Int
ern
at
i
onal
Journ
al of Ele
ctrical
an
d
Co
mput
er
En
gin
eeri
ng
(IJ
E
C
E)
Vo
l.
11
,
No.
1
,
Febr
uar
y
2021
, pp.
59
6
~
601
IS
S
N: 20
88
-
8708
,
DOI: 10
.11
591/
ije
ce
.
v11
i
1
.
pp
59
6
-
601
596
Journ
al h
om
e
page
:
http:
//
ij
ece.i
aesc
or
e.c
om
D
iffer
ential
evolu
tion detecti
on m
odels
f
or
SMS
s
pam
Sa
r
ab
M. H
am
eed
Depa
rt
m
ent
o
f
C
om
pute
r
Scie
n
ce,
Col
le
ge
of
Science
,
Univer
si
t
y
of
Baghda
d
,
Ir
aq
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
N
ov
23
, 201
9
Re
vised
Jun
17
,
20
20
Accepte
d
J
u
n
27
,
20
20
W
it
h
the
growth
of
m
obil
e
phone
s
,
short
m
essage
service
(SM
S)
bec
ame
an
essenti
a
l
t
ext
co
m
m
unic
at
ion
ser
vic
e
.
How
eve
r
,
t
he
low
cost
and
ea
se
use
of
SM
S
le
d
to
an
inc
rea
se
in
SM
S
Spam
.
In
t
his
pape
r,
the
c
har
acte
rist
ics
of
SM
S
spam
h
as
studie
d
and
a
set
of
f
ea
tu
res
has
int
roduc
ed
t
o
get
r
id
o
f
SM
S
spam
.
In
a
ddit
ion,
th
e
pro
ble
m
of
SMS
sp
am
det
ec
t
ion
was
addr
essed
as
a
c
luste
r
ing
ana
l
y
sis
tha
t
req
uire
s
a
m
eta
heur
isti
c
a
lgori
th
m
to
find
the
cl
uster
ing
str
uct
ure
s.
Thr
ee
d
iffe
ren
ti
a
l
evol
ut
ion
var
ia
nts
v
iz
DE/ra
nd/
1
,
jDE/
ran
d
/1,
jDE
/be
st/1
,
are
ado
pte
d
for
solving
the
SM
S
spam
proble
m
.
Expe
riment
al
r
e
sults
il
lustr
at
e
th
at
th
e
jDE
/be
st/
1
produc
es
best
result
s
over
othe
r
var
i
ant
s in te
rm
s of
ac
cur
a
c
y
,
f
al
se
-
posit
ive
rat
e
and
f
al
se
-
n
e
gat
iv
e
rate.
Moreove
r,
it surpass
es
the
b
ase
l
i
ne
m
et
hods.
Ke
yw
or
d
s
:
D
iffe
re
ntial
evoluti
on
F
eat
ur
e
ex
tr
act
ion
M
achine lea
rn
i
ng
SMS s
pam
cla
ssificat
ion
SMS s
pam
d
et
ect
ion
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
:
Sara
b
M.
Ham
eed
,
Dep
a
rtm
ent o
f C
om
pu
te
r
Scie
nce
,
C
ollege
of
Science
,
Un
i
ver
sit
y o
f B
aghda
d
,
Ba
ghda
d,
Ir
a
q
.
Em
a
il
:
sarab
.m
@sc.
uoba
ghda
d.
e
du.iq
1.
INTROD
U
CTION
Th
e
us
age
of
s
hort
m
essaging
serv
ic
e
(S
MS
)
becam
e
increasing
rap
i
dly
as
a
resu
lt
of
it
s
cheap
c
os
t
and
ease
of
use
.
This
directe
d
to
a
n
i
ncr
eas
e
in
the
am
ount
of
s
pam
,
wh
i
ch
is
un
s
olici
te
d
a
nd
unde
sire
d
SMS
sent
to
a
la
r
ge
nu
m
ber
of
rec
ei
ver
s.
The
pr
e
ci
se
de
finiti
on
of
s
pam
do
es
no
t
e
xist.
Esse
ntial
ly
,
sp
am
r
egards
as an
un
desira
ble em
ai
l, h
owever, it
is not a
ll
u
ndesi
ra
ble e
-
m
ails ar
e sp
a
m
.
SMS sp
am
h
as a sign
ific
ant inf
l
uen
ce
on
use
rs
beca
us
e u
s
ers
view
eac
h
SMS they get, so
SMS sp
am
aff
ect
s
the
use
rs
im
m
ediat
e
ly
[1
,
2].
In
add
it
io
n,
to
di
sturbin
g,
us
ers
require
a
de
gr
ee
of
sec
rec
y
with
their
m
ob
il
e
phones
a
nd
un
a
ff
ect
ed
by
s
pa
m
and
vir
us
es
intru
si
ons
[
3
-
5]
.
Ther
e
f
or
e,
c
on
si
der
a
ble
at
te
ntio
n
to
the
SMS
s
pam
pr
oble
m
is
pro
vid
e
d
t
o
de
velo
p
a
set
of
a
ppro
a
c
he
s
to
bypas
s
th
is
pro
blem
.
Am
ong
the
ap
proach
e
s
dev
el
op
e
d
to
com
bat
the
SMS
sp
am
i
s
a
cl
assifi
cation
of
m
essa
ges
into
SM
S
sp
am
and
ham
.
The
chall
en
ge
is
t
hat
the
m
essages
a
re
s
hort
and
co
ntains
f
ew
words
a
nd
these
wor
ds
m
ay
be
abbre
viate
d [6
,
7].
Detect
ing
SM
S
sp
am
turn
s
to
be
of
sig
nifi
cant
worth
due
to
the
huge
loss
that
co
uld
resu
lt
from
the
SMS
s
pam
.
Tw
o
ty
pe
s
of
m
et
ho
ds
that
a
re
c
ollaborat
iv
e
base
d
an
d
c
onte
nt
-
base
d
m
et
hods
ca
n
be
use
d
t
o
detect
SMS
s
pam
.
