TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol.12, No.4, April 201
4, pp. 2936 ~ 2
9
4
0
DOI: http://dx.doi.org/10.11591/telkomni
ka.v12i4.4806
2936
Re
cei
v
ed Se
ptem
ber 5, 2013; Re
vi
sed
No
vem
ber 1
0
,
2013; Accep
t
ed No
vem
b
e
r
23, 2013
An Improved RBF Neural Network Method for
Information Security Evaluatio
n
Liu Yinfeng
Xi
’a
n
Internati
o
nal
U
n
ivers
i
t
y
; Shaa
n
x
i Xi’
an 710
07
7
Chi
n
a
Abstract
It is w
e
ll-know
n that infor
m
ation s
e
curity
means
th
e prot
e
c
tion of i
n
for
m
ation, a
nd
ens
urin
g the
avail
a
b
ility, con
f
identi
a
lity a
nd
integr
it
y of info
rmati
on. The p
u
rpos
e of th
is pap
er is to pre
s
ent an i
m
pr
ov
ed
RBF
neura
l
ne
tw
ork method f
o
r infor
m
ati
on
eval
uatio
n.
Ant colony o
p
ti
mi
zation is a
mu
lti
-
age
nt appr
oac
h
for difficu
lt co
mb
in
atoria
l o
p
t
i
mi
z
a
tio
n
prob
l
e
ms, w
h
ic
h
ha
s be
en
ap
pli
e
d to v
a
rio
u
s N
P
har
d pr
ob
le
ms.
Here, a
n
t colo
ny opti
m
i
z
a
t
io
n
alg
o
rith
m is a
ppli
ed to
opti
m
i
z
e
the
para
m
eters of RBF
n
eura
l
netw
o
rk. I
n
this p
aper, w
e
empl
oy “u
naut
hori
z
e
d
access
”, “u
na
uthor
i
z
e
d
acc
e
ss to
a s
ystem r
e
so
urce
”
,
“
data
le
aka
g
e
”
,
“denial
of service”
, “
unauthor
i
z
ed
m
o
dific
a
tion data and
software”
, “syst
em
cras
h”
as
the features
of
infor
m
ati
on sec
u
rity eval
uati
o
n
.
It is indicated
that t
he infor
m
ation s
e
curity e
v
alu
a
tion
error
of the i
m
prov
e
d
RBF
neural n
e
tw
ork is smal
ler than that o
f
the
RBF
neural netw
o
rk. Thus, the impro
v
ed RBF
neur
a
l
netw
o
rk is very suitabl
e for in
formation securi
ty evaluation.
Ke
y
w
ords
:
i
m
prove
d
RBF
ne
ural n
e
tw
ork, informati
on sec
u
rity, evaluati
on
meth
od
Copy
right
©
2014 In
stitu
t
e o
f
Ad
van
ced
En
g
i
n
eerin
g and
Scien
ce. All
rig
h
t
s reser
ve
d
.
1. Introduc
tion
It is well
-kno
wn that info
rmation
se
curity
means th
e protectio
n
of informatio
n, and
ensurin
g the
availability, confidentiality and integ
r
it
y of information
[1-5]. RBF n
eural
network is
a type of fee
d
-forwa
rd net
work, whi
c
h
has three
different layers.
The input layer is used to
colle
ct the i
n
put inform
ation [6, 7]. Th
e ce
nter la
ye
r in
clud
es
ra
dial ba
si
s fu
nction
s, which is
con
n
e
c
ted
directly to all
th
e elem
ents in
the o
u
t
put la
yer. The
outp
u
t of the
neu
ral net
work i
s
a
linear
co
mbin
ation of the radial ba
si
s fu
nction
s.RBF
neural net
work can
a
pproximate co
ntinu
o
u
s
function m
a
p
p
ing
with the
excelle
nt accura
cy. An
t col
ony optimization is
a multi-agent a
pproa
ch
for
difficult combinato
r
ial optimizatio
n probl
em
s,
wh
ich h
a
s be
en
applie
d to
solve vario
u
s
NP
hard p
r
o
b
le
ms [8-1
1]. Here, ant col
ony optim
iza
t
ion algorith
m
is appli
e
d
to optimize
the
para
m
eters o
f
RBF neural netwo
rk.
