
e
m
ents
a
nd
popula
rizat
ion
of
in
f
or
m
at
ion
a
nd
netw
ork
te
ch
no
l
og
ie
s
[1
]
-
[12]
.
H
uma
n
-
base
d
IDSs
can
ei
ther
w
arn
or
intercept
netw
ork
int
ru
si
on, b
ut this is
not t
he
case f
or the t
rad
it
io
nal n
et
w
ork defe
ns
e m
echan
ism
s.
The foc
us
of
m
os
t
stud
ie
s
in
this
fiel
d
has
bee
n
on
the
i
m
pr
ovem
e
nt
of
the
perform
ance
of
sm
art
netw
ork
i
nt
ru
si
on
detect
ion
syst
e
m
s
(
I
DS
s
)
[
13
]
-
[
17
]
as
th
ey
are
c
onsid
ered
an
ef
fecti
ve
s
olu
ti
on
t
o
netw
ork
s
e
cur
it
y.
Con
si
der
i
ng
th
e
lo
w
detect
ion
rate
(D
R
)
of
the
e
xisti
ng
I
D
Ss
in
the
prese
nce
of
new
at
ta
cks,
co
uple
d
w
it
h
th
e
high
over
hea
d
associat
ed
with
au
dit
data,
ef
forts
are
bein
g
channele
d
t
o
m
achine
le
ar
ning
-
base
d
m
et
ho
ds
a
nd
op
ti
m
iz
ation
al
gorithm
s f
or
ne
twork
intr
us
io
n detec
ti
on
[
18]
-
[29]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Op
ti
mize
d m
ac
hin
e le
arnin
g a
lgo
rit
hm f
or
in
trusio
n detec
ti
on
(
Royid
a A.
I
br
ahe
m
Al
ha
y
al
i
)
591
This
era
of
big
data
is
ass
ocia
te
d
with
issues
on
the
secu
rity
of
netw
ork
s
yst
e
m
.
In
tru
sio
n
detect
io
n
is
receiving
m
uch
at
te
ntio
n
du
e
to
the
nee
d
for
bette
r
se
cur
it
y
of
net
w
o
r
k
inf
rastr
uct
ur
e
in
rece
nt
tim
es.
Diff
e
re
nt
m
ac
hin
e
le
a
rn
i
ng
(
ML
)
m
et
ho
ds
hav
e
bee
n
c
om
bin
ed
with
optim
iz
at
ion
al
gorithm
s
fo
r
e
ff
ic
ie
nt
intru
si
on
detec
ti
on
;
f
or
insta
nc
e,
s
om
e
of
th
e
cu
rr
e
nt
com
bin
at
io
ns
i
nclu
de
fu
zzy
l
og
ic
,
k
-
nea
rest
neig
hbors
(
K
NN
)
,
a
rt
ific
ia
l
neural
netw
ork
(
A
NN
)
,
pa
rtic
le
swar
m
op
tim
iz
at
ion
(
P
SO
)
,
s
upport
ve
ct
or
m
achine
(
SV
M
)
,
Cutt
le
fish
opti
m
iz
at
ion
al
go
r
it
h
m
,
and
arti
f
ic
ia
l
i
m
m
un
e
syst
e
m
(A
IM)
app
r
oac
hes
[
30]
-
[
36]
.
Most
of
the
appr
oach
es
tha
t
com
bin
e
ML
with
op
ti
m
iz
a
ti
on
al
gorit
hms
ha
ve
s
how
n
bette
r
pe
r
form
ance
as
c
om
par
ed
t
o
the
tra
diti
on
al
cl
assifi
cat
ion
te
chn
i
ques
[
37
]
-
[
40
]
.
Va
riou
s
ML
a
nd
op
ti
m
iz
ation
-
ba
sed
IDS
have
bee
n
pro
po
se
d
i
n
t
he
li
te
ratur
e
[
41
]
-
[
43
]
;
f
or
instance
,
a
c
om
bin
at
ion
of
K
-
m
eans
c
lust
erin
g,
NB,
C
4.5,
an
d
Krus
kal
-
Wall
is
has
been
pr
opos
e
d
by
Louvi
eris
et
al.
[44]
fo
r
the
detect
ion
of
net
work
intru
s
i
on
with
high
accuracy.
This
appr
oach
c
an
be
us
ed
t
o
cl
assify
releva
nt
featu
re
set
s
a
s
it
includes
a
sta
ti
sti
cal
tool
fo
r
validit
y.
A
stud
y
by
Čr
epinše
k
et
al.
[45]
pro
po
se
d
a
sel
f
-
orga
nizi
ng
m
ap
(S
OM
)
an
d
pr
i
ncipal
com
po
nen
t
analy
sis
(
PCA
)
-
base
d
I
DS
th
at
filt
ers
no
ise
in
dataset
and
lo
w
-
va
riance
featur
e
s
us
i
ng
the
PCA
a
nd
f
isher
discrim
inant
rati
o
(F
DR
).
It
reli
es
on
t
he
m
os
t
discri
m
in
at
ive
pro
j
ect
io
ns
that
do
no
t
dep
e
nd
so
le
ly
on
t
he
exp
la
ine
d
va
riance
by
the
prototypes
of
the
ei
genvecto
rs
ge
ner
at
e
d
by
th
e
SO
M
proces
s.
H
oweve
r,
t
he
m
aj
or
pro
blem
of
this
syst
e
m
is
low
detect
ion
r
at
e
(
DR
)
[
15]
.
A
ti
m
e
-
var
yi
ng
ch
aos
-
PS
O
m
et
h
od
h
as b
een p
r
ov
i
de
d
by
Bam
akan
[
14]
as
a
new
M
L
-
ba
sed
IDS;
the
pro
pose
d
m
et
hod
is
base
d
on
2
c
onve
ntio
nal
optim
iz
ers
wh
ic
h
are
m
ult
iple
crit
eria
li
near
program
m
ing
(
MC
LP)
an
d
S
VM.
The
par
a
m
et
ers
of
thes
e
op
ti
m
iz
ers
wer
e
set
us
in
g
the
pro
pose
d
m
et
ho
d;
it
was
al
so
us
e
d
to
sel
ect
the
m
os
t
ap
pro
pr
ia
te
featu
re
s
ub
set
s
[
46
]
-
[49]
.
T
he
only
pro
blem
of
this
new
m
et
ho
d
is
the
pr
ol
onge
d
trai
ning
pe
r
iod
w
hich
nee
ds
to
be
im
pr
oved
.
Alth
ough
these
com
bin
at
ion
s
c
an
im
pr
ove
I
D
S
pe
rfor
m
ance
in
te
rm
s
of
D
R
and
le
ar
ning
sp
ee
d,
f
ur
the
r
i
m
pr
ovem
ent
is
sti
l
l
need
e
d
[
50]
.
Ra
o
an
d
Fate
l
et
al.
[51]
pro
posed
R
ao
-
S
VM
al
gorithm
that
requires
no
us
e
r
-
def
i
ned
par
a
m
et
er
durin
g
the
op
ti
m
iz
at
ion
proce
ss
for
m
echan
ic
al
desig
n
pro
bl
e
m
s.
The
ne
w
m
et
ho
d
was
e
valuated
on
di
f
fer
e
nt
ben
c
hm
ark
f
unct
ion
s
a
nd
f
ou
nd
t
o
pe
rfo
rm
bette
r
tha
n
som
e
pf
the
e
xisti
ng
o
ne
s.
T
he
po
te
ntial
of
a ne
w
R
ao
-
SV
M
al
gorith
m
in
opti
m
a
l
fr
ee
par
am
et
ers
sel
ect
ion
f
or
SV
M
regressi
on
m
od
el
s
has
been
re
porte
d
by
[
30]
us
in
g
m
ulti
-
com
m
od
it
y
fu
tur
es
ind
e
x
data
retrieve
d
from
m
ult
i
-
cut
cr
osso
ver
(MCX
)
.
From
the
re
su
lt
s,
SV
M
-
R
ao
-
SVM
p
er
f
or
m
ed
well
in fin
ding
the opti
m
al
p
ara
m
et
ers
com
par
ed
to
t
he
cl
ass
ic
al
SV
M.
The
hybr
i
d
S
V
M
-
R
a
o
-
SV
M m
od
el
was
pr
e
sented
by
Das
et
al.
[52]
via
i
ntr
oductio
n
of a
dim
ension
-
reducti
on
a
ppr
oach
that
al
lo
ws
t
he
re
duct
ion
of
t
he
nu
m
ber
of
in
put
va
riables
us
in
g
PCA,
ke
rn
el
pri
ncipal
com
po
ne
nt
an
al
ysi
s
(
KP
CA
)
,
an
d
in
de
pend
ent
com
ponen
t
analy
sis
(I
C
A
).
T
he
st
ud
y
a
lso
in
vestigat
e
d
the
po
s
sibil
it
y
of
us
ing
the
m
ulti
-
com
m
od
it
y
fu
tures
in
de
x
data
extracte
d
fro
m
the
M
CX
in
the
pr
op
os
ed
m
od
el
.
The
perform
ance
of
the
pro
pose
d
m
od
el
w
as
co
nf
irm
ed
to
be
s
up
e
rio
r
t
o
that
of
so
m
e
existi
ng
po
pu
l
at
ion
-
insp
ire
d
m
od
el
s.
I
n
a
no
t
her
stud
y,
the
e
ff
ect
of
num
ber
of
ge
ner
at
io
ns
a
nd
sam
ple
siz
e
on
the
perform
ance
of
op
ti
m
iz
ation
fra
m
ewo
r
ks
was
evaluated
by
Ra
o
and
Pate
l
[50]
wh
il
e
Cre
pin
še
k
et
al.
[45]
fo
c
us
ed
on
us
in
g
R
ao
-
S
VM
to
so
lve
the
e
xa
ct
pro
blem
s
hig
hlig
hted
i
n
[42]
an
d
[53]
.
A
m
ulti
-
obj
e
ct
ive
R
ao
-
SVM
was
dev
el
op
e
d
by
Nayak
a
nd
R
out
[47]
w
her
e
a
m
at
rix
of
so
l
utions
was
de
velo
ped
for
ea
ch
ob
j
ect
ive.
F
or
the
R
ao
-
S
VM
m
od
el
,
t
he
te
ache
r
sel
ect
io
n
pro
cess
wa
s
reli
a
nt
on
the
best
-
fou
nd
so
l
ution
in
t
he
sea
rch
sp
ac
e
wh
il
e
t
he
le
ar
ne
rs
a
re
ta
ught
just
to
m
axi
m
i
z
e
that
obj
ect
iv
e.
T
he
a
vaila
bl
e
so
l
utions
in
t
he
sea
rc
h
s
pac
e
we
re
arr
a
ng
e
s
to
ar
r
ive
at
a
set
of
op
ti
m
al
so
luti
on
s.
T
he
stu
dy
by
Shu
kla
et
al.
[54]
reli
ed
on
dif
fer
e
nt
te
a
chin
g
m
et
ho
ds
t
o
pr
esent
a
m
ulti
-
obj
ect
ive
R
ao
-
SV
M
in
wh
ic
h
the
c
ro
ss
ove
r
operat
or
(in
s
te
ad
of
us
i
ng
a
scal
er
functi
on)
was
util
iz
ed
in
-
between
s
olu
ti
on
s
in
the
te
achin
g
&
le
ar
ning
phases.
Kizi
loz
et
al.
[40]
pres
ented
3
m
ul
ti
-
obj
ect
iv
e
R
ao
-
S
VM
fr
a
m
ewo
r
ks
f
or
F
SS
-
BC
P.
Am
on
g
th
e
pr
e
sent
ed
m
et
ho
ds
,
a
m
ul
ti
-
obj
ect
ive
R
ao
-
SV
M wit
h
scal
ar tran
s
form
at
i
on
(
MR
ao
-
SVM
-
ST) was the
f
ast
est
d
espite
p
rovi
ding a li
m
it
ed
nu
m
ber
o
f
no
n
-
do
m
inate
d
so
l
utions.
Re
ga
rdi
ng
the
m
ulti
-
obj
ect
ive
R
ao
-
SV
M
with
non
-
dom
inate
d
sel
ect
ion
(MR
ao
-
S
VM
-
NS
)
,
it
searc
hes
the
so
l
ution
sp
ace
,
ge
ner
at
e
a
set
of
non
-
dom
inate
d
so
l
utions
,
an
d
re
quire
s
m
uch
i
m
ple
m
entat
io
n
ti
m
e.
Mult
i
-
ob
j
ect
ive
R
ao
-
S
VM
with
m
ini
m
u
m
dist
ance
(MR
ao
-
SV
M
-
M
D)
ge
ner
at
e
s
si
m
il
ar
so
luti
on
s
to
that
of
MR
ao
-
S
VM
-
NS
;
ye
t,
in
a
sign
if
ic
antly
le
sser
a
m
ou
nt
of
tim
e.
The
propose
d
R
ao
-
SV
M
wer
e
eva
luate
d
for
perf
or
m
ance
us
i
ng
LR,
S
VM,
a
nd
ELM
.