Coll
abo
r
at
ive
base
d
de
pends
on
us
a
ge
and
us
e
r
ex
per
ie
nce.
Wh
il
e
the
co
ntent
-
base
d
m
et
ho
d
co
nce
ntrates
on
exa
m
ining
the
te
xtu
al
co
ntent
of
m
essages
[8
,
9].
Zai
nal
and
Jal
i
perf
orm
ed
a
stud
y
of
t
he
disti
nguish
i
ng
con
t
ro
l
of
t
he
f
eat
ur
es
a
nd
ex
a
m
ining
it
'
s
info
rm
at
ion
al
or
i
m
pact
ci
rcu
m
s
ta
nce
in
SMS
sp
am
m
essages
cl
assifi
cat
ion
[
10
]
.
Kaya
and
E
rtu
ğrul
intr
oduce
d
a
m
e
tho
d
ba
sed
on
local
te
rn
a
ry
patte
rn
s
t
o
ext
ract
two
disti
nc
t
featur
es
set
low
a
nd
up
fe
at
ur
es
f
r
om
S
MS
m
essages
and
se
ve
ral
m
achine
le
arn
in
g
m
et
h
od
s
wer
e
a
ppli
ed
f
or
cl
assify
ing
SMS
s
pam
.
The
eval
uation
resu
lt
s
ove
r
three
s
epa
rate
SMS
dataset
s g
ai
ne
d ac
cu
racy 9
3.3
18%,
87.
15
%
, a
nd 94.1
0% [11]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Diff
erenti
al ev
olu
ti
on
detect
ion m
odel
s for
SM
S
s
pam
(
Sa
ra
b M.
Ha
mee
d)
597
Su
la
im
an
and
Jal
i
intro
duce
d
a
m
et
ho
d
for
detect
ion
SM
S
sp
am
based
on
well
-
kn
own
cha
racters
us
e
d
at
tra
ns
m
it
ti
ng
SMS,
S
MS
le
ngth
a
nd
keyw
ords
.
Fi
ve
dif
fer
e
nt
al
gorithm
s
an
d
th
ree
diff
e
re
nt
da
ta
set
s
wer
e
us
e
d
to
ver
ify
t
he
pro
po
s
ed
m
et
ho
d.
T
he
resu
lt
s
i
nd
ic
at
ed
that
t
he
pro
po
se
d
m
et
ho
d
co
uld
pro
du
ce
a reas
on
a
ble
de
te
ct
ion
r
at
e
[
12]
.
Chou
dh
a
ry
an
d
Jai
n
prese
nted
a
n
ap
proac
h
for
detect
ing
s
pam
m
essages
us
in
g
te
n
feat
ures
a
nd
fiv
e
m
achine
le
ar
nin
g
al
gorithm
s
viz
naï
ve
Ba
ye
s,
lo
gisti
c
re
gressi
on,
J
48,
de
ci
sion
ta
ble
an
d
ra
ndom
fores
t.
SMS
Sp
am
Corpus
was
us
ed
as
a
n
evaluati
on
dat
aset
and
the
re
su
lt
s
sho
wed
that
the
rand
om
forest
is
the
be
st
on
e
with
detect
ion
rate eq
uals
96.
1% [13]
.
Abd
ulh
a
m
id
et
al
.
prese
nted
a
re
view
of
th
e
possible
way
s
an
d
c
halle
ng
es
of
s
pam
detect
ion
a
nd
fu
t
ur
e
resea
rc
h
directi
on
t
hat
ca
n
help
sp
eci
al
ist
re
searche
rs
t
o
know
the
are
as
that
need
f
ur
th
e
r
i
m
pr
ovem
ent
[1
]
.
Kali
ya
r
et
al
.
pr
opos
e
d
a
general
m
od
e
l
to
disti
nguis
h
SMS
sp
am
m
essages
ba
se
d
on
sever
al
m
achine
le
arn
i
ng
al
gorithm
s.
The
pro
po
se
d
SMS
sp
am
fil
te
ring
m
od
el
was
ab
le
to
filt
er
m
e
ssages
from
sever
al
fa
m
ilies:
Sing
apore,
Am
erican,
an
d
I
nd
ia
n
En
glish.
T
he
r
esults
sh
owe
d
the
pr
op
os
e
d
m
od
el
achieve
d
a
h
i
gh
pr
eci
sio
n usi
ng In
dian
E
ngli
sh
SMS
[14
]
.
Sh
a
raff
pr
ese
nt
ed
a
com
par
at
ive
stud
y
of
the
eff
ect
of
featur
e
sel
ect
ion
te
ch
niques
on
sev
era
l
cl
assifi
ers.
The
resu
lt
s
showe
d
that
the
featur
e
s
el
ect
ion
t
echn
i
qu
e
im
pacts
the
per
f
orm
ance
of
the
c
la
ssifie
r
and
can
assist
in
e
nh
a
ncin
g
t
he
perf
or
m
ance
of
so
m
e
cl
assifi
ers
[
15
]
.
O
jugo
a
nd
Eb
oka
de
velo
ped
SMS
sp
am
detect
ion
m
et
ho
d
base
d
on
B
ay
esi
an
Netw
ork
.
A
set
of
fea
tures
is
determ
inist
ic
al
ly
sel
e
ct
e
d
f
ro
m
SMS
us
i
ng
the
ge
netic
al
go
rithm
(G
A
).
The
res
ults
sho
wed
t
hat
the
fe
at
ur
e
sel
ect
ion
te
chn
iq
ue
im
pacts
the
perfor
m
ance
of the classi
fie
r
[
16]
.
A
tw
o
-
f
old
c
ontrib
utio
n
is
c
oupled
in
t
his
pa
per.
Fir
st,
we
raise
the
f
ollow
i
ng
quest
ion
:
How
t
o
desig
n
a
n
ef
fici
ent
featu
re
e
xtracti
on
m
od
el
to
i
m
pr
ove
the
accu
racy.