The pu
rpo
s
e
of this pape
r is to pre
s
e
n
t an improv
ed RBF ne
ural netwo
rk m
e
thod for
informatio
n evaluation. The inform
ation intr
u
s
ion
types inclu
d
ing “u
nauth
o
rized a
c
cess”,
“una
utho
rized
access to a system resou
r
ce
”, “
data le
aka
ge”, “deni
al of
service”,
“unauth
o
ri
ze
d
modificatio
n
data
an
d software”,
“syste
m
cra
s
h”
hav
e a g
r
e
a
t influen
ce o
n
info
rmation
se
cu
rity.
Thus,
we em
ploy “un
auth
o
rized a
c
ce
ss”, “una
uthori
z
ed
acce
ss to a syste
m
reso
urce
”, “da
t
a
leakage
”, “de
n
ial of
se
rvice”, “
una
uthori
z
ed
modifi
cat
i
on d
a
ta an
d
softwa
r
e
”
, “system
crash”
as
the feature
s
of informatio
n se
cu
rity e
v
aluation. It
is indicated
that the information se
cu
rity
evaluation e
r
ror of the im
proved
RBF
neural net
wo
rk i
s
sm
aller
than that of the RBF n
eural
netwo
rk. T
h
u
s
, the imp
r
o
v
ed RBF n
e
u
ral n
e
two
r
k is very suitable for i
n
fo
rmation
se
cu
rity
evaluation.
2. The Des
c
r
i
ption of RBF Neur
al Netw
o
r
k
RBF neu
ral n
e
twork is a type of feed-fo
rward
network, which ha
s three diffe
rent
layers.
The in
put lay
e
r i
s
u
s
e
d
to
coll
ect th
e i
nput info
rmat
ion. The
cent
er laye
r in
clu
des ra
dial
ba
sis
function
s, wh
ich i
s
called
the
hidd
en la
yer. It is co
n
necte
d directl
y
to all the e
l
ements i
n
th
e
output laye
r,
whi
c
h ca
n re
spo
n
s
e
d
e
c
r
e
a
se
s,
o
r
i
n
c
r
eas
es,
mon
o
tonically with
distan
ce
from
a
cente
r
point.
The output of
the
ne
ura
l
net
work i
s
a
line
a
r
combi
nation
of the
ra
d
i
al ba
si
s
function
s.
RB
F neural net
work
can ap
p
r
oximate cont
i
nuou
s fun
c
tion mappi
ng
with the exce
llent
ac
cur
a
cy
.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
An Im
proved
RBF Ne
ural
Network Met
hod for Info
rm
ation Securi
ty Evaluatio
n (Liu Yinfen
g)
2937
The ra
dial ba
sis fun
c
tion of
the hidden n
ode
s is sho
w
n as follo
ws:
2
2
2
exp
i
j
i
j
c
x
c
x
G
(1)
W
h
er
e
j
x
is the
input ve
ctor
of the
j
th
inp
u
t nod
e,
i
c
is th
e cente
r
of th
e
i
th RBF unit,
and
is
the width of RBF unit.
The output of
RBF neu
ral
netwo
rk
is
formed
by
a li
n
early
weig
hte
d
sum
of the
numbe
r
of radial ba
si
s functio
n
s in
the hidde
n la
yer, which ca
n be de
scribe
d as follo
ws:
i
j
n
i
ik
k
c
x
G
w
x
f
1
(2)
Whe
r
e
ik
w
is the weig
ht from the
i
th hidde
n layer to the
k
t
h
output layer.
3. Optim
i
zin
g
the Param
e
ter
s
of R
B
F
Neur
al N
e
t
w
ork by
Ant Colony
Optim