S
ultan
a
an
d
Ja
bbar
[
55
]
sta
te
d
that
FSS
i
n
the W
ra
pper
m
et
hod
is
m
ade
as
a
black
box,
m
eaning
that
there
is
no
kn
owle
dge
of
the
unde
rly
ing
al
gorithm
.
The
sel
ect
io
n
of
f
eat
ure
s
ubs
et
s
is
done
us
i
ng
in
du
ct
ive
al
gorithm
s
and
t
he
sel
ect
ed
fea
ture
s
ubset
are
us
e
d
to
est
i
m
at
e
the
accuracy
of
the
trai
ning
m
od
e
l.
The
acc
urac
y
le
vel
will
gu
ide
the
m
od
el
in
deci
ding
w
hethe
r
featur
e
s can
be
ad
de
d
or
rem
ov
ed
f
r
om
the sele
ct
ed
su
bs
et
. Hence
, th
e
Wrapp
e
r
m
et
ho
ds
are
co
ns
ide
red
m
or
e
com
pu
ta
ti
on
al
ly
com
plex
[
56]
-
[58]
.
T
he
Fil
te
r
m
et
ho
d
is
ano
t
her
m
et
ho
d
wh
e
re
t
he
m
od
el
i
niti
at
es
with
al
l
the
avail
able
f
eat
ur
es
be
fore
sel
ect
ing
t
he
best
feat
ur
e
subset
ba
sed
on
certai
n
sta
ti
sti
cal
m
et
rics,
suc
h
as
Pears
on’s
co
rrel
at
ion
[59]
,
A
NOV
A,
LD
A,
Chi
sq
ua
re,
a
nd
m
utu
al
info
r
m
at
ion
[60],
[
61
]
.
T
hese
sta
ti
sti
cal
m
et
rics
are
ba
s
ed
on
the f
eat
ure
a
nd
res
ponse
va
riables
i
n
t
he
dataset
;
ho
wev
e
r,
the
c
om
m
on
ly
us
ed
s
ta
ti
sti
ca
l
m
et
rics are
pea
rson’s
c
orrelat
ion (PC
)
a
nd
m
utu
al
Inform
at
i
on m
et
ho
ds
[61]
-
[
64]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
24
, N
o.
1
,
Oct
ober
20
21
:
590
-
59
9
592
Most
of
the
f
eat
ur
e
sel
ect
io
n
m
et
ho
ds
ea
r
li
er
discusse
d
are
de
pe
nd
e
nt
on
featu
re
s
ubset
at
the
pr
e
processi
ng
le
vel;
hen
ce,
t
he
m
et
ho
ds
to
be
disc
us
se
d
her
e
a
re
the
e
m
bed
ded
m
et
ho
ds
a
s
they
w
ork
in
a
way
al
lows
th
e
sel
ect
ion
of
the
best
feat
ur
es
durin
g
th
e
le
arn
in
g
phase
[65]
-
[
67]
.
The
ad
va
nta
ges
of
inco
rpor
at
in
g
the
featu
re
sel
ect
ion
process
in
t
he
le
ar
ning
process
incl
ud
e
im
pr
ov
e
d
com
pu
ta
ti
on
a
l
cost
,
bette
r
cl
assifi
c
at
ion
acc
ur
acy
,
an
d
a
vo
i
dan
ce
of
the
nee
d
to
retr
ai
n
m
od
el
s
w
hen
e
ve
r
ne
w
featu
res
a
re
a
dde
d.
The
em
bed
de
d
m
et
ho
d
perfor
m
s
featur
e
s
ubset
sel
ect
ion
,
a
nd
t
he
le
ar
ning
al
gorithm
interact
s
in
a
di
fferent
m
ann
er
com
par
ed
t
o
oth
e
r
f
eat
ur
e
sel
ect
io
n
m
et
ho
ds.
Fil
te
r
-
ba
sed
le
arni
ng
al
gorithm
s
are
no
t
c
omm
on
ly
e
m
p
loye
d
f
or
featur
e
sel
ect
ion,
but
the
Wrapp
e
r
m
et
ho
d
te
sts
the
qual
it
y
of
t
he
sel
ect
ed
feat
ur
es
us
i
ng
t
he
le
arn
in
g
al
gorithm
[68]
.
F
or
the
e
m
bed
de
d
m
et
ho
d,
it
ad
dresses
t
he
iss
ue
of
com
pu
ta
ti
on
al
c
om
plexity
as
it
perform
s
the
appr
opriat
e
m
o
del
le
ar
ning
a
nd
featu
re
sel
e
ct
ion
at
the
sa
m
e
tim
e;
featur
e
sel
ect
ion
is
done
durin
g
t
he
m
od
el
trai
ni
ng
st
age,
t
her
e
by
r
e
du
ci
ng
the
c
om
pu
ta
ti
on
al
co
st
com
par
ed
t
o
that
of
t
he
W
rapper
m
et
ho
d.
Wh
e
n
the
acc
ur
acy
of
detec
ti
on
is
inc
reas
ed,
the
exec
ut
ion
ti
m
e
will
so
m
et
i
m
es
increase
by
a
su
bst
antia
l
am
ount.
O
n
t
he
oth
er
ha
nd,
t
her
e
m
a
y
be
reduct
ion
i
n
t
he
e
xec
ution
ti
m
e,
le
a
ding
t
o
lo
w
ac
cur
acy
.
Hen
ce
,
the
FS
S
prob
le
m
can
be
seen
as
a
m
ul
ti
-
obj
ect
ive
op
ti
m
iz
at
ion
(MOO)
pro
blem
that
req
uire
s
m
or
e
than
one
so
l
ution
[
45
]
,
[52],
[
69
]
.
For
s
om
e,
accuracy
is
ve
ry
i
m
po
rta
nt;
the
s
olu
ti
on
tha
t
of
fe
rs
acc
ura
cy
is
chosen
.
Me
an
wh
il
e,
f
or
ot
he
rs,
the
best
s
olu
ti
on
is
th
e
on
e
that
reduces
t
he
exec
utio
n
ti
m
e
even
if
acc
ur
acy
is
com
pr
om
ise
d.
R
ao
was
de
vel
op
e
d
as
a
new
m
et
aheu
risti
c
for
var
i
ous
intr
act
able
op
ti
m
i
zat
ion
pro
blem
s
and
has
perform
ed
well
in
su
c
h
app
li
cat
io
ns
c
om
par
ed
to
ot
her
fr
am
eworks,
su
c
h
as
ge
ne
ti
c
al
go
rith
m
s
(
GA
)
,
PSO,
an
d
a
nt
c
olony
opti
m
iz
a
ti
on
(
ACO
)
.
T
he
com
bin
at
io
n
of
the
ne
w
m
ulti
-
obj
ect
ive
R
ao
-
S
VM
f
ra
m
ewo
r
k
with
s
uper
vise
d
m
achine
le
ar
ning
(ML)
te
c
hn
i
qu
e
s
is
pro
po
s
ed
in
t
his
pa
per
f
or
FS
S
i
n
b
ina
ry
cl
assifi
cat
ion
pro
blem
s.
W
hi
le
try
ing
to
s
el
ect
the
le
ast
nu
m
ber
of
f
eat
ur
es
with
out
i
m
pacti
ng
th
e
accuracy
of
FSS
pro
blem
s,
the
f
irst
obj
ect
i
ve
s
hould
be
the
se
le
ct
ion
of
t
he
r
igh
t
nu
m
ber
of
featu
res,
w
hile
the
sec
ond
c
on
ce
r
n
sh
oul
d
be
the
accuracy
of
th
e
detect
ion
.
T
he
pe
rfor
m
ance
of
R
ao
-
SVM
has
been
r
e
po
rted
as
rem
ark
a
ble
wh
e
n
c
om
par
e
d
to
ot
her
m
e
ta
heurist
ic
s
al
gorithm
s.
The
ne
w
te
ac
hing
-
le
arn
i
ng
-
base
d
op
ti
m
iz
ation
(R
ao
-
SV
M)
an
d
a
s
et
of
s
up
e
rv
ise
d
ML
te
ch
niques
wer
e
em
plo
ye
d
f
or
opti
m
al
featur
es
s
ub
s
et
sel
ect
ion
in
t
his
stud
y.
This
w
ork
c
on
t
rib
utes
the
f
ollo
wing
to
li
te
ratur
e:
i)
the
util
iz
at
ion
of
the
R
ao
-
S
VM
al
gorithm
for
featur
e
sel
ect
ion
in
I
DS
for
th
e
first
tim
e;
ii
)
the
ne
w
R
ao
-
S
VM
al
gorithm
pro
po
se
d
in
t
his
stud
y.
T
he
re
st
of
this
arti
cl
e
is
a
rr
a
ng
e
d
as
f
ollow
s:
intr
oducti
on
of
the
F
SS
pro
blem
pr
es
e
nted
in
s
ect
io
n
2
;
the
propose
d
R
ao
-
SV
M
prese
nted
in
s
ect
io
n
3
;
introdu
ct
io
n
of
the
m
achin
e
le
arn
in
g
te
chn
i
qu
e
s
ap
plied
with
R
ao
-
S
VM
an
d
exp
e
rim
ental
s
et
up
prese
nted
in
s
ect
ion
4
;
th
e
resu
lt
s
of
the
R
ao
-
S
VM
al
gorithm
in
co
m
par
is
on
to
R
ao
-
SV
M
pr
ese
nted
in
s
e
ct
ion
5
;
an
d
th
e co
nclusi
on o
the stu
dy
pr
ese
nted
i
n
s
ect
io
n
6
.
2.
FEATU
RE S
UBSET
SELE
CTIO
N
P
ROB
LE
M
Feat
ur
e
subset
sel
ect
ion
is
th
e
proces
s
of
s
el
ect
in
featu
re
su
bse
ts
f
r
om
a
la
rg
e
set
of
fe
at
ur
es.
It
is
aim
ed
at
red
uc
ing
the
c
om
pl
ex
cal
culat
ions
by
rely
ing
on
few
er
num
ber
of
feat
ur
es
to
achieve
d
i
m
pr
ove
d
perform
ance
of
cl
assifi
ers
.
V
ario
us
sc
ho
la
rs
hav
e
pro
vid
e
d
di
ff
e
ren
t
defi
niti
on
s
of
FS
S
[
62
]
;
f
or
i
nst
ance,
so
m
e
def
ined
i
t
as
the
reducti
on
of
t
he
siz
e
of
t
he
sel
ect
ed
featur
e
subset
wh
il
e
s
om
e
c
on
si
der
e
d
it
a
way
of
i
m
pr
ovin
g
the
pr
e
dicti
on
ac
cur
acy
of
cl
as
sifie
rs.
F
SS
is
reg
a
rd
e
d
as
a
way
of
est
a
bl
ishing
t
he
effe
ct
ive
su
bse
ts
that
ca
ptures
the
inf
orm
ation
hidde
n
in
a
dataset
by
rem
ov
ing
the
irreleva
nt
and
re
dunda
nt
featur
es
.
Hen
ce
,
t
he
ai
m
of
F
SS
is
to
fi
nd
the
le
ast
nu
m
ber
of
feat
ures
with
out
sig
ni
ficantl
y
aff
ect
ing
t
he
cl
assi
ficat
ion
accuracy.
Extr
act
ion
of
op
ti
m
al
featur
e
subsets
is
a
com
plex
ta
s
k
that
currently
has
no
poly
no
m
ia
l
tim
e
al
g
ori
thm
to
address
it
;
this
i
m
pl
ie
s
that
FS
S
is
an
N
P
-
hard
prob
le
m
[63
]
.
A
ty
pical
FS
S
in
vo
l
ves
4
t
ypic
al
ste
ps
[
64]
-
[
67]
:
i)
a
searc
h
for
the
sel
ect
ion
of
in
div
id
ual
fea
tures
t
hat
will
m
ake
up
the
subsets;
ii
)
eval
ua
ti
on
of
the
subsets
and
their
c
om
par
iso
n
with
ea
ch
oth
e
r;
ii
i)
de
te
rm
inati
on
of
wh
et
her
the
t
erm
inati
on
co
ndit
ion
has bee
n
m
et
;
and
iv
)
c
heck f
or th
e esta
blish
m
ent o
f
the
optim
al
f
eat
ur
e s
ubset
based o
n p
re
-
knowle
dge.
Pr
oble
m
def
in
it
ion
:
t
his
stud
y
involves
t
wo
m
ajo
r
part
s:
best
featur
e
subset
sel
ect
ion
,
a
nd
perform
ance
evaluati
on.
Sinc
e
there
are
two
m
ajo
r
obj
ect
i
ves,
FS
S
is
con
side
red
a
m
ul
ti
-
obj
ect
ive
pr
ob
le
m
.
A
f
orm
al
d
efin
it
ion
of
fin
ding
optim
al
so
luti
on
s
th
rou
gh m
eet
ing
bo
t
h o
bject
ives
is p
r
ov
i
ded in
(
1)
a
nd
(2)
.
1
=
|
|
(1)
2
=
(
)
ℎ
⊆
(2)
Wh
e
re k
re
pr
es
ents
the
s
ub
set
o
f
th
e
ori
gin
al
d
at
aset
(K)
tha
t
op
ti
m
iz
es
f1
and
f2
(
the objec
ti
ves)
.