Seco
nd,
to
e
nsure
m
or
e
ef
fic
aci
ou
s
perform
ance,
we
ex
plo
it
m
ulti
ple
var
ia
nts
of
diff
e
ren
t
ia
l
evo
luti
on
(D
E
)
m
od
el
s
that
of
fe
r
posit
iv
e
coope
rati
on
w
it
h
a
set
of
f
eat
ur
es
for
S
MS
sp
am
to
correct
ly
div
i
de
the
sea
rch
sp
ace
int
o
two
a
nd
non
-
overla
pp
i
ng classe
s
SMS
sp
am
an
d ham
.
The
rest
of
t
his
pa
per
is
or
ga
nized
a
s
f
ollo
ws.
Sect
io
n
2
el
aborates
on
an
a
ppr
oach
f
or
detect
in
g
SMS
sp
am
by
su
ggest
in
g
of
a
featur
e
set
.
I
n
Sect
io
n
3,
th
e
resu
lt
s
a
nd
di
scus
sio
n
of
th
e
pro
po
se
d
a
ppr
oac
h
are
r
ep
or
te
d.
Finall
y,
the
c
oncl
udin
g
rem
a
rk
s
an
d
f
uture
w
ork
f
or
othe
r
re
searc
h
dir
ect
ion
s
a
re
pr
ov
i
ded
in secti
on
4
.
2.
RESEA
R
CH MET
HO
D
This
sect
ion
cl
arifies
the
m
e
t
hodolo
gy
us
e
d
fo
r
SMS
s
pa
m
detect
ion
wh
ic
h
co
ns
ist
s
of
tw
o
sta
ges
.
The
first
sta
ge
at
tem
pts
to
extract
fr
om
SM
S
sp
eci
fic
feat
ur
es
that
m
ay
be
us
e
d
to
cha
racteri
ze
SMS
sp
am
.
Ba
sed
on
the
extracte
d
feat
ures,
three
dif
fere
ntial
evo
luti
on
al
og
irt
hm
s
ar
e
us
ed
in
the
s
econd
sta
ge
to
detect
SMS
sp
am
.
Con
si
der
a
n
SM
S
co
ll
ect
ion
of
te
xt
m
essages,
i.e.
=
{
1
,
…
,
}
.
Each
m
essage
,
,
1
≤
≤
con
sist
s
of
w
ords
,
num
ber
s,
et
c.
and
it
s
siz
e
is
lim
it
ed
to
160
c
har
act
er
s.
Figure
1
dem
on
st
rates
so
m
e
sam
ples o
f
SM
S sp
am
and
ha
m
in
[
17
]
.
Figure
1.
Sam
ple o
f
SMS s
pa
m
an
d
ham
fr
om
SMS
s
pam
c
ollec
ti
on
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
1
1
, No
.
1,
Febr
uar
y
2021
:
59
6
-
6
0
1
598
The
idea
pro
po
sed
he
re
for
de
al
ing
with
the
SMS
sp
am
pr
oble
m
is
based
on
var
ia
nt
D
E
al
gorithm
s.
Up
t
o
the
best
of
our
knowle
dg
e
,
this
is
t
he
fir
st
tim
e
t
o
use
DE
al
go
rithm
s
as
cl
us
te
rin
g
f
or
SMS
sp
am
.
The
SMS
s
pam
detect
ion
ta
sk
is
t
o
i
den
ti
fy
w
heth
er
is
SMS
s
pam
or
ham
us
i
ng
cl
us
te
r
ing
is
ℂ
∶
→
{
,
ℎ
}
.
Detect
in
g
di
sti
nctive
feat
ures
t
hat
hav
e
hi
gh
discri
m
inatory
pow
er
to
disti
nguish
bet
ween
SMS
s
pa
m
and
ham
to
be
cr
ucial
in
S
MS
sp
am
detect
ion
.
T
o
s
upport
the
cl
us
te
ri
ng
a
set
of
featur
es
=
{
1
,
2
,
…
,
}
s
houl
d be e
xtract
ed fr
om
.
This r
esea
rch
i
nv
e
sti
gates
th
e
chall
en
ge
of
S
MS
s
pam
detect
ion
t
hat
is
ho
w
t
he
disti
nctive featu
res
are
ide
ntifie
d
t
o
cl
assify
the
s
pam
SMS
m
es
sages.
I
n
this,
the
feat
ur
es
o
f
the
ham
and
SMS
s
pam
m
e
ssages
are
extracte
d
t
o
form
featur
es
set
rep
rese
nting
a
sig
n
for
the
corres
ponding
m
essage.
These
featu
res
a
re
us
e
d
to
trai
n
DE
f
or
identify
in
g
t
he
centr
oid
s
of
t
he
SMS
sp
am
and
ham
.
The
n,
the
trai
ne
d
ce
ntr
oid
s
a
re
give
n
to
the te
sti
ng stag
e to
detect
SM
S sp
am
o
r
h
am
.
The
first
sta
ge
of
the
pro
pose
d
SMS
sp
a
m
detect
ion
m
od
el
is
pr
epr
oc
c
es
sin
g.
A
f
te
r
retrievi
ng
the
SMS
m
es
sage
from
a
data
s
ource,
te
xt
pr
e
process
ing
that
is
ve
ry
im
po
rtant
i
s
ap
plied
to
m
ak
e
the
m
essages
t
o
be
a
naly
zed.
Sever
al
ste
ps
are
invol
ved
in
pr
e
processi
ng
in
orde
r
to
pr
epar
e
t
he
m
ess
age
t
o
an
SMS
sp
am
detect
ion.
Fir
stl
y,
the
m
ess
age
is
c
onver
t
ed
to
l
ow
e
rcas
e
to
eva
de
dif
fer
e
ntiat
ion
be
tween
the
sam
e
wo
rds
that
are
di
ff
e
ren
t
in
case
.
S
econdly
,
the
punct
uation
is
r
e
m
ov
ed
.
A
fter
that,
the
m
essa
ges
ar
e
tok
e
nized
base
d
on
delim
i
te
d
wh
it
es
-
pace
to
identify
the
t
ok
e
ns.
The
n,
s
top
w
ords
a
re
exclu
ded
from
the
set
of
t
ok
e
ns
recogn
iz
e
d
in
t
he
f
or
m
er
ste
p
to
ob
ta
in
a
m
essa
ge
as
a
li
st
of
keyw
ords.