T
he
s
eco
nd
par
t
in
vo
l
ves
the
evaluati
on
of
the
sel
ect
ed
featur
e
s
ub
set
s
based
on
acc
ur
acy
(a
n
est
ablishe
d
perf
orm
ance
evaluati
on
m
etr
ic
),
a
s
pr
ov
i
de
d
in
(
3
)
.
Acc
ur
acy
cal
c
ulati
on
re
quires
t
he
div
isi
on
of
t
he
in
sta
nces
t
hat
are
cl
assifi
ed
c
orre
ct
ly
b
y al
l i
ns
ta
nces.
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Op
ti
mize
d m
ac
hin
e le
arnin
g a
lgo
rit
hm f
or
in
trusio
n detec
ti
on
(
Royid
a A.
I
br
ahe
m
Al
ha
y
al
i
)
593
=
(
+
)
/
(
+
+
+
)
(3)
Wh
e
re T
P = t
r
ue posi
ti
ve,
T
N
= t
ru
e
n
e
gative,
FP
=
f
al
se
po
sit
ive
, a
nd F
N
=
false
neg
at
ive.
The
pro
po
se
d
R
ao
-
S
VM
al
go
rithm
was
i
m
ple
m
ented
at
the
FSS
ph
a
se
.
I
t
was
init
ia
li
zed
via
ra
ndom
gen
e
rati
on
of
a
n
init
ia
l
po
pula
ti
on
cal
le
d
the
Teacher
a
nd
a
set
of
Stu
den
ts
.
The
featu
res
wer
e
re
present
ed
in
the
R
ao
-
S
VM
by
com
bin
ing
R
ao
-
S
VM
with
G
A
,
an
d
the
featur
e
s
wer
e
r
epr
ese
nted
as
a
chr
om
os
om
e
as
this
is
on
e
of
the
fe
at
ur
es o
f
G
A.
Th
e
ch
r
om
os
ome
s
wer
e
upda
te
d
by
ap
plyi
ng
cr
os
s
ov
e
r
an
d
m
utati
on
para
m
et
er
s
of
GA
a
nd
ea
ch
so
l
ution
i
n
the
popula
ti
on
is
consi
de
re
d
a
ch
ro
m
os
om
e
or
an
in
div
id
ual
,
as
s
ho
wn
i
n
Fig
ure
1.
Ch
rom
os
o
m
es
that
hav
e
a
featu
re
gen
e
with
a
va
lue
of
1
are
c
onside
red
sel
ect
ed
w
hile
tho
se
with
a
value
of
0
a
re
no
t
sel
ect
ed
.
T
he
R
ao
-
S
VM
execu
te
d
var
io
us
it
erati
ons
a
d
the
te
ac
her
i
s
co
ns
ide
red
t
he
be
st
ind
ivi
du
al
i
n
the
po
pu
la
ti
on
wh
i
le
the
rest
are
the
st
ud
e
nt
s.
A
fter
sel
ect
ing
t
he
te
ache
r,
the
three
phas
es
of
R
ao
-
S
VM
are
init
ia
te
d
wh
ic
h
are
the
Teach
er,
Be
st
Cl
assm
at
es
(Learn
er
Ph
ase
1),
an
d
Learn
e
r
Phase
2.
Th
e
Teacher
P
hase
involves
t
he
te
acher
s
har
i
ng
kn
ow
le
dge
with
eac
h
stu
de
nt
t
o
im
pr
ov
e
their
un
der
st
and
i
ng
wh
il
e
the
Be
st
Cl
ass
m
at
e
Ph
ase
involves
s
el
ect
ion
of
tw
o
be
st
stude
nt
s
that
will
interact
with
the
oth
e
r
stud
e
nts.
Lea
r
ner
Ph
ase
2
in
vo
l
ves
ra
ndom
interact
ion
a
m
on
g
the
eac
h
to
en
han
ce
t
he
ir
unde
rstan
di
ng.
T
he
gen
e
rati
on
of
ne
w
ch
r
om
os
om
es
in
the
ne
w
R
ao
-
S
VM
was
done
usi
ng
s
pecial
cr
os
s
ov
e
r
operat
or
s
cal
le
d
half
-
unif
or
m
cro
ss
over
an
d
bi
t
-
flip
m
utati
on
op
e
rato
rs
,
as
show
n
in
Fig
ur
e
1
an
d
Fig
ur
e
2
.
T
he
c
rosso
ver
op
e
rato
r
need
s
two
parent
c
hrom
os
om
es
(m
ay
be
a
te
ache
r
an
d
a
st
uden
t,
or
t
wo
stu
de
nts).
The
cr
os
s
ov
e
r
op
e
rato
r
de
pends
on
the
inf
orm
at
ion
of
the
two
par
e
nt
chro
m
os
o
m
es
.
So
,
if
the
sa
m
e
gen
e
is
pr
ese
nt
in
both
par
e
nts,
t
he
ge
ne
is
kep
t;
bu
t
w
hen
t
he
parents
feat
ur
e
di
ff
e
ren
t
genes,
the
ge
ne
of
ei
ther
pa
ren
t
is
c
ho
s
en
rand
o
m
l
y
[32]
.
This
ope
rati
on
re
s
ults
in
the
ge
nerat
ion
of
on
e
ne
w
c
hrom
os
om
e.
The
bit
-
flip
m
utati
on
op
e
rates
on
a
s
ing
le
ch
r
om
os
om
e
to
al
te
r
a
sing
le
gen
e
us
i
ng
a
pro
ba
bili
sti
c
rati
o;
if
the
gen
e
has
a
val
ue
of
zero, it
w
il
l be
updated
as
one
, or vice
ver
sa
.
1
= s
el
ect
ed fe
at
ur
es
0
=
unsel
ect
ed feat
ur
e
s
Figure
1. Sc
he
m
at
ic
r
epr
ese
ntati
on
of a c
hro
m
os
o
m
e
Figure
2
.
Muta
ti
on
op
e
rato
r
3.
PROP
OSE
D RAO
-
S
V
M M
ET
HOD
In
the
pro
pose
d
al
gorithm
,
R
ao
was
im
plem
ented
at
the
FSS
phase;
it
was
init
ia
li
zed
via
rand
om
gen
e
rati
on
of
t
he
init
ia
l
po
pu
la
ti
on
cal
le
d
the
Teache
r
an
d
a
set
of
Stu
den
ts
(t
hese
re
pr
ese
nt
the
po
te
ntial
so
luti
ons
).
T
he
n,
the
cr
os
s
ov
e
r
an
d
m
utati
on
op
e
rato
rs
of
GA
was
bor
rowed
t
o
re
pre
se
nt
the
fe
at
ur
e
s
in
the
R
ao
via
t
he
re
pr
ese
ntati
on
of
the
feat
ur
es
a
s
ch
ro
m
os
om
e
s.
Cr
os
s
ov
e
r
w
as
us
e
d
to
up
da
te
the
ch
ro
m
os
om
e.
Each
so
l
ution
in
t
he
popu
la
ti
on
is
c
on
sidere
d
a
n
i
ndivid
ual/c
hrom
os
om
e
,
as
s
how
n
i
n
Fig
ure
1.
Chrom
os
om
es
that
ha
ve
a feat
ur
e
g
e
ne
with a
value
of 1
a
r
e
co
ns
ide
red
se
le
ct
ed
w
hile
th
os
e w
it
h
a
val
ue
of
0
are
no
t sel
ect
e
d.
Detai
ls o
f
th
e pro
posed
m
eth
od a
re
pr
e
sent
ed
in
A
l
gorith
m
1
and
Fig
ur
e
3
.
A
l
g
o
r
i
t
hm
1:
D
e
t
a
i
l
s
o
f
t
h
e
R
ao
-
S
V
M
a
l
g
o
r
i
t
hm
S
t
e
p
1
:
I
n
i
t
i
a
l
i
z
e
t
h
e
p
o
p
u
l
a
t
i
o
n
r
a
n
d
o
m
l
y
w
i
t
h
e
a
c
h
p
o
p
u
l
a
t
i
o
n
h
a
v
i
n
g
d
i
f
f
e
r
e
n
t
s
e
t
o
f
f
e
a
t
u
r
e
s
S
t
e
p
2
:
B
a
s
e
d
o
n
t
h
e
a
c
c
u
r
a
c
y
o
f
t
h
e
c
l
a
s
s
i
f
i
c
a
t
i
o
n
f
o
r
e
a
c
h
s
e
t
o
f
f
e
a
t
u
r
e
s
,
s
p
e
c
i
f
y
b
e
s
t
a
n
d
w
o
r
s
t
s
et
(
p
o
p
u
l
a
t
i
o
n
)
S
t
e
p
3
:
M
o
d
i
f
y
s
ol
u
t
i
on
s
b
a
s
e
d
o
n
t
h
e
b
e
s
t
a
n
d
w
o
r
s
t
s
o
l
ut
i
o
n
s
a
nd
r
a
n
d
o
m
i
n
t
e
r
a
c
t
i
o
n
s
b
a
s
e
d
o
n
N
e
w
_
s
e
t
=
r
a
n
d
o
m
_
s
e
t
c
r
os
s
o
v
e
r
w
i
t
h
(
b
e
s
t
_
s
e
t
c
r
o
s
s
o
v
e
r
w
i
t
h
w
o
r
s
t
_
s
e
t
)
S
t
e
p
4
:
I
f
t
h
e
n
e
w
s
e
t
of
f
e
a
t
u
r
e
s
b
e
t
t
e
r
t
h
a
n
t
h
e
o
l
d
b
e
s
t
s
e
t
(
i
n
t
e
r
m
of
a
c
c
u
r
a
c
y
o
f
c
l
a
s
s
i
f
i
c
a
t
i
o
n
)
t
h
e
n
k
e
e
p
t
h
e
n
e
w
s
e
t
e
l
s
e
k
e
e
p
t
h
e
o
l
d
s
e
t
S
t
e
p
5
:
I
s
t
h
e
t
e
r
m
i
n
a
t
i
o
n
c
r
i
t
e
r
i
a
s
a
t
i
s
f
i
e
d
o
r
n
o
t
?
i
f
y
e
s
r
e
p
o
r
t
t
h
e
b
e
s
t
s
e
t
o
f
f
e
a
t
u
r
e
s
,
e
l
s
e
g
o
t
o
s
t
e
p
3
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
24
, N
o.
1
,
Oct
ober
20
21
:
590
-
59
9
594
S
t
a
r
t
I
n
i
t
i
a
l
i
z
e
p
o
p
u
l
a
t
i
o
n
s
i
z
e
,
n
u
m
b
e
r
o
f
d
e
s
i
g
n
v
a
r
i
a
b
l
e
s
a
n
d
t
e
r
m
i
n
a
t
i
o
n
c
r
i
t
e
r
i
o
n
I
d
e
n
t
i
f
y
b
e
s
t
a
n
d
w
o
r
s
t
s
o
l
u
t
i
o
n
i
n
t
h
e
p
o
p
u
l
a
t
i
o
n
M
o
d
i
f
y
t
h
e
s
o
l
u
t
i
o
n
s
b
a
s
e
d
o
n
b
e
s
t
a
n
d
w
o
r
s
t
s
o
l
u
t
i
o
n
s
a
n
d
r
a
n
d
o
m
i
n
t
e
r
a
c
t
i
o
n
s
N
e
w
-
s
e
t
=
r
a
n
d
o
m
e
-
s
e
t
c
r
o
s
s
o
v
e
r
w
i
t
h
(
b
e
s
t
-
s
e
t
c
r
o
s
s
o
v
e
r
w
i
t
h
w
o
r
s
t
-
s
e
t
)
I
s
t
h
e
s
o
l
u
t
i
o
n
s
c
o
r
r
e
s
p
o
n
d
i
n
g
t
o
N
e
w
-
s
e
t
B
e
t
t
e
r
t
h
a
n
t
h
a
t
c
o
r
r
e
s
p
o
n
d
i
n
g
t
o
C
u
r
r
e
n
t
-
s
e
t
K
e
e
p
t
h
e
p
r
e
v
i
o
u
s
s
o
l
u
t
i
o
n
A
c
c
e
p
t
a
n
d
r
e
p
l
a
c
e
t
h
e
p
r
e
v
i
o
u
s
s
o
l
u
t
i
o
n
I
s
t
e
r
m
i
n
a
t
i
o
n
c
r
i
t
e
r
i
o
n
s
a
t
i
s
f
i
e
d
R
e
p
o
r
t
t
h
e
b
e
s
t
s
e
t
o
f
f
e
a
t
u
r
e
s
E
n
d
No
Y
es
Y
es
No
Figure
3. Ra
o
-
SV
M al
go
rith
m
4.
E
X
PERI
MEN
TAL SET
UP
The
prese
nt
stud
y
evalu
at
ed
the
so
l
utions
achieve
d
usi
ng
R
ao
-
S
VM
by
dep
loyi
ng
three
ML
te
chn
iq
ues
(L
R,
SV
M
,
an
d
ELM
).
LR
is
a
com
m
on
,
fast
,
an
d
easi
ly
im
plem
ented
cl
assifi
er;
SV
M
is
well
-
known
f
or
it
s
eff
ect
ive
ness
in
bi
nar
y
cl
as
sific
at
ion
;
w
he
reas
ELM
is
a
ne
wly
introdu
ce
d
but
pro
m
isi
ng
cl
assifi
er.
LR:
cl
assifi
cat
ion
with
LR
is
perform
ed
by
est
i
m
at
ing
an
eve
nt’s
occ
urre
nc
e
prob
a
bili
ty
based
on
the
sim
il
arit
y
of
giv
e
n
data
po
i
nts.
It
fin
ds
the
pro
ba
bili
ty
of
the
eve
nt
occurre
nce
by
e
m
plo
yi
ng
a
sigm
oid
functi
on.
I
f
th
e
occurre
nce
pro
ba
bili
ty
of
a
n
eve
nt
is
>0.
5,
the
n
the
LR
pr
edict
s
the
e
ven
t
as
“occ
urred
”
or
“no
t
occurre
d”
,
as
the
case
m
ay
be.
SVM
:
cl
assifi
cat
i
on
ta
s
ks
us
in
g
S
VM
are
pe
rfor
m
ed
thr
ough
the
const
ru
ct
io
n
of
a
separ
at
ing
li
ne
betwee
n
th
e
giv
en
data
points
[
37
]
.
T
he
data
po
ints
cl
os
est
to
this
li
ne
are
desig
nated
a
s
su
pp
or
t
vecto
r
s
(SVs)
.
T
his
li
ne
is
it
erati
vely
con
str
ucte
d
th
rou
gh
the
m
axi
m
iz
ation
of
t
he
m
arg
in
bet
wee
n
the
S
V
an
d
the
li
ne
of
the
c
la
sses.
This
id
ea
or
igi
nates
f
r
om
the
assu
m
ption
that
an
inc
rease
in
the
m
arg
in
can
re
duce
the
gen
e
rali
zat
ion
error.
ELM
:
ELM
is
bu
il
t
as
a
feed
f
orw
ard
ne
ur
al
net
wor
k
(F
F
NN)
with
a
hidden
la
ye
r
,
an
in
pu
t
la
ye
r,
an
d
an
outp
ut
l
ay
er.
The
t
rainin
g
data
ar
e
fed
int
o
the
m
od
el
thr
ough
the
in
pu
t
la
ye
r,
w
he
re
they
are
the
n
weig
hte
d
an
d
f
orwarde
d
to
the
hidden
la
ye
r
via
a
funct
ion
.
A
si
m
il
ar
transfo
rm
ation
is
execu
te
d
bet
ween
the
hidden
a
nd
ou
tp
ut
la
ye
rs.
The
FF
NN
requires
it
erati
ve
tu
ni
ng
of
it
s
par
am
et
e
r;
howe
ve
r,
no
par
am
et
er
tuni
ng
occ
ur
s
i
n
t
he
ELM
.
The
r
efore,
t
he
le
ar
ning
ti
m
e
of
ELM
is
lowe
r
as c
om
par
ed
to
t
ho
se
of c
onve
ntion
al
FFNNs.
The
ex
pe
rim
en
ta
l
scenario
,
prob
le
m
instances,
an
d
the
outc
om
e
of
the
experim
ents
are
all
pr
esent
e
d
in
this
sect
ion
.
In
this
stud
y,
the
exp
e
rim
e
nts
wer
e
perform
ed
on
two
intr
us
io
n
datas
et
s
(K
D
DCup
99
a
nd
CICID
S)
,
w
hi
ch
wer
e
reduc
ed,
beca
us
e
of
the
f
oc
us
on
bin
a
ry
cl
assif
ic
at
ion
to
acc
omm
od
at
e
on
l
y
two
cl
asses
(nor
m
al
and
intr
us
i
on).
T
o
ens
ure
fa
ir
validat
io
n,
K
-
f
old
validat
io
n
was
us
e
d,
w
her
e
th
e
valu
e
of
K
is
set
to 10
[
39
]
.
KDDCu
p
99
da
ta
set
was
first
us
ed
to
bu
il
d
an
I
DS
at
the
3rd
inter
natio
na
l
kn
owle
dge
disco
ver
y
an
d
data
m
ining
t
oo
ls
c
om
petit
i
on
[50]
.
T
he
def
e
ns
e
a
dv
a
nc
ed
resea
rc
h
pro
j
ect
s
age
ncy
(
D
ARPA
)
int
ru
si
on
detect
ion
e
valu
at
ion
pro
gr
am
was
set
up
in
1998
by
the
MI
T
Lincoln
La
borat
or
y
as
a
sim
ula
te
d
env
ir
onm
ent
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Op
ti
mize
d m
ac
hin
e le
arnin
g a
lgo
rit
hm f
or
in
trusio
n detec
ti
on
(
Royid
a A.
I
br
ahe
m
Al
ha
y
al
i
)
595
for
gat
her
i
ng
r
aw
tra
ns
m
issi
o
n
co
ntr
ol pro
t
oc
ol/i
ntern
et
pro
tocol
(
TCP/
IP
)
du
m
p
data f
or
a
local
area n
et
work
(LAN)
[46]
.
It
was
set
up
w
it
h
the
aim
of
com
par
ing
va
rio
us
intr
us
io
n
detect
ion
m
eth
ods
based
on
their
perform
ance.
A
ve
rsion
of
th
e
DA
RP
A
’98
dataset
was
use
d
in
the
K
DDC
up
99
dataset
[31]
.
Th
e
D
A
RPA’9
8
dataset
consi
sts
of
c
om
pr
essed
ra
w
TCP
dum
p
data
of
7
week
s
of
network
patte
r
n.
It
is
app
r
oxim
at
el
y4
gig
a
byte
s in
siz
e and
ca
n be
processe
d
int
o about 5
,00
0,000 con
necti
on
r
ecords,
eac
h of about 10
0
byt
es
[33]
.
In
the
dataset
,
the
t
wo
wee
ks’
te
st
data
c
onta
ins
a
ppr
oxim
at
ely
2,000,0
00
c
onnecti
on
rec
ords.
T
he
KDD
trai
ning
datase
t
is
com
pr
ise
d
of
a
bout
4,900,000
si
ng
le
c
onnecti
on
ve
ct
or
s
of
41
feat
ur
es
eac
h,
w
hich
a
r
e
la
beled
ei
the
r
a
s nor
m
al
o
r
a
n at
ta
ck
of a s
pe
ci
fic ty
pe
[
1]
.
The
at
ta
ck
ty
pe
s in
t
he data
se
t were
cate
gori
zed in
t
o four
m
ajor
cat
e
gories:
a)
Pr
obi
ng
at
ta
ck
:
this
i
s
a
n
ef
f
or
t
by
a
n
at
ta
cker
to
gain
ne
twork
inf
orm
a
ti
on
sim
ply
to
ci
rcu
m
ven
t
th
e
netw
ork’s
sec
uri
ty
con
tr
ols.
The
C
ICI
DS
da
ta
set
con
ta
in
s
both
be
nign
a
nd
the
rece
nt
f
or
m
s
of
at
ta
ck
s
that
m
i
m
ic
real
-
w
or
ld
data
pr
i
ncipa
l
com
pone
nt
analy
sis
of
prote
om
i
cs
(P
CAPs
).
The
dataset
al
so
con
ta
in
s d
at
a
f
ro
m
networ
k
tr
aff
ic
a
naly
sis
wh
ic
h
was
perf
or
m
ed
us
i
ng
a CIC
f
lo
wm
et
er.
The
la
belin
g
of
the
fl
ow
s
is
ba
sed
on
the
tim
est
a
m
p,
the
s
ource
port,
the
destinat
io
n
por
t,
the
s
ource
I
P,
t
he
de
sti
nation
IP
, a
tt
ack,
a
nd
protoc
ols.
b)
Den
ia
l
-
of
-
se
rv
i
ce
(DoS
)
at
ta
ck:
in
t
his
ty
pe
of
at
ta
ck
,
the
intr
ud
e
r
intenti
on
al
ly
de
nies
le
gitim
at
e
netwo
r
k
acce
ss b
y m
aking the
syst
e
m
t
oo busy t
o pro
cess le
giti
m
a
te
re
qu
est
s
.
c)
User
-
to
-
r
oot
(
U2
R
)
at
ta
ck:
the
a
tt
acker
ga
ins
acce
ss
to
the
netw
ork
by
acce
ssin
g
t
he
syst
em
as
a
le
gitim
at
e u
ser,
b
e
f
or
e e
xploit
ing
t
he
la
pse
s i
n
s
om
e syst
e
m
s to gain
roo
t a
ccess.
d)
Rem
ote
-
to
-
u
se
r
(R2L
)
at
ta
ck:
this
is
a
fo
rm
of
at
ta
ck
wh
e
r
e
an
inv
a
de
r
ex
plo
it
s
vulne
rabi
li
t
y
in
m
achines
by
se
nd
i
ng p
ac
kets to
them
o
ve
r
a
netw
ork
i
n a bi
d
to
g
ai
n
l
ocal acce
ss
as
a legal
user.
Althou
gh
se
ve
ral
ty
pes
of
R2
U
at
ta
cks
exist
,
the
m
os
t
com
m
on
ty
pes
are
tho
se
exec
uted
via
s
ocial
eng
i
neer
i
ng.
T
hese
at
ta
cks
(
DoS,
U2
R,
R
2L
,
an
d
prob
i
ng)
are
cl
assifi
e
d
into
22
diff
e
re
nt
at
ta
ck
ty
pes
in
the
KDDCu
p
99
dataset
,
as
sho
wn
in
Ta
ble
1.
These
do
not
on
ly
ref
e
r
to
the
sp
eci
fic
c
ase
of
KDDC
up
99
dataset
;
ad
diti
on
al
ly
,
se
ve
ral
know
n
cl
assif
ic
at
ion
s
a
nd
ta
xonom
ie
s
of
c
om
pu
te
r
syst
e
m
a
tt
acks
we
r
e
al
so
analy
zed i
n
thi
s stu
dy
[
37]
.
The
CIC
IDS
2017
dataset
[1
8],
[43]
sat
isfie
s
the
11
m
and
at
or
y
at
trib
utes
of
a
tr
ue
ID
S
dataset
,
wh
ic
h
are
a
va
il
able
protoc
ols,
feat
ur
e
s
et
,
com
plete
i
nteracti
on,
a
nonym
ity,
com
plete
captu
re,
at
ta
ck
div
e
rsity
,
com
plete
traff
ic
,
m
et
adata,
com
plete
network
c
onfi
gurati
on,
la
beling,
an
d
heteroge
neity
[1
]
,
[18],
[38]
.
The
datas
et
con
ta
ins
3,0
57,50
3
r
ow
s (
de
vised
on
8
file
s)
an
d
each
r
ow
co
ntains
79 f
eat
ur
e
s.
Eac
h
row
is
ei
ther
la
beled
as
ben
i
gn
or
as
a
ny
of
the
14
ty
pes
of
at
ta
ck.
Table
2
su
m
m
arized
the
ty
pes
of
at
ta
ck
distrib
ution i
n
the b
e
nign
r
ows.
In
this
st
ud
y,
the
exp
e
rim
e
nts
we
re
pe
rfor
m
ed
on
a
c
om
pu
te
r
r
unni
ng
a
n
I
ntel
Core
i7
-
4810
process
or
with
a
CP
U
cl
ock
rate
of
2.8
0
G
Hz
a
nd
a
n
8
GB
m
ai
n
m
e
m
or
y.
Th
e
cl
assi
ficat
ion
a
sp
ect
of
the
al
gorithm
s
wa
s
done
us
in
g
MATLAB
2017a.
The
tw
o
im
po
rtant
par
a
m
et
ers
that
mu
st
be
decide
d
pr
io
r
to
run
ning
R
ao
-
S
VM
we
re
popula
ti
on
siz
e
a
nd
num
ber
of
gen
e
rati
ons.
A
higher
val
ue
of
t
hese
par
a
m
et
ers
ens
ur
es
a
hi
gh
er
res
ult
of
acc
ur
acy
,
eve
n
t
houg
h
the
com
pu
ta
ti
on
ti
m
e
will
be
inc
reased
.
A
n
in
vestigat
i
on
of
a n
e
w
in
div
i
dual
is tim
e ineff
ic
ie
nt.
T
a
b
l
e
1
.
K
D
D
d
a
t
a
s
e
t
Attack
class
Ty
p
es o
f
attacks
Do
S
s
m
u
rt,
nep
tu
n
e,
p
o
d
,
teardrop
,
b
ack,
l
an
d
,
R2
L
p
h
f
,
f
t
p
-
write,
i
m
a
p
,
m
u
ltih
o
p
,
war
ez
clien
t,
wa
rez
m
aste
r,
sp
y
,
g
u
ess
pas
swo
rd
U2
R
p
erl,
load
m
o
d
u
le,
b
u
ff
er
-
o
v
erflo
w,
roo
tk
it
Prob
in
g
p
o
rtsweep, ips
wee
p
,
satan
,
n
m
ap
T
a
b
l
e
2
.