Fin
al
ly
,
the
tok
e
ns
are
stemm
ed
to r
ec
ognize thei
r root
s
.
On
ce
a
pp
ly
in
g
pr
e
processin
g
to
the
colle
ct
ion
of
m
essages
,
the
m
essage
rep
rese
ntati
on
sta
ge
shoul
d
be per
form
ed
to r
e
present t
he m
essage b
y
dis
ti
nct term
s.
Each term
then
is
involve
d
a
w
ei
gh
t cal
c
ulate
d usin
g
te
rm
-
fr
equ
e
n
c
y
inv
e
rse
-
se
nte
nce
-
fr
e
quency.
Af
te
rwo
rd,
ea
ch
m
essage
is
r
epr
ese
nted
as
a
vecto
r
of
wei
ghts o
f
te
rm
s.
In
ad
di
ti
on
,
s
om
e
featur
es
a
re
e
xtracted
from
the
ra
w
dat
a
to
f
ur
t
her
i
m
pr
ov
e
the
m
essage
represe
ntati
on
as ex
plaine
d
in
what f
ollo
ws:
-
The
le
ng
t
h
of
the
m
essage
is
co
ns
ide
red
one
featu
re
t
o
be
ad
ded
to
re
present
the
m
ess
age
si
nce
SM
S
sp
am
s leng
th t
ends to
b
e
lo
nger th
a
n ham
.
-
The
prese
nce
of
a
num
ber
is
con
si
der
e
d
a
s
on
e
featu
re
to
be
ad
de
d
to
re
pr
ese
nt
th
e
m
essage
sin
ce
the s
pam
m
er pref
e
rs usin
g
t
he
phon
e
num
ber
to
claim
a lotte
ry or
pr
iz
e.
The
ext
racted
f
eat
ur
e
set
can
be
us
e
d
by
the
DE
to
de
fine
t
he
centr
oid
of
each
cl
us
te
r
.
T
he
pro
posed
DE
SM
S s
pam
d
et
ect
ion c
on
s
ist
s o
f
t
wo stag
es:
trainin
g
a
nd test
in
g
sta
ge
s
. T
he go
al
of th
e trai
ning sta
ge
is to
tun
e
t
he
val
ue
of
t
w
o
cen
troid
vecto
rs
accor
ding
to
a
set
=
{
1
,
2
,
…
,
}
of
tra
ining
m
essages.
Wh
il
e
the
goa
l
of
t
he
te
sti
ng
sta
ge
is
to
c
la
ssify
the
inc
om
ing
m
essage
into
SM
S
spam
or
ham
based
on
the
tw
o
cent
roi
d
vect
or
s
pro
du
ce
d
from
the
trai
n
in
g
sta
ge
.
The
f
ollow
i
ng
subsect
io
ns
il
lustrate
how
DE
i
s
util
iz
ed
f
or the
cl
us
te
rin
g pur
pose
base
d on th
e extracte
d feat
ur
es
.
2.1
.
DE
based
c
lusteri
n
g fo
r
SMS sp
am
de
tection
Af
te
r
ext
racti
ng
the
m
essage
featu
res
set
,
an
op
ti
m
iz
ation
m
od
el
base
d
on
DE
t
hat
us
es
thes
e
featur
e
s
to
identify
the
centro
ids
f
or
cl
ust
ering
of
SM
S
sp
am
and
ha
m
is
pr
opose
d
.
Th
ree
dif
fe
ren
ti
al
evo
l
ution
var
i
ants
nam
ely
DE/ra
nd
/
1,
jD
E/ran
d/1
,
j
DE/
best/1
a
re
util
iz
ed.
T
he
det
ai
le
d
ex
planat
ion
of
the m
ai
n
char
a
ct
erist
ic
co
m
po
nen
ts
o
f
the
pr
opos
e
d
m
od
el
s
is provide
d
i
n wh
at
f
ollow
s
.
2.2.1
.
I
ndivid
ua
l repre
sent
ati
on and p
opu
lation ini
tializ
at
i
on
In
t
he
pro
pos
ed
m
od
el
s,
cl
us
te
r
ce
ntr
oids
are
e
nc
od
e
d
in
the
in
div
i
du
al
.
Eac
h
in
div
id
ual
is
represe
nted
as
a
ve
ct
or
of
ge
nes
that
c
or
res
ponds
to
the
2
c
luster
centr
oids.
T
he
m
at
he
m
at
ic
al
form
ulatio
n
of the
i
nd
i
vidu
al
is as f
ollo
ws
:
=
{
1
,
2
,
…
,
}
,
∀
∈
{
1
,
…
,
}
,
∈
[
,
]
: i
s the m
ini
m
u
m
v
al
ue
featu
r
e
can
g
et
,
a
nd
:
i
s the m
axi
m
um
v
al
ue
feat
ure
can
ge
t.
DE
is
a
popul
at
ion
-
base
d
op
tim
iz
at
ion
al
gorithm
and
sta
rts
wit
h
a
po
pula
ti
on
ℙ
of
so
luti
ons.
Fo
rm
al
ly
sp
eakin
g,
ℙ
can
b
e
for
m
ulate
d
as
f
ollow
s:
ℙ
=
{
1
,
2
,
…
,
}
w
he
re
is t
he p
opulati
on size.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Diff
erenti
al ev
olu
ti
on
detect
ion m
odel
s for
SM
S
s
pam
(
Sa
ra
b M.
Ha
mee
d)
599
2.2.2
.
Mut
at
io
n and cr
os
s
ov
er op
er
ati
on
s
This
op
e
rati
on
is
consi
der
e
d
a
s
the
m
ai
n
op
e
rati
on
t
hat
is
re
sp
onsi
ble
f
or
m
ai
ntaining
th
e
popula
ti
on
div
e
rsity
in
ev
olu
ti
on.