C
I
C
I
D
S
d
a
t
a
s
e
t
Attack
class
Ty
p
es o
f
attacks
DOS
DDo
S,
slo
wlo
ris,
Hera
tb
leed
,
Hu
lk
,
Go
ld
en
Ey
e,
Slo
wh
ttp
test
Po
rtScan
Po
rtscan
Bo
t
Bo
t
Bru
te
-
Fo
rce
FTP
-
Pa
tato
r,
SSH
-
Patato
r
W
eb
Att
ack
W
eb
atta
ck
XSS
,
web
attack S
QL
in
jectio
n
,
web
atta
ck
bru
te f
o
rce
Inf
iltration
Inf
iltration
5.
RESU
LT
S
AND DI
SCUS
S
ION
The
pa
ram
et
er
s
us
e
d
in
t
his
stud
y
a
re
s
ho
wn
in
Ta
ble
3.
T
he
acc
ur
acy
res
ult
of
t
he
KDDCu
p
99
dataset
is
pr
e
s
ented
in
Table
4,
a
nd
t
he
re
s
ults
of
t
he
CI
CIDS
2017
da
ta
set
are
pr
e
se
nted
in
Table
5.
T
he
Tables
4
a
nd
5
prese
nt
the
ac
cur
acy
res
ults
for
bo
t
h
data
se
ts.
From
Table
4,
bo
t
h
R
ao
-
S
VM
a
nd
R
ao
-
SV
M
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
24
, N
o.
1
,
Oct
ober
20
21
:
590
-
59
9
596
offer
e
d
the
sa
m
e
execu
ti
on
tim
e
fo
r
each
ML
te
chn
iq
ue.
Fo
r
each
ML,
the
nu
m
ber
of
featur
e
s,
accu
r
acy
,
and
execu
ti
on
tim
e
wer
e
cal
culat
ed.
The
nu
m
ber
s
in
red
s
uggest
the
best
r
esults
for
both
R
ao
-
S
VM
an
d
R
ao
-
SV
M.
R
ao
-
S
V
M
co
ns
ist
ently
prese
nted
bet
te
r
accu
racies
as
com
par
ed
to
R
ao
-
SV
M
usi
ng
the
thre
e
ML
te
chn
iq
ues
.
It
al
so
pr
e
sente
d
bette
r
tim
e
acc
ur
acy
us
i
ng
L
R
and
S
VM
ML
te
chn
iq
ues
.
Howe
ver,
R
ao
-
S
V
M
pro
vid
e
d
a
bette
r
exec
ution
ti
m
e
with
ELM
as
com
par
ed
to
R
ao
-
SV
M
.
W
it
h
the
CICI
DS
20
17
datas
et
,
R
ao
-
SV
M
c
onsist
ently
showe
d
be
tt
er
accu
racy
than
m
any
ot
he
r
al
gorit
hm
s
us
in
g
t
he
th
ree
ML
te
ch
nique
s.
W
it
h
the
LR
te
chn
i
que,
R
ao
-
S
VM
pr
ese
nted
a
bet
te
r
execu
ti
on
ti
m
e
co
m
par
ed
t
o
the
S
VM
an
d
ELM
te
ch
niq
ue
s,
as sho
wn in T
a
ble
5.
T
h
e
d
e
t
e
c
t
i
o
n
r
a
t
e
(
D
R
)
a
s
s
ho
w
n
i
n
(
4
)
r
e
f
e
r
s
t
o
t
h
e
p
e
r
c
e
nt
a
g
e
o
f
t
h
e
c
o
r
r
e
c
t
l
y
c
l
a
s
s
i
f
i
e
d
s
a
m
pl
e
s
b
y
t
h
e
c
l
a
s
s
i
f
i
e
r
i
nt
o
t
h
e
i
r
c
o
r
r
e
c
t
c
l
a
s
s
.
=
+
(4)
A
n
o
t
h
e
r
s
t
a
t
i
s
t
i
c
a
l
t
e
s
t
i
s
t
h
e
e
r
r
o
r
r
a
t
e
,
w
h
i
c
h
i
s
t
h
e
p
r
o
p
o
r
t
i
o
n
o
f
p
a
t
t
e
r
n
s
t
h
a
t
h
a
v
e
b
e
e
n
i
n
c
o
r
r
e
c
t
l
y
c
l
a
s
s
i
f
i
e
d
b
y
t
h
e
m
o
d
e
l
.
E
R
i
s
c
a
l
c
u
l
a
t
e
d
b
a
s
e
d
o
n
(
5
)
.
=
+
+
+
+
(5)
T
a
b
l
e
6
i
l
l
u
s
t
r
a
t
e
s
t
h
e
r
e
s
u
l
t
s
of
(
4
)
a
n
d
(
5
)
.
A
n
o
t
h
e
r
s
t
a
t
i
s
t
i
c
a
l
t
e
s
t
(
T
-
t
e
s
t
)
w
a
s
a
p
p
l
i
e
d
t
o
d
e
m
o
n
s
t
r
a
t
e
t
h
e
s
u
p
e
r
i
o
r
i
t
y
o
f
R
ao
-
S
V
M
o
v
e
r
R
ao
-
S
V
M
.
T
a
b
l
e
7
s
h
o
w
s
t
h
e
p
-
v
a
l
u
e
s
a
n
d
t
-
v
a
l
u
e
s
w
h
e
r
e
t
h
e
s
m
a
l
l
v
a
l
u
e
s
p
o
r
t
r
a
y
e
d
t
h
e
s
i
g
n
i
f
i
c
a
n
c
e
o
f
t
h
e
n
e
w
R
ao
-
S
V
M
a
l
g
o
r
i
t
hm
.
C
om
p
a
r
i
s
o
n
r
e
s
u
l
t
s
w
i
t
h
e
x
i
s
t
i
n
g
w
o
r
k
s
s
h
o
w
e
d
t
h
a
t
t
h
e
p
r
o
p
o
s
e
d
m
o
d
e
l
p
e
r
f
o
r
m
e
d
b
e
t
t
e
r
t
h
a
n
m
a
n
y
e
x
i
s
t
i
n
g
w
o
r
k
s
i
n
t
e
r
m
s
o
f
a
c
c
u
r
a
c
y
a
s
s
h
o
w
n
i
n
T
a
b
l
e
8
.
T
a
b
l
e
3
.
P
a
r
a
m
e
t
e
r
s
u
s
e
d
i
n
t
h
i
s
s
t
u
d
y
P
a
r
am
e
t
e
r
V
a
l
u
e
P
o
p
u
l
a
t
i
o
n
s
i
z
e
f
o
r
R
a
o
40
N
u
m
b
e
r
o
f
g
e
n
e
r
a
t
i
o
n
s
f
o
r
R
a
o
60
C
r
o
s
s
o
v
e
r
t
y
p
e
H
a
l
f
-
u
n
i
f
o
rm
M
u
t
a
t
i
o
n
t
y
p
e
B
i
t
-
f
l
i
p
T
a
b
l
e
4
.
A
c
c
u
r
a
c
y
r
e
s
u
l
t
o
f
K
D
D
C
u
p
9
9
d
a
t
a
s
e
t
Cla
ss
ifier
Ra
o
LR
No
.
o
f
f
eatu
res
Accurac
y
2
0
.99
2
4
0
.99
4
Total ti
m
e
1
1
.40
2
3
SVM
3
0
.97
4
0
.99
8
5
1
.00
Total ti
m
e
1
3
0
5
.0
3
5
5
ELM
2
0
.99
3
3
0
.99
5
4
0
.99
9
Total ti
m
e
4
.42
6
1
T
a
b
l
e
5
.
A
c
c
u
r
a
c
y
r
e
s
u
l
t
o
f
C
I
C
I
D
S
2
0
1
7
d
a
t
a
s
e
t
Cla
ss
ifier
Ra
o
LR
No
.
o
f
f
eatu
res
Accurac
y
8
0
.94
9
0
.95
4
14
0
.95
5
2
15
0
.96
23
0
.97
6
Total ti
m
e
2
9
.08
9
SVM
14
0
.90
17
0
.92
2
22
0
.93
Total ti
m
e
5
4
8
4
.0
9
7
ELM
6
0
.90
1
7
0
.92
Total ti
m
e
6
.53
1
2
T
a
b
l
e
6
.
R
e
s
u
l
t
s
o
f
(
4
)
a
n
d
(
5
)
KDDCu
p
99
CICIDS 20
1
7
Detectio
n
Rate
0
.99
9
0
.99
0
Er
ror
Rate
0
.00
4
4
5
0
.02
6
7
T
a
b
l
e
7
.
T
-
T
e
s
t
K
D
D
C
u
p
9
9
C
I
C
I
D
S
2
0
1
7
P
-
V
a
l
u
e
0
.
0
1
5
7
0
.
0
0
6
7
8
T
-
V
a
l
u
e
3
.
1
7
5
4
.
0
5
1
T
a
b
l
e
8
.
C
om
pa
r
i
s
o
n
w
i
t
h
e
x
i
s
t
i
n
g
w
o
r
k
s
R
e
f
.
D
a
t
a
s
e
t
A
c
c
u
r
a
c
y
[
4
7
]
C
I
C
I
D
S
9
7
.
9
0
%
[
2
]
C
I
C
I
D
S
9
7
.
0
8
%
[
4
1
]
K
D
D
99
.
7
5
%
[
4
9
]
K
D
D
9
9
.
8
9
%
P
r
o
p
o
s
e
d
m
e
t
h
o
d
C
I
C
I
D
S
9
7
.
6
%
P
r
o
p
o
s
e
d
m
e
t
h
o
d
K
D
D
1
0
0
%
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Op
ti
mize
d m
ac
hin
e le
arnin
g a
lgo
rit
hm f
or
in
trusio
n detec
ti
on
(
Royid
a A.
I
br
ahe
m
Al
ha
y
al
i
)
597
6.
CONCL
US
I
O
N
This
pa
per
pro
po
s
es
a
ne
w
(
R
ao
-
S
VM)
for
featur
e
s
ubset
sel
ect
ion
pro
bl
e
m
s
in
intru
sion
detect
ion.
The
pe
rfor
m
ance
of
th
e
new
al
gorithm
was
dem
on
strat
e
d
to
be
s
up
e
rio
r
to
m
any
oth
e
r
al
gorithm
s
in
FS
S
pro
blem
s
on
two
la
rg
e
intr
usi
on
dataset
s.
The
pro
po
se
d
R
ao
-
S
VM
c
onsist
ently
pr
ese
nted
b
et
te
r
acc
ur
acy
i
n
the
exec
utio
n
tim
e.
On
the
s
ta
ti
sti
ca
l
te
sts
(confusi
on
m
at
rix)
a
ppli
ed
t
o
the
R
ao
-
SVM
detect
ion
ra
te
and
error
rate
ext
r
act
ed
f
ro
m
the
co
nfusion
m
at
rix,
R
ao
-
S
VM
sho
wed
a
higher
detect
ion
rate
f
or
both
t
he
KDDCu
p
99
a
nd
IC
IDS2
017
dataset
s
.
It
s
howe
d
a
lo
w
e
rror
rate
for
t
he
t
wo
dataset
s.
A
s
a
rec
omm
end
at
ion,
the
pro
po
s
ed
R
ao
-
S
VM
sho
uld
be
ap
plied
to
m
ulti
-
cl
ass
cl
assifi
cat
ion
pro
blem
s,
and
m
or
e
ML
te
chn
iq
ues
cou
l
d be
us
e
d
f
or eval
uating i
ts pe
rfor
m
ance.
ACKN
OWLE
DGE
MENTS
The
a
uthors
w
ou
l
d
li
ke
to
t
ha
nk
U
niv
e
rsity
of
Diya
la
and
A
l
Sala
m
Un
iversity
Coll
eg
e
for
their
facil
it
ie
s an
d
s
upport; a
nd Un
iversita
s
Ah
m
ad Dahla
n
to
s
uppo
rt this c
ollaborat
ive
resea
r
ch
.
REFERE
NCE
S
[1]
I.
Alj
ara
h
and
S.
A.
Ludwig
,
“
MapReduc
e
in
t
rusion
det
e
ction
s
y
stem
base
d
on
a
par
ti
c
le
s
warm
opti
m
iz
ati
o
n
cl
uster
ing
a
lgori
t
hm
,
”
2013
I
EE
E
Congress
on
E
v
o
.
Comp
.
,
2013
,
pp.
955
-
962
,
doi
:
10.
1109
/CE
C.
2
013.
6557670
.
[2]
S.
Alja
warn
eh,
M.
Aldwairi,
a
nd
M.
B.
Y
assein,
“
Anom
aly
-
base
d
int
rusion
det
e
ct
ion
s
y
st
e
m
through
feat
ure
sele
c
ti
on
anal
y
s
i
s
and
buil
ding
hy
brid
ef
ficien
t
m
odel
,
”
Journal
of
Computati
onal
Sci
en
ce
,
vol
.
25,
pp.
152
–
160,
2018
,
doi
:
10
.
10
16/j
.
jo
cs.
2017
.
0
3.
006
.
[3]
I.
Idrissi,
M.
Boukabous,
M.
A
zi
z
i,
O.
Mous
saoui,
and
H.
E.
Fadil
i,
“
Towa
rd
a
dee
p
le
arn
in
g
-
base
d
int
r
usio
n
det
e
ct
ion
s
y
s
te
m
for
iot
aga
inst
b
otne
t
atta
cks,
”
I
AE
S
Inte
rnat
ion
al
Journal
of
Arti
fi
c
ial
Int
el
l
ige
n
ce
,
vol
.