T
he
m
ut
a
ti
on
eq
ua
ti
on
s
for
DE/r
and
/
1,
jDE/ra
nd/1
a
nd
jDE/
be
st/
1
are
i
n
(
1
)
,
(
2
)
,
and
(
3
)
res
pect
ively
[
18
-
20]
.
DE/ra
nd
/
1
=
1
+
(
2
−
3
)
(1)
j
D
E/ran
d/1
=
1
+
(
2
−
3
)
(2)
j
D
E/best/
1
=
+
(
1
−
2
)
(3)
w
he
re
1
,
2
an
d
3
are
ra
ndom
nu
m
ber
s
in
[
1,N]
an
d
they
a
re
m
utu
al
ly
diff
e
re
nt,
is
the
in
div
i
du
al
i
n
the curre
nt
popula
ti
on
.
is a ra
ndom
n
um
ber
i
n (0,1
)
a
nd
=
+
×
<
1
is l
ower
boun
d of
m
utati
on
.
is u
pper
boun
d
of m
utati
on
.
1
is t
he
pro
bab
il
it
y t
o
al
te
r
fa
ct
or
.
The
cr
os
s
over
op
e
rati
on
ai
m
is
to
com
bin
e
the
ta
rg
et
ve
ct
or
with
the
dono
r
vect
or
to
pro
duce
the
tria
l
vect
or
.
E
qu
at
io
n
(
4
)
[
21
]
il
lustrate
s
the
c
ro
ss
over
oper
at
ion
f
or
DE/
rand/1
a
nd
(
5
)
[
22
]
dem
on
strat
es t
he
c
ro
s
sover
ope
rati
on
for j
D
E/ran
d/1
a
nd
jDE/best/
1.
,
=
{
,
1
≤
=
,
2
,
1
>
≠
,
2
(4)
,
=
{
,
1
≤
=
,
2
,
1
>
≠
,
2
(5)
wh
e
re
is
a
ra
ndom
nu
m
ber
in
(
0,1]
that
dete
rm
ines
t
he
value
of
the
tria
l
ve
ct
or
whi
ch
is
i
nh
e
rite
d
from
the do
nor
.
1
is a ra
ndom
num
ber
s in (
0,1]
2
is a ra
ndom
num
ber
s in [
1,n]
a
nd
=
<
2
2
is t
he pr
obabil
it
y t
o
al
te
r
CR
factor.
2.2.3
.
E
valua
tion
DE selec
tio
n opera
tio
n
Accor
ding
to
the
SMS
s
pam
pro
blem
,
the
fo
rm
ulati
on
of
t
he
obj
ect
ive
f
unct
ion
re
quires
m
axi
m
iz
ing
the
accu
racy
that
m
eans
the
nu
m
ber
of
m
essage
s
that
are
co
rr
ect
ly
detect
ed
as
SMS
sp
am
(
)
and
the
num
ber
of
m
essages
ar
e
correct
ly
cl
assifi
ed
as
ha
m
(
)
sho
uld
be
m
axi
m
iz
ed
.
In
oth
e
r
wor
ds
the
num
ber
of
m
essages
th
at
are
m
isclassifie
d
as
SM
S
sp
am
(
)
an
d
t
he
nu
m
ber
of
m
ess
ages
are
m
isc
la
ssifie
d
as
ham
(
)
sh
ould
be
m
ini
m
iz
ed.
Each
in
div
i
du
a
l
is
evaluated
usi
ng
the
obj
e
ct
ive
functi
on
a
s
in
(
6
)
.
T
he
c
om
pu
ta
ti
on
of
t
he
ob
j
ect
ive
f
un
ct
io
n
f
or
ea
ch
in
div
id
ual
is
as
fo
ll
ows.
First,
the
ce
ntro
i
ds
,
ℂ
=
{
,
ℎ
}
enco
de
d
in
t
he
ind
i
vidual
are
extracte
d
an
d
t
he
cl
us
te
rs
are
form
ed.
The
n,
the
cl
us
te
r
is
at
ta
ined
by
assigni
ng
t
he
m
essages
to
a
cl
us
te
r
co
rr
es
pondin
g
to
the
cl
os
est
centr
oid
.
E
uclidea
n
s
qu
a
re
d
distance m
et
ric
is ad
opte
d for
the co
m
pu
ta
ti
on
of the
dis
ta
nc
e b
et
wee
n
a
m
essage a
nd the
centr
oid
.
(
)
=
+
+
+
+
(6)
Af
te
r
e
valu
at
ion
t
he
in
div
i
dual
s,
DE
sel
ec
ti
on
ope
rati
on
is
app
li
ed
as
s
how
n
in
(
7
)
,
t
he
res
ultant
vecto
r
from
th
is
op
erati
on
is
the
vector
w
it
h
the
higher
obj
ect
ive
f
unct
ion
that
will
be
passed
to
th
e
next
gen
e
rati
on.
=
{
(
)
>
(
)
ℎ
(7)
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
1
1
, No
.
1,
Febr
uar
y
2021
:
59
6
-
6
0
1
600
3.
E
X
PERI
MEN
TAL RES
UL
TS A
ND DIS
CUSSIO
N
SMS
Sp
am
Coll
ect
ion
[17c]
is
a
public
data
set
colle
ct
ed
for
SMS
sp
am
research
t
hat
co
nt
ai
ns
5,574
En
glish,
real
and
non
-
en
co
de
d
m
essages.
E
ach
m
essage
is
la
beled
to
in
dicat
e
wh
et
her
a
giv
e
n
m
ess
age
is
a
le
gitim
a
te
(h
a
m
)
or
SMS
sp
am
.
It
con
ta
ins
ab
ou
t
86.
6
7%
of
ham
and
13
.
33%
of
S
MS
sp
am
.
Evaluati
on
m
et
rics
represe
nted
by
acc
ur
a
cy
(
)
[23], f
al
se ne
gative rate
(
)
[24]
an
d
false posit
ive
rate
(
)
[
25]
are
us
e
d
for
as
sessing
the
propose
d
SMS
s
pam
.
Evaluati
on
of
the
pro
pose
d
SMS
s
pa
m
detect
ion
m
od
el
is
perform
ed
by
app
ly
in
g
the
k
-
f
old
cr
os
s
-
validat
ion
a
ppr
oa
ch.
I
n
this
paper,
3
-
f
old
c
ros
s
-
validat
io
n
a
ppr
oac
h
and
2
-
fo
l
d
cro
ss
-
validat
ion
are
adopted
to
sh
ow
th
e
i
m
pact
of
trai
nin
g
a
nd
te
sti
ng
dataset
on
the
perform
ance
of
the
pr
opos
e
d
m
od
el
.
T
he
e
val
uation
is
prese
nted
in
te
rm
s
of
a
verage
a
ccu
racy
(
)
,
false
ne
gative
r
at
e
(
)
and
false
posit
ive
rate
(
)
over
k
-
f
old.
Ta
bl
e
1
re
ports
the
set
ti
ng
pa
ram
e
te
rs
of
DE
m
od
el
s.
Tables
2
an
d
3
pr
ese
nt
the
res
ults
of
the
pro
po
s
ed
a
ppro
a
c
h
unde
r
3
-
f
old
and
2
-
fo
l
d
f
or
var
io
us
cl
assifica
ti
on
.
Fr
om
tho
se
t
wo
ta
bles,
w
hi
ch
pr
esent
th
e
aver
a
ge
pe
r
form
ance
of
the
thr
ee
va
riant
D
E
al
gorithm
s
DE/rand/1,
jDE/ra
nd
/
1,
jDE/best
/1
agai
ns
t
k
-
m
eans
baseli
ne
m
et
ho
d
i
n
te
r
m
s
of
,
a
nd
ov
e
r
te
n
dif
f
eren
t
r
uns,
it
is
cl
ear
that
the
th
ree
va
ria
nt
DE
m
od
el
s
hav
e
su
cce
ss
fu
ll
y
cl
us
te
re
d
the
m
essages
i
nto
SMS
s
pam
and
ham
fo
r
the
dif
fe
ren
t
k
-
fo
l
d
ap
proac
he
s
against
k
-
m
e
ans
th
rou
gh
re
du
ci
ng
a
nd
a
nd
inc
r
easi
ng
the
acc
ur
acy
.
Also,
it
ca
n
be
ob
se
r
ve
d
that
t
he
jD
E/best/
1
pro
du
ces
the
best
resu
l
t
ov
e
r oth
er
var
ia
nt
DE
m
od
el
s
.
Table
1.
DE
p
a
ram
et
ers
s
et
ti
ng
Para
m
eter
Valu
e
Po
p
u
latio
n
size,
100
Maxi
m
u
m
nu
m
b
er
of
gen
eration
s
100
Cro
ss
o
v
er
p
rob
ab
il
ity
,
0
.9
Mutatio
n
pro
b
ab
ility,
0
.5
Lower bo
u
n
d
of
m
u
tatio
n
,
0
.1
Up
p
er
b
o
u
n
d
o
f
m
u
tatio
n
,
0
.9
p
rob
ab
ilit
y
to alter
,
1
0
.1
p
rob
ab
ility
to
alter
,
2
0
.1
Table
2.
C
om
par
at
ive
r
esults
o
btaine
d
o
ver
10
i
ndepe
ndent
r
un
s
for
t
he
p
r
opos
e
d
m
od
el
against
K
-
m
eans
r
e
gardin
g 3
-
f
ol
d
Mod
el
DE/r
an
d
/1
0
.96
4
1
0
.01
1
6
0
.19
2
5
jDE/r
an
d
/1
0
.96
6
0
0
.01
2
2
0
.17
4
2
jDE/bes
t/1
0
.96
6
7
0
.01
1
6
0
.17
3
8
k
-
m
eans
0
.86
0
0
0
.06
8
2
0
.60
2
4
Table
3.
C
om
par
at
ive
r
esults
o
btaine
d
o
ver
10
i
ndepe
ndent
r
un
s
for t
he
p
r
opos
e
d
m
od
el
against
K
-
m
eans
r
e
gardin
g 2
-
f
ol
d
Mod
el
DE/r
an
d
/1
0
.96
4
8
0
.01
2
8
0
.18
0
1
jDE/r
an
d
/1
0
.96
5
2
0
.01
2
0
0
.18
1
4
jDE/bes
t/1
0
.96
7
1
0
.01
0
7
0
.17
5
6
k
-
m
eans
0
.82
9
0
0
.09
9
1
0
.63
5
0
4.
CONCL
US
I
O
N
In
t
his
pa
per,
the
pr
ob
le
m
of
SMS
sp
am
detect
ion
w
as
ad
dresse
d
as
a
cl
ust
ering
a
naly
sis.
Fin
di
ng
the clusterin
g
s
tructu
res
that
pr
ovide
high
discrim
inati
on
b
et
ween
SMS s
pa
m
an
d
ham
h
as
b
een c
onside
r
ed
as
NP
-
ha
rd
w
hich
nee
ds
a
m
etah
eu
risti
c
al
gorithm
.
Thr
ee
DE
var
ia
nts
w
ere
util
iz
ed
to
disti
nguish
i
nc
om
ing
m
essages
int
o
SMS
s
pam
and
ham
.
Exp
e
ri
m
ental
resu
lt
s
rev
eal
t
hat
the
DE/
best/1
s
ur
passes
ot
her
va
riants
and
k
-
me
ans
in
te
rm
s
of
acc
ur
acy
,
false
-
ne
gative
rate,
a
nd
false
-
posit
iv
e
rate.
F
uture
work
sho
uld
in
cl
ud
e
a
n
inv
est
igati
on
t
o
c
onsider
m
ul
ti
-
obj
ect
ive e
voluti
onary al
gorithm
s
t
hat
m
a
y pro
du
ce
b
et
t
er r
e
su
lt
s
.
REFERE
NCE
S
[1]
S.
M.
Abdulha
m
id
,
et
al
.,
“
A
Revi
ew
on
Mo
bil
e
SM
S
Spam
Filt
eri
ng
Tech
nique
s,
”
in
I
EEE
Acce
ss
,
vol
.
5,
pp.