10,
no.
1
,
pp.
110
–
120
,
20
21,
doi
:
10
.
1159
1/i
jai.
v10
.
i1
.
pp1
10
-
120.
[4]
A.
M.
Al
-
Ghail
i
et
al.
,
“
A
rev
ie
w
of
anomal
y
d
et
e
ct
ion
t
ec
hn
ique
s
in
adva
n
ce
d
m
eteri
ng
infr
astruc
t
ure
,
”
Bul
l
et
in
o
f
El
e
ct
rica
l
Eng
in
ee
ring a
nd
Infor
matic
s
,
vo
l. 10,
no.
1
,
pp
.
266
–
2
73,
2021
,
doi
:
10
.
11591/eei
.
v10i1
.
2026.
[5]
B.
Altay
,
T
.
Dokerogl
u,
and
A.
Cosar,
“
Conte
xt
-
sensiti
ve
and
ke
y
word
density
-
b
ase
d
supervise
d
m
ac
hin
e
learni
n
g
te
chn
ique
s
for
m
al
ic
ious
webp
age
de
tecti
on
,
”
Soft
Computing
-
A
Fusion
of
Foundat
ions,
M
et
hodologies
an
d
Appl
ic
a
ti
ons
,
vo
l.
23
,
no
.
12
,
pp
.
4177
–
4191,
201
9
,
doi
:
10
.
1007/s
00500
-
018
-
3066
-
4
.
[6]
P.
W
anda
,
M.
E. Hiswati
,
and
H. J
.
Jie,
“
Dee
pOS
N:
Bringi
ng
deep l
ea
rning as mali
ci
ous de
tecti
on
sche
m
e
in
onli
n
e
socia
l
net
work
,
”
IA
ES
In
te
rna
ti
onal
Journal
of
Artifi
ci
al
Int
el
li
g
ence
,
vol.
9,
no.
1,
pp.
146
–
154,
2020
,
doi:
10
.
11591/ij
ai
.
v9
.
i1
.
pp146
-
1
54.
[7]
A.
W
.
Muham
m
ad,
C
.
F.
M.
Foo
z
y
,
and
K.
M.
B
.
Moham
m
ad,
“
Multi
sche
m
e
f
eedforward
ar
ti
fi
cial
n
eur
al
ne
twork
arc
hi
te
c
ture
for
ddos
at
ta
ck
d
et
e
ct
ion
,
”
Bu
ll
e
t
in
of
El
e
ct
rica
l
Engi
nee
ring
a
nd
Informatic
s
,
vol.
10,
no.
1,
pp.
458
–
465
,
20
20,
doi
:
10
.
1159
1/e
e
i.
v10i1
.
2383
.
[8]
V.
Mac
h,
M.
A
damek,
J.
Va
lou
ch,
and
K.
B
arcova,
“
Control
a
nd
indi
c
at
ing
eq
uipment
comm
unic
a
ti
ng
vi
a
th
e
per
iphe
r
al
component
in
te
rco
nn
e
ct
expr
ess
bus,”
Bul
letin
o
f
Elec
t
rical
Eng
ine
erin
g
and
Informati
cs
,
vol
.
9,
no
.
2
,
pp.
729
–
738
,
20
20,
doi
:
10
.
1159
1/e
e
i.
v9i2
.
1753.
[9]
M
.
S
.
I
.
S
h
a
r
i
f
u
d
d
i
n
,
S
.
N
o
r
d
i
n
,
a
n
d
A
.
M
.
A
l
i
,
“
C
o
m
p
a
r
i
s
o
n
o
f
C
N
N
s
a
n
d
S
V
M
f
o
r
v
o
i
c
e
c
o
n
t
r
o
l
w
h
e
e
l
c
h
a
i
r
,
”
I
A
E
S
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
A
r
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
9
,
n
o
.
3
,
p
p
.
3
8
7
-
3
9
3
,
2
0
2
0
,
d
o
i
:
1
0
.
1
1
5
9
1
/
i
j
a
i
.
v
9
.
i
3
.
p
p
3
8
7
-
3
9
3
.
[10]
A.
Boukhal
fa
,
N.
Hm
ina
,
and
H.
Chaoui
,
“
Para
l
l
el
proc
essing
using
big
dat
a
and
m
ac
hine
l
ea
rnin
g
te
chni
qu
es
for
int
rusion
detec
t
i
on,
”
IAE
S
In
te
r
nati
onal
Journa
l
of
Arti
ficia
l
Inte
lligen
ce
,
vol.
9,
no.
3,
pp.
553
–
560,
2020,
doi:
10
.
11591/ij
ai
.
v9
.
i3
.
pp553
-
5
60.
[11]
A.
J.
Moham
m
e
d,
M.
H.
Arif,
and
A.
A.
Ali,
“
A
m
ult
il
a
y
er
p
erc
ep
tron
artificia
l
neur
al
ne
twork
appr
oac
h
fo
r
improving
the
accura
c
y
of
in
trusi
on
det
e
ction
s
y
st
ems
,
”
IAE
S
Inte
r
nati
onal Journal
of
Art
if
i
ci
al
Intelli
gen
ce
,
vol
.
9
,
no.
4
,
pp
.
609
–
6
15,
2020
,
doi
:
10
.
11591/i
j
ai.v9.i4
.
pp609
-
615.
[12]
T.
Ghrib,
M.
B
enmoham
m
ed,
and
P.
S.
Pande
y
,
“
Autom
at
ed
dia
gnosis
of
attac
ks
in
interne
t
of
thi
ngs
usin
g
m
ac
hine
l
ea
rnin
g
and
fre
quen
c
y
distr
ibut
ion
t
ec
hniqu
es,
”
Bu
l
le
ti
n
o
f
Elec
tri
c
al
En
gineering
and
Informatic
s
,
vol.
10
,
no
.
2
,
pp
.
950
–
961
,
2021
,
doi: 10.
11591
/eei.
v10
i2.
2766
.
[13]
M.
Alsajri
,
M.
A.
Ism
ai
l,
and
S.
Abdul
-
Baqi,
“
A
Revi
ew
on
the
Recent
A
ppli
c
at
ion
of
Ja
y
a
Opt
imization
Algorit
hm
,
”
201
8
1st
Annual
Inte
rnational
Conf
ere
nce
on
Information
and
Sci
e
nce
s
(
Ai
CIS)
,
20
18,
pp.
129
-
132,
doi:
10
.
1109/Ai
CIS.2018.
00034
.
[14]
S.
M.
H.
Bamakan,
H.
W
ang,
T.
Yingjie,
and
Y.
Shi,
“
An
eff
ec
t
ive
int
rusion
det
ection
fra
m
ework
base
d
on
MCLP
/SVM
o
pti
m
iz
ed
b
y
tim
e
-
var
y
ing
ch
a
os
par
ti
c
le
sw
arm
opti
m
iz
a
tion,”
N
euroc
omputing
,
vo
l.
19
9,
pp.
90
–
102
,
201
6
,
doi
:
10
.
1016/j.ne
ucom.2016
.
0
3.
031
.
[15]
J.
Cai
,
J.
Luo
,
S.
W
ang,
an
d
S.
Yang,
“
Feat
ure
sel
ectio
n
in
m
ac
hine
le
arn
ing:
A
ne
w
per
spec
ti
v
e,
”
Neurocomputi
ng
,
vol
.
300
,
pp
.
70
–
79,
2018
,
doi
:
1
0.
1016/j.ne
u
com.2017.
11
.
077
.
[16]
S.
Raj
agopal,
P.
P.
Kundapur,
and
H.
K.
Siddara
m
appa
,
“
A
pre
dic
ti
v
e
m
odel
for
net
work
int
rusio
n
det
ection
using
stac
king
appr
oa
c
h,
”
Int
ernati
ona
l
Journal
of
El
e
c
tric
al
and
Comp
ute
r
Engi
n
ee
rin
g
,
vol
.
10,
no
.
3
,
pp.
2734
–
2741,
2020,
doi
:
10
.
11
591/i
jece
.
v10i3
.
pp2734
-
2741.
[17]
S.
M.
Shara
th,
P.
Manjuna
th
a,
an
d
H.
R.
Shw
et
ha,
“
Insights
on
cri
ti
c
al
ene
rg
y
ef
fi
ci
en
c
y
appr
oa
ch
es
in
int
ern
e
t
-
of
-
thi
ngs
appl
i
catio
n,
”
Int
ernati
onal
Journal
of
Elec
tric
al
and
Comp
ute
r
E
ngin
ee
rin
g
,
vol.
11
,
no.
4
,
pp.
2925
–
2933,
2021,
doi
:
10
.
11
591/i
jece
.
v11i4
.
pp2925
-
2933.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
24
, N
o.
1
,
Oct
ober
20
21
:
590
-
59
9
598
[18]
A.
Chaudhar
y
,
V.
Ti
wa
ri,
and
A.
Kum
ar,
“
A
novel
int
rusion
det
e
ct
ion
s
y
st
e
m
for
ad
ho
c
f
l
ooding
a
tt
a
ck
u
sing
fuz
z
y
logi
c
in
m
obil
e
ad
hoc
n
etw
orks,”
Int
.
Con
f
.
Rec
.
Adv
.
and
Innov
.
Eng
.
,
201
4,
pp
.
1
-
4
.
[19]
M.
K.
Hos
sain
and
M.
M.
Haque
,
“
Sem
i
-
supervise
d
le
arn
ing
a
pproa
ch
using
m
odifi
ed
self
-
training
al
gor
it
hm
to
count
er
burst
he
ade
r
pa
cke
t
floo
ding
at
t
ac
k
in
op
ti
c
al
burst
sw
it
c
hing
net
work,
”
Inte
rnational
Jou
rnal
o
f
El
ectric
a
l
and
Computer
E
ngine
ering
,
vol
.
10,
no
.
4
,
pp
.
43
40
–
4351,
2020
,
doi:
10
.
11591/ij
ec
e
.
v10i4
.
pp434
0
-
4351.
[20]
K
.
S
.
Y
i
n
a
n
d
M
.
A
.
K
h
i
n
e
,
“
O
p
t
i
m
a
l
r
e
m
o
t
e
a
c
c
e
s
s
T
r
o
j
a
n
s
d
e
t
e
c
t
i
o
n
b
a
s
e
d
o
n
n
e
t
w
o
r
k
b
e
h
a
v
i
o
r
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
E
l
e
c
t
r
i
c
a
l
a
n
d
C
o
m
p
u
t
e
r
E
n
g
i
n
e
e
r
i
n
g
,
v
o
l
.
9
,
n
o
.
3
,
p
p
.
2
1
7
7
–
2
1
8
4
,
2
0
1
9
,
d
o
i
:
1
0
.
1
1
5
9
1
/
i
j
e
c
e
.
v
9
i
3
.
p
p
2
1
7
7
-
2
1
8
4
.
[21]
Y.
Kham
a
y
seh
,
M.
B.
Yass
ei
n
,
and
M.
Abu
-
Jaz
oh,
“
Inte
l
li
gen
t
bla
ck
hole
detec
ti
on
in
m
obil
e
AdH
oc
net
work
s,
”
Inte
rnational
J
ournal
of
Elec
tric
al
and
Co
mpute
r
Engi
ne
ering
,
vol
.
9
,
no.
3,
pp.
1968
–
1977,
20
19,
doi:
10
.
11591/ij
ec
e
.
v9i3
.
pp1968
-
1977.
[22]
H.
H.
Ibra
h
im
,
et
al
.
,
“
A
comprehe
nsive
st
ud
y
of
d
istri
bu
te
d
Deni
al
-
of
-
S
erv
ice
a
tt
a
ck
w
it
h
th
e
de
te
c
ti
o
n
te
chn
ique
s,”
Int
ernati
onal
Journal
of
E
le
c
tric
a
l
and
Computer
Engi
ne
ering
,
vol
.
10,
no
.
4,
pp.
3685
–
3694,
202
0,
doi:
10
.
11591/ij
ec
e
.
v10i4
.
pp368
5
-
3694.
[23]
A.
Boukhal
fa
,
A.
Abdell
aoui,
N.
Hm
ina
,
and
H.
Chaoui
,
“
LSTM
dee
p
le
arn
i
ng
m
et
hod
for
net
work
int
rusio
n
det
e
ct
ion
s
y
st
em,”
In
te
rnationa
l
Journal
of
El
e
ctr
ic
al
and
Comp
ute
r
Engi
n
ee
rin
g
,
vol
.
10,
no.
3
,
pp.
3315
–
332
2
,
2020,
doi
:
10
.
11
591/i
jece
.
v10i3
.
pp3315
-
3322.
[24]
P
.
I
.
P
r
i
y
a
d
a
r
s
i
n
i
a
n
d
G
.
A
n
u
r
a
d
h
a
,
“
A
n
o
v
e
l
e
n
s
e
m
b
l
e
m
o
d
e
l
i
n
g
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
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
E
l
e
c
t
r
i
c
a
l
a
n
d
C
o
m
p
u
t
e
r
E
n
g
i
n
e
e
r
i
n
g
,
v
o
l
.
1
0
,
n
o
.
2
,
p
p
.