15650
-
1566
6,
2017
.
[2]
G.
Rawa
shdeh
,
et
a
l.
,
“
Com
par
at
ive
B
et
wee
n
O
pti
m
iz
ation
Fea
t
ure
Selecti
on
b
y
Us
ing
Cla
ss
ifi
e
rs
Algorit
hm
s
o
n
Spam
Emai
l,
”
In
te
rnational
Jour
nal
of
Elec
tri
cal and
Computer
E
ngine
ering
(
IJE
CE)
,
vol.
9,
n
o.
6,
pp.
5479
-
548
5
,
2019
.
[3]
N.
Desai
and
M.
Narve
kar
,
“
No
r
m
al
iz
a
ti
on
of
Nois
y
Te
xt
Dat
a,
”
Proce
dia
Computer
Sc
ie
nc
e
,
vol
.
45,
pp.
127
-
132
,
2015.
[4]
N.
Jiang
,
e
t
al.,
“
Understa
n
ding
SM
S
Spam
in
La
rge
C
el
lu
la
r
Ne
twork:
Char
ac
t
er
isti
cs,
Strategi
es
and
D
efe
nses,
”
Inte
rnational
W
orkshop on
R
ec
e
nt
Ad
vanc
es
in I
ntrusion De
tection
,
pp
.
328
-
347
,
201
3
.
[5]
H.
Saje
di
,
et
a
l
.
,
“
SM
S
Spam
Filt
eri
ng
Us
ing
Mac
hine
L
ea
rn
i
ng
Te
chni
qu
es:
A
Survey
,
”
M
achi
ne
Learning
Re
search
,
vol
.
1
,
no.
1,
pp.
1
-
14
,
2016
.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Diff
erenti
al ev
olu
ti
on
detect
ion m
odel
s for
SM
S
s
pam
(
Sa
ra
b M.
Ha
mee
d)
601
[6]
T.
M
.
Mahm
oud
and
A.
M
.
M
ahf
ouz,
“
SM
S
Spam
Filt
eri
ng
Te
chn
ique
B
a
se
d
on
Artifici
al
I
m
m
une
Sy
stem
,
”
Inte
rnational
Jo
urnal
of
Comput
er
Scienc
e
Iss
ue
s
,
vol. 9,
no.
2,
p
p.
589
-
597
,
201
2.
[7]
N.
Sjari
f
,
et
a
l
.
,
“
SMS
Spam
Me
ss
age
Detect
ion
Us
ing
Te
rm
Fre
quency
-
Inve
rse
Docum
ent
Frequ
ency
and
R
andom
Forest
Algorit
h
m
,
”
Proc
edi
a
Co
mputer
Scienc
e
,
vol.
161
,
pp
.
509
-
515
,
2019
.
[8]
X.
Qian,
“
SM
S
Spam
Dete
ct
ion
Us
ing
Nonconte
nt
Feat
ure
s
,
”
IE
EE
Intelli
g
ent
S
yste
ms
,
vol.
27
,
n
o.
6,
pp.
44
-
51
,
2012.
[9]
A.
Kara
m
i
and
L.
Zhou
,
“
Im
proving
Stat
ic
SM
S
Spam
Dete
ction
b
y
Us
ing
New
Conte
nt
-
b
ase
d
Feat
ur
es,
”
T
he
20
th
Ame
ricas
Conf
ere
nce on
In
formation
S
yste
ms
(
AMCIS)
,
2014
.
[10]
K.
Za
in
al
and
M
.
Z.
Jal
i,
“
A
Revi
ew
of
Feat
ure
E
xtra
c
ti
on
Optimi
za
t
ion
in
S
MS
S
pam
Mess
age
s
C
la
ss
ifi
c
at
ion
,
”
i
n
Berr
y
M.,
et
al.
(
e
ds)
,
“
Soft
Co
m
puti
ng
in
Dat
a
Scie
nc
e
(
SC
DS
),
”
Comm
unic
a
tions
in
Computer
and
Informatio
n
Sci
en
ce
,
vol
.
65
2
,
pp
.
158
-
170
,
201
6
.
[11]
Y.
Ka
y
a
and
O.
F.
Ert
uğru
l,
“
A
Novel
Feat
ure
Ext
r
ac
t
ion
Approac
h
in
SM
S
Spa
m
filte
ring
for
Mobi
le
Com
m
unic
at
ion: One
-
Dim
ensiona
l Te
rn
ar
y
Pat
te
r
ns,
”
S
e
curity
an
d
communic
at
io
n
net
works
,
201
6.
[12]
N.
F.
Sulai
m
an
and
M.
Z.
Jali
,
“
A
New
SM
S
Sp
am
Dete
ct
ion
M
et
hod
Us
ing
both
Conte
nt
-
Based
and
Non
Conte
nt
-
Based
Fea
ture
s
,
”
Adv
an
ce
d
Computer
and
Comm
unic
ati
on
Eng
in
ee
ring Te
chnol
o
gy
,
pp
.
505
-
514
,
2016
.
[13]
N.
Choudhar
y
and
A.
K.
Jain
,
“
Towa
rds
Filt
e
ring
of
SM
S
Spam
Mess
age
s
Us
ing
M
ac
hine
Learni
ng
B
ase
d
Te
chn
ique
,”
i
n
Singh
D.,
et
al
.
,
(ed
s)
,
“
Advan
ce
d
Inform
at
i
cs
for
Com
puti
ng
Resea
rch
,”
Co
mm
unic
ati
ons
i
n
Computer
and
I
nformation
Sc
ience
,
vol
.
712
,
p
p
.
18
-
30
,
2017
.
[14]
R.
K.
Kaliy
a
r,
et
al
.
,
“
SM
S
Spam
Filt
eri
ng
on
Mu
lt
iple
Bac
kg
roun
d
Data
sets
Us
ing
Mac
hine
L
ea
r
ni
ng
Te
chn
ique
s
:
A
Novel
Approa
ch
,
”
2018
I
EEE
8th
Int
ernati
onal
Adv
an
ce Computi
ng
Conf
ere
nc
e
(
IACC)
,
pp.