1
9
6
3
–
1
9
7
1
,
2
0
2
0
,
d
o
i
:
1
0
.
1
1
5
9
1
/
i
j
e
c
e
.
v
1
0
i
2
.
p
p
1
9
6
3
-
1
9
7
1
.
[25]
M.
Nare
nder
an
d
B.
N.
Yuvara
j
u,
“
Pree
m
pti
ve
m
odel
li
ng
towar
ds
cl
assif
y
ing
vu
lne
rab
il
i
t
y
of
D
DoS
at
ta
ck
in
SD
N
envi
ronm
ent
,
”
I
nte
rnational
Jou
rnal
of
El
e
ct
rica
l
a
nd
Comput
er
Engi
ne
ering
,
vo
l
.
10,
no.
2,
pp.
1
599
–
1611,
2020
,
doi:
10
.
11591/ij
ec
e
.
v10i2
.
pp159
9
-
1611.
[26]
M.
C.
Bel
av
agi
and
B.
Muni
y
al,
“
Multi
ple
int
ru
sion
det
ection
in
RP
L
base
d
net
works
,
”
Inte
rnat
ional
Journal
of
El
e
ct
rica
l
and
C
omputer
Engi
n
e
ering
,
vo
l
.
10,
n
o.
1
,
pp
.
467
–
47
6,
2020
,
doi
:
10
.
11591/i
jece
.
v10i
1.
pp467
-
476.
[27]
A.
Sara
v
ana
n
,
S.
Sath
y
a
Bam
a,
S.
Kadr
y
,
a
nd
L.
K.
Ramasam
y
,
“
A
ne
w
fra
m
ework
t
o
alleviate
DD
oS
vulne
rab
il
i
ti
es
i
n
cl
oud
computi
ng,
”
Int
ernati
on
al
Journal
of
Elec
tri
cal
and
Computer
Engi
neer
ing
,
vol.
9,
no
.
5,
pp.
4163
–
4175
,
2019,
doi
:
10
.
11
591/i
jece
.
v9i5
.
p
p4163
-
4175.
[28]
M.
S.
Vid
y
a
a
nd
M.
C.
Pati
l,
“
Revi
ewing
eff
ec
t
ivi
t
y
in
sec
urity
appr
oa
ches
towar
ds
strengt
heni
ng
intern
e
t
arc
hi
te
c
ture,”
In
te
rnational
Jour
nal
of
El
ectric
al
and
Com
pute
r
Engi
ne
ering
,
vol
.
9,
no.
5,
pp.
3862
–
3871,
2019,
doi:
10
.
11591/ij
ec
e
.
v9i5
.
pp3862
-
3871.
[29]
Y.
N.
Doddam
ani
and
U.
C.
Ka
pal
e
,
“
A
tra
nsiti
on
from
m
anua
l
to
int
e
ll
ig
ent
autom
at
ed
power
sy
stem
oper
at
ion
-
A
indi
cative
r
eview,”
Int
ernati
on
al
Journal
of
E
l
ec
tri
cal
and
Co
mputer
Engi
ne
ering
,
vol
.
9,
no
.
4
,
pp.
2274
–
2280
,
2019,
doi
:
10
.
11
591/i
jece
.
v9i4
.
p
p2274
-
2280.
[30]
S.
P.
Das
and
S.
Padh
y
,
“
A
nov
el
h
y
brid
m
odel
using
teac
hing
–
le
arn
ing
-
base
d
o
pti
m
iz
ation
and
a
support
ve
ct
or
m
ac
hine
fo
r
com
m
odity
future
s
i
ndex
fore
ca
st
ing
,
”
In
t. J.
Ma
ch.
Learn.
Cybe
rn
.
,
vol.
9
,
no
.
1
,
pp
.
97
–
111,
2018
.
[31]
E.
De
l
a
Hoz
,
E
.
De
L
a
Hoz
,
A.
Ortiz
,
J.
Ort
ega,
and
B.
Priet
o
,
“
PC
A
fil
te
ring
an
d
proba
bi
li
sti
c
S
OM
for
net
work
int
rusion
d
et
e
ct
i
on,
”
N
euroc
omp
uti
ng
,
vol. 164
,
pp.
71
–
8
1
,
2015
,
doi:
10.
1016
/j
.
n
euc
om
.
2014.
09
.
083
.
[32]
D
.
D
i
n
g
,
Q
.
-
L
.
H
a
n
,
Y
.
X
i
a
n
g
,
X
.
G
e
,
a
n
d
X
.
-
M
.
Z
h
a
n
g
,
“
A
s
u
r
v
e
y
o
n
s
e
c
u
r
i
t
y
c
o
n
t
r
o
l
a
n
d
a
t
t
a
c
k
d
e
t
e
c
t
i
o
n
f
o
r
i
n
d
u
s
t
r
i
a
l
c
y
b
e
r
-
p
h
y
s
i
c
a
l
s
y
s
t
e
m
s
,
”
N
e
u
r
o
c
o
m
p
u
t
i
n
g
,
v
o
l
.
2
7
5
,
p
p
.
1
6
7
4
–
1
6
8
3
,
2
0
1
8
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
n
e
u
c
o
m
.
2
0
1
7
.
1
0
.
0
0
9
.
[33]
T.
Dokerog
lu,
“
H
y
brid
te
a
chi
ng
–
le
arn
ing
-
bas
ed
opti
m
iz
ation
a
lg
orit
hm
s
for
the
quadr
atic
assignm
ent
proble
m
,
”
Computers
&
In
dustrial
Engi
n
eer
ing
,
vol
.
85
,
pp
.
86
–
101,
2015
,
d
oi:
10
.
1016/j.c
ie.
2015.
03.
001
.
[34]
J.
A.
Jupin,
T.
S
uti
kno,
M
.
A.
Is
m
ai
l,
M.
S
.
Moham
ad,
S.
Kasim
,
and
D.
Sti
awa
n
“
Revi
ew
of
the
m
ac
hine
l
ea
rn
in
g
m
et
hods
in
th
e
c
la
ss
ifi
c
at
ion
of
p
hishing
a
tt
a
ck,”
Bul
letin
of
Elec
t
rical
Engi
ne
erin
g
and
Informati
cs
,
vol
.
8
,
no
.
4
,
pp.
1545
–
1555
,
2019,
doi
:
10
.
11
591/e
e
i.
v8i4
.
134
4
.
[35]
S.
Raj
agopal,
K.
S.
Hare
esha
,
an
d
P.
P.
Kundap
ur,
“
Perform
anc
e
ana
l
y
sis
of
bin
ar
y
and
m
ult
ic
l
a
ss
m
odel
s
using
az
ure
m
ac
hin
e
le
a
rning
,
”
Int
e
rnational
Journal
of
El
ectric
al
and
Computer
Engi
nee
ring
,
vol.
10,
no.
1
,
pp.
978
–
986
,
20
20,
doi
:
10
.
1159
1/i
jece
.
v10i1
.
pp
978
-
986.
[36]
K.
Farha
na
,
M.
Rahman,
and
M
.
T
.
Ahm
ed,
“
An
int
rusion
de
tecti
on
s
y
s
te
m
for
pac
ke
t
and
f
lo
w
base
d
net
wor
ks
us
ing
dee
p
neur
al
n
et
work
appr
oac
h,
”
In
te
rnati
onal
Journal
o
f
Elec
tri
cal
and
Computer
Enginee
ring
,
vol
.
10
,
no.
5
,
pp
.
5514
–
5525,
2020
,
doi
: 10.11591/
i
je
c
e.
v
10i5.
pp5514
-
55
25.
[37]
S.
Dum
ai
s,
J.
P
la
tt,
D.
Hec
k
er
m
an,
and
M
.
Sa
hami,
“
Induc
ti
v
e
l
ea
rning
al
gor
it
hm
s
and
rep
r
e
senta
ti
ons
fo
r
t
e
xt
ca
t
egor
izati
on
,
”
Proce
ed
ings
of
t
he
sev
ent
h
in
te
r
nati
onal
conf
ere
nce
on
In
formati
on
and
knowled
ge
management
,
1998,
pp
.
148
–
1
55
,
doi
:
10
.
1145
/288627.
288651
.
[38]
A.
S.
Ee
sa,
Z.
Orm
an,
and
A.
M.
A.
Brifc
ani
,
“
A
novel
fea
ture
-
sel
ec
t
ion
a
pproa
ch
base
d
on
the
cut
tlefis
h
opti
m
iz
ation
a
l
gorit
hm
for
intrus
ion
det
e
ct
ion
s
y
stems
,
”
E
xp
ert
Syste
ms
wit
h
Appl
ic
a
ti
ons
,
vol.
42,
no
.
5,
pp.
2670
–
2679
,
2015
,
doi
:
10
.
10
16/j
.
eswa.
2014.
1
1.
009
.
[39]
C.
Guo,
Y.
Ping
,
N.
Li
u,
and
S.
-
S.
Luo,
“
A
two
-
le
ve
l
h
y
brid
app
roa
ch
for
int
rusi
on
det
ection,”
Neurocomputi
ng
,
vol.
214
,
pp
.
391
–
400,
2016
,
doi
:
10.
1016/j.ne
u
co
m
.
2016.
06.
021
.
[40]
H.
E.
Ki
zi
lo
z,
A.
Deniz,
T.
D
oker
oglu,
and
A.
Cosar,
“
Novel
m
ult
io
bjecti
v
e
TL
BO
al
gor
it
h
m
s
for
the
fea
tu
re
subs
et
select
ion
proble
m
,
”
Neuro
computi
ng
,
vol
.
306,
pp
.
94
–
107
,
2018
,
doi:
10.
10
16/j
.
n
eu
com.201
8.
04.
020
.
[41]
M.
K.
Khal
ee
l
,
M.
A.
Ism
ai
l,
U
.
Yunan,
and
S.
Kasim
,
“
Revi
ew
on
int
rusion
d
etec
t
ion
s
y
st
em
ba
sed
on
the
go
al
of
the
d
et
e
ction
s
y
s
te
m
,
”
Int. J.
In
tegr
.
Eng
.
,
vol. 10, no. 6, 2018
,
doi:
10.
30880/
ij
i
e.
20
18.
10.
06
.
028
.
[42]
W.
-
C.
Li
n
,
S.
-
W
.
Ke,
and
C
.
-
F.
Tsai
,
“
CAN
N:
An
int
rusion
d
et
e
ct
ion
s
y
stem
base
d
on
combi
ning
cl
ust
er
c
enters
and
ne
are
st
nei
g
hbors,”
Knowl
ed
ge
-
based
Syst
.
,
v
ol.
78
,
pp
.
13
–
21
,
2015
,
doi
:
10
.
1
016/j
.
knos
y
s
.
20
15.
01.
009
.
[43]
Y.
Li,
J.
-
L.
W
a
ng,
Z
.
-
H.
Ti
an
,
T.
-
B.
Lu
,
and
C
.
Young,
“
Buil
d
ing
li
gh
twei
ght
int
rusion
de
tection
s
y
stem
using
wrappe
r
-
base
d
f
ea
tur
e
se
le
c
ti
on
m
ec
hani
sm
s,”
C
omputers
&
Sec
urity
,
vol
.
28
,
no
.
6
,
pp
.
466
-
475
,
2009
.
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Op
ti
mize
d m
ac
hin
e le
arnin
g a
lgo
rit
hm f
or
in
trusio
n detec
ti
on
(
Royid
a A.
I
br
ahe
m
Al
ha
y
al
i
)
599
[44]
P.
Louvi
eri
s,
N
.
Cle
wle
y
,
and
X.
Li
u,
“
Eff
e
cts
-
base
d
fea
tur
e
ide
nti
f
icati
on
fo
r
net
work
int
ru
sion
det
ection,”
Neurocomputi
ng
,
vol
.
121
,
pp
.
26
5
–
273,
2013
,
doi
:
10.
1016
/j.ne
uc
om
.
2013.
04.
038
.
[45]
M.
Črepi
nšek
,
S.
-
H.
Li
u
,
and
L.
Mernik,
“
A
not
e
on
t
eachi
ng
–
l
earning
-
ba
sed
op
tim
iz
at
ion
al
gor
ithm
,
”
Informatio
n
Sci
en
ce
s
,
vol. 21
2,
pp
.
79
–
93
,
20
12
,
doi
10.
1016
/
j.
ins.20
12
.
05.
00
9
.
[46]
S.
Mahda
vif
ar
a
nd
A
.
A.
Ghorb
ani
,
“
Applic
a
ti
o
n
of
de
ep
l
ea
rn
i
ng
to
c
y
ber
sec
u
r
ity
:
A
surv
e
y
,
”
Neurocomputi
ng
,
vol.
347
,
pp
.
149
–
176,
2019
,
doi
:
10.
1016/j.ne
u
co
m
.
2019.
02.
056
.
[47]
M.
A.
Mohamm
ed,
et
al
.
,
“
A
Fo
ca
l
lo
ad
bal
an
ce
r
base
d
al
gorit
hm
for
ta
sk
assign
m
ent
in
cl
oud
envi
ronm
ent
,
”
201
8
10th
Int
.
Con
f
.
E
le
c
.
,
Comp
t.
Artif
.
In
te
l
l.,
2018
,
p
p.
1
-
4
,
doi
:
10
.