59
-
65
,
2018
.
[15]
A.
Shara
ff
,
“
Spam
Dete
c
ti
on
i
n
SM
S
Based
on
Feat
ur
e
Sel
e
ct
ion
Techni
qu
e
s
,
”
i
n
Abraha
m
A.,
et
al.,
(ed
s
)
,
“
Emerging
Te
c
hnologi
es
in
D
at
a
Mining
and
Inform
at
i
on
Secur
i
t
y
,”
Adv
an
ce
s
in
Int
el
li
g
e
nt
Syste
ms
and
Computing
,
vo
l.
813,
pp
.
555
-
56
3,
2019
.
[16]
A.
A.
Ojugo
a
nd
A.
O.
Ebok
a
,
“
Mem
et
ic
Al
gorit
hm
for
Short
Mess
agi
ng
Servic
e
Spam
Filt
er
Us
ing
Text
Norm
al
iz
ation
a
nd
Sem
ant
ic
Approac
h,
”
Inte
rn
ati
onal
Journal
of
Info
rm
ati
cs
and
Comm
unic
ati
on
Technol
og
y
(
IJ
-
ICT
)
,
vol. 9,
no.
1,
pp.
9
-
18
,
2
020
.
[17]
T.
A.
Alm
ei
da
,
et
al.
,
“
Contributions
to
the
Stud
y
of
SM
S
Spam
filte
ring:
New
Coll
e
ct
ion
and
Resul
ts
,”
i
n
Proce
ed
ings o
f
t
he
2011
ACM
S
y
mpos
ium
on
Document
Eng
ineering
(
DO
CENG’1
1
)
,
pp.
259
-
262
,
2011
.
[18]
M.
Georgi
ouda
kis
,
et
a
l.
,
“
On
the
Perform
anc
e
of
Diffe
r
ent
i
al
Evol
u
ti
on
Vari
an
ts
in
Constrai
n
ed
Struct
ur
al
Optimiza
ti
o
n,
”
Proce
dia
Manuf
act
uring
,
vol
.
44
,
pp
.
371
-
378
,
2
020.
[19]
T.
E
lt
a
ei
b
and
A.
Mahm
ood
,
“
Diffe
ren
tial
E
volut
ion
:
A
Surve
y
and
Ana
l
y
sis
,
”
App
li
ed
Sci
en
ce
s
,
vo
l.
7
,
p.
1945
,
2018
.
[20]
M.
Georgi
ouda
kisa
and
V.
Pl
evr
is,
“
On
the
Perform
anc
e
of
Diffe
ren
t
ia
l
E
volut
ion
Vari
an
ts
in
Constrai
n
ed
Struct
ura
l
Optim
iz
a
ti
o
n,
”
Proce
d
ia
Manufa
ct
urin
g
,
vol
.
44
,
pp
.
37
1
-
378
,
2020
.
[21]
S.
Das
and
P.
N.
Sugantha
n,
“
Diffe
ren
t
ia
l
Evo
lut
ion:
A
Surve
y
of
th
e
Sta
te
-
of
-
the
-
Art,
”
IE
EE
Tr
ansacti
ons
on
Ev
olutionar
y
Co
mputati
on
,
vol
.
15,
pp
.
4
-
31
,
20
11.
[22]
J.
Brest,
e
t
al.,
“
Dy
n
amic
optim
iz
at
ion
using
Self
-
Adapti
v
e
Diffe
ren
t
ia
l
Evo
lut
ion,
”
2009
I
E
EE
Congress
on
Ev
olutionar
y
Co
mputati
on
,
Tron
dhei
m
,
pp
.
415
-
422
,
2009
.
[23]
K.
S
.
Redd
y
an
d
E.
S
.
Redd
y
,
“
Inte
gra
te
d
App
roa
ch
to
Det
ec
t
Spam
in
Socia
l
Media
Networ
ks
Us
ing
Hy
b
ri
d
Feat
ure
s,
”
In
te
rn
ati
onal Journ
al of
E
le
c
tric
al
and
Computer
Eng
i
nee
ring (
IJE
C
E)
,
vol
.
9
,
no
.
1
,
pp
.
562
-
569
,
2019
.
[24]
A.
Boukhal
f
a,
e
t
al
.
,
“
LSTM
D
ee
p
Learni
ng
Method
for
Network
Intrusion
Dete
c
ti
on
S
y
ste
m
,
”
Inte
rnation
al
Journal
of
Elec
t
rical
and
Computer
Eng
ine
ering
(
IJE
CE)
,
vol. 10
,
n
o
.
3
,
pp
.
3315
-
3322
,
2020
.
[25]
S.
Sheikhi
,
et
al.
,
“
An
Eff
ec
ti
v
e
Model
for
S
M
S
Spa
m
Dete
ct
i
on
Us
ing
Conte
nt
-
b
ase
d
Feat
ur
e
s
and
Avera
ged
Neura
l
Network
,
”
Inte
rnat
ional
J
ournal
of
Engi
n
ee
ring,
Tr
ansact
ions
B:
Appl
i
cations
,
vol.
33
,
no
.
2,
pp
.
221
-
228
,
2020
.
BIOGR
AP
H
Y
O
F
AU
TH
OR
Sarab
M.
Ha
m
ee
d
re
ce
iv
ed
B
.
Sc.
degr
ee
in
co
m
pute
r
scie
n
ce
f
rom
univ
ersity
o
f
Baghda
d
,
in
1992,
M.
Sc.
fro
m
unive
rsit
y
of
Baghda
d,
in
199
9
and
Ph.
D.
fro
m
unive
rsit
y
of
te
chno
log
y
in
2005.
She
is
cur
ren
tly
assit
ant
profe
ss
or
at
depa
rtment
of
c
om
pute
r
scie
nc
e
,
unive
rsit
y
of
B
aghda
d.
Her
r
ese
arc
h
intere
st
inc
lud
es
computer
and
data
s
ec
uri
t
y
and
soft
computing
in
the
se
cur
ity
fi
el
d
.
Evaluation Warning : The document was created with Spire.PDF for Python.