1
109/E
CAI.2018.8679043
.
[48]
M
.
R
.
N
a
ya
k
,
C
.
K
.
N
a
ya
k
,
a
n
d
P
.
K
.
R
o
u
t
,
“
A
p
p
l
i
c
a
t
i
o
n
o
f
m
u
l
t
i
-
o
b
j
e
c
t
i
v
e
t
e
a
c
h
i
n
g
l
e
a
r
n
i
n
g
b
a
s
e
d
o
p
t
i
m
i
z
a
t
i
o
n
a
l
g
o
r
i
t
h
m
t
o
o
p
t
i
m
a
l
p
o
w
e
r
f
l
o
w
p
r
o
b
l
e
m
,
”
P
r
o
.
T
e
c
h
.
,
v
o
l
.
6
,
p
p
.
2
5
5
–
2
6
4
,
2
0
1
2
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
p
r
o
t
c
y
.
2
0
1
2
.
1
0
.
0
3
1
.
[49]
R.
V.
Rao
,
V.
J.
Savsani,
an
d
D.
P.
Vakha
r
ia
,
“
T
eachi
ng
–
l
ea
rning
-
b
ase
d
o
pti
m
iz
ation:
a
novel
m
et
h
od
f
or
constra
in
ed
m
echani
c
al
d
esign
o
pti
m
iz
ation
prob
le
m
s,”
Comput
e
r
-
Ai
ded
Design
,
vol.
43
,
no
.
3
,
pp
.
303
–
315
,
2011
.
[50]
R.
V.
Ra
o,
V.
J.
Savsani
,
and
J.
Balic
,
“
Teac
hi
ng
–
le
arn
ing
-
bas
ed
opti
m
izati
on
al
gorit
hm
for
u
nconstra
in
ed
an
d
constra
in
ed
r
ea
l
-
par
am
et
er
opt
imization
prob
le
m
s,”
Eng
.
Opt
.
,
vo
l
.
44
,
no
.
12
,
pp
.
1447
–
1462,
201
2.
[51]
R.
V.
Rao
and
V.
Pate
l
,
“
An
improved
te
ac
h
in
g
-
le
arn
ing
-
base
d
opti
m
iz
ation
algorithm
for
solving
unconstra
in
e
d
opti
m
iz
ation
pro
ble
m
s,”
S
ci
en
ti
a
Iranica
,
vo
l. 20,
no.
3
,
pp
.
7
10
–
7
20,
2013
,
doi
:
10
.
1016/j.sci
en
t.
20
12.
12.
005
.
[52]
S.
P.
Das,
N.
S.
Acha
r
y
,
and
S.
Padh
y
,
“
Novel
h
y
brid
SV
M
-
TL
BO
fore
ca
st
ing
m
o
del
in
cor
pora
ting
dimensionali
t
y
red
uction
t
ec
hn
i
ques,
”
Appl.
In
tell.
,
vol
.
45
,
no.
4
,
pp
.
1148
–
1165
,
2016
,
doi:
10.
10
07/s10489
-
016
-
0801
-
3
.
[53]
R.
Sen,
M.
Chattopadh
y
a
y
,
and
N.
Sen,
“
An
eff
ic
ie
n
t
appr
oa
ch
t
o
deve
lop
an
in
t
rusion
det
ection
s
y
stem
base
d
on
m
ult
i
lay
e
r
b
ac
k
propa
gation
neur
al
n
et
work
a
lgor
it
hm
:
IDS
using BPNN
al
gorit
hm
,
”
Proceedi
ngs
o
f
the
2015
ACM
SIGMIS Conf
ere
nce
on
Compute
rs
and
P
eople R
ese
arch
,
2015,
p
p.
105
–
108
,
doi
:
10.
1145/275195
7.
2751979
.
[54]
A.
Shukla,
S.
Kum
ar,
and
H.
Singh,
“
AN
N
base
d
ex
ec
u
ti
on
ti
m
e
pr
edicti
on
m
odel
and
ass
essm
ent
of
inpu
t
par
amete
rs
throu
gh
ISM
.
,
”
Int
.
A
rab J
.
Inf. Tec
hn
ol.
,
vol
.
17
,
no
.
5
,
pp
.
683
–
691
,
2
020
,
doi
:
10
.
340
28/i
ajit/
17
/5/
1
.
[55]
A.
Sulta
na
and
M.
A.
Jabba
r,
“
I
nte
lligent
n
et
wo
rk
int
rusion
dete
ct
ion
s
y
s
te
m
using
dat
a
m
ini
ng
t
ec
hniqu
es,
”
201
6
2nd
Inte
rnationa
l
Confe
renc
e
on
Appl
ie
d
and
The
oreti
cal
Comput
ing
and
Comm
unic
ati
on
Te
chnology
(
iCA
TccT)
,
2016,
pp
.
329
-
3
33,
doi
:
10
.
1109
/ICATCCT
.
2016
.
7912017
.
[56]
P.
Ta
o,
Z.
Sun
a
nd
Z.
Sun,
“
An
Im
prove
d
Intrusion
Dete
c
ti
on
Al
gorit
hm
Ba
sed
on
GA
and
SVM,”
IEEE
Acce
ss
,
vol.
6
,
pp
.
13624
-
13631,
2018
,
d
oi:
10
.
1109/AC
CESS
.
2018.
281
0198
.
[57]
Y.
Ti
an
,
M.
Mir
za
bagh
eri,
S.
M.
H.
Bamakan,
H
.
W
ang,
and
Q.
Qu,
“
Ramp
loss
one
-
class
support
vector
m
ac
hin
e
;
a
robust
and effect
iv
e
appr
o
ac
h
t
o
anomal
y
d
etec
ti
on
prob
le
m
s,”
Neurocomputi
ng
,
vol
.
310
,
pp
.
22
3
–
235,
2018
.
[58]
A.
H.
Ali
,
Z.
F
.
Hus
sain,
and
S.
N.
Abd,
“
Big
Data
Cl
assificat
ion
Ef
ficien
c
y
Based
on
Li
n
e
ar
Discriminant
Anal
y
sis
.
”
Iraq
i
Journal
For
Computer
Scienc
e
and
Math
emati
cs,
vol.
1
,
no.
1,
pp.
7
-
12
,
2020,
do
i
:
10.
52866/ijcsm
.
2019.
01.
01
.
001
.
[59]
S.
N.
Abd
,
M.
Alsajri
and
H
.
R.
Ibra
he
em,
“
Rao
-
SV
M
Mac
hine
L
ea
rning
Algorit
hm
for
Intrusion
Detect
i
on
S
y
stem.
”
Iraqi
Journal
For
Computer
Sci
e
nce
and
Mathem
ati
cs
,
vol
.
1
,
no.
1,
pp.
23
-
27
,
2020
,
do
i
:
10.
52866/ijcsm
.
2019.
01.
01
.
004
.
[60]
M.
A.
Moham
me
d,
I
.
A.
Moha
m
m
ed,
R.
A.
Hasan,
N.
Ţă
puş,
A.
H.
Ali,
and
O.
A.
Ham
m
oo
d,
“
Gree
n
Ener
g
y
Source
s:
Iss
ues
and
Cha
ll
eng
e
s,”
2019
18th
RoE
duNet
Conf
ere
nce
:
Net
wor
ki
ng
in
Educ
a
tion
and
Re
searc
h
(
RoE
duNet
)
,
201
9,
pp
.
1
-
8
,
doi
:
1
0.
1109/ROEDUN
ET
.
2019.
8909
595
.
[61]
M.
A.
Mohamm
ed,
Z.
H.
Sali
h
,
N.
Ţă
puş,
and
R.
A.
K.
Hasan,
“Sec
urity
and
acc
ounta
bilit
y
for
sharing
the
data
stored
in
the
cl
o
ud,
”
2016
15th
RoE
duNet
Con
fer
enc
e:
Ne
tworking
in
Educ
at
ion
and
Re
search
,
2
016,
pp.
1
-
5,
doi
:
10.
1109/RoE
du
Net.
2016.
77532
01
.
[62]
M.
A.
Moham
m
ed
and
N.
Ţ
ăpu
ş
,
“
A
novel
app
roa
ch
of
red
u
cing
ene
rg
y
cons
um
pti
on
b
y
utili
zi
ng
ent
ha
lp
y
in
m
obil
e
cl
oud
computing,
”
St
udie
s
in
Informatic
s
and
Control
,
vol.
2
6,
no.
4,
pp.
425
–
434,
2017
,
doi:
10
.
24846/v
26i4
y
201706
.
[63]
N.
Q.
Moham
me
d,
M.
S.
Ahmed,
M.
A.
Moham
m
ed,
O.
A.
Ham
m
ood,
H.
A.
N.
Alshara
and
A.
A.
Kam
il
,
“
Com
par
at
ive
Anal
y
sis
bet
we
en
Solar
and
W
ind
Turbi
ne
Ene
rg
y
Sourc
es
in
Io
T
Based
on
Ec
onom
ic
al
and
Eff
iciency
Cons
ide
ra
ti
ons,”
201
9
22nd
Inte
rnational
Confe
ren
c
e
on
Control
Syste
ms
and
Computer
Sci
ence
(
CSCS
)
,
2019,
p
p.
448
-
452
,
doi
: 10.1109/
CS
CS
.
2019.
00082
.
[64]
R.
A.
I.
Alha
y
al
i
,
M.
A.
Ahm
ed,
Y.
M.
Mohial
de
n,
and
A.
H.
Ali,
“
Eff
ic
ie
nt
m
et
h
od
for
bre
ast
ca
n
ce
r
class
ifi
cation
base
d
on
ense
m
ble
hoff
edi
ng
tr
e
e
and
n
aï
v
e
Ba
yes,
”
Indon
esian
Journal
of
El
e
ctr
ic
al
Eng
ineering
and
Computer
Sci
en
ce
,
vol
.
18
,
no.
2,
pp.
1074
–
1080,
2020
,
doi
:
10.
11591/ijeecs.
v18.
i2.
pp1074
-
1
080
.
[65]
Z.
H.
Sal
ih,
G
.
T.
Hasan
,
and
M.
A.
Moham
m
ed,
“
Inve
st
iga
t
e
and
an
aly
z
e
the
le
ve
ls
of
e
lectr
o
m
agne
ti
c
rad
i
ati
ons
emitt
ed
from
under
ground
po
wer
ca
b
le
s
e
x
tended
in
m
oder
n
ci
t
ie
s,
”
2017
9th
Int
ernati
o
nal
Confe
ren
ce
on
El
e
ct
ronics,
Co
mputers and
Artific
ia
l
In
te
l
li
gen
c
e
(
ECA
I)
,
2017,
pp.
1
-
4
,
doi
:
10
.
1109/E
CAI.201
7.
8166452
.
[66]
Z.
H.
Sa
li
h
,
G.
T.
Hasan
,
M.
A.
Moham
m
ed,
M.
A.
S.
Kl
ib,
A.
H.
Ali,
and
R.
A.
Ibr
ahi
m
,
“
St
ud
y
the
E
ffe
c
t
o
f
Inte
gra
ti
ng
the
Solar
Ene
rg
y
Source
on
Stabi
l
i
t
y
of
Elec
tri
c
al
Distribut
ion
Sy
stem,
”
2019
22nd
Inte
rnation
al
Confe
renc
e
on
C
ontrol
Syst
ems and
Computer
Sc
i
enc
e
,
2019
,
pp
.
443
-
447,
doi
:
10
.
1109/CSCS
.
2019.
00081
.
[67]
N.
D.
Za
ki
,
N.
Y.
Hashim
,
Y.
M.
Mohial
d
en,
M.
A.
Moham
m
ed,
T.
Sutikno,
a
nd
A.
H.
Ali
,
“
A
rea
l
-
ti
m
e
b
ig
dat
a
senti
m
ent
an
aly
s
is
for
ira
qi
twee
t
s
using
spark
strea
m
ing,
”
Bulleti
n
of
Elec
tri
cal
E
ngine
ering
and
I
nformatic
s
,
vol
.
9,
no
.
4
,
pp
.
141
1
–
1419,
2020
,
d
oi:
10
.
11591/eei
.
v9i4.
1897
.
[68]
N.
M.
Hus
sien,
Y.
M.
Mohialden,
N.
T.
Ah
m
ed,
M.
A.
Moham
m
ed,
and
T.
Sutikno
,
“
A
sm
art
gas
le
akage
m
onit
oring
s
y
st
em
for
use
in
h
ospita
ls,
”
Indon
esian
Journal
o
f
Elec
tric
al
Eng
ine
ering
and
C
omputer
Scienc
e
,
vol.
19
,
no
.
2
,
pp
.
1048
–
1054
,
20
20
,
doi
:
10
.
1159
1/i
jeec
s.v19
.
i2
.
p
p1048
-
1054
.
[69]
M.
Dash
and
H.
Li
u,
“
Feat
ur
e
select
ion
for
c
la
s
sific
a
ti
on,
”
Intel
l.
data
Ana
l.
,
vol.
1,
no
.
1
–
4,
pp
.
131
–
156,
1997
,
doi:
10
.
1016/S1
088
-
467X(97)00008
-
5
.
Evaluation Warning : The document was created with Spire.PDF for Python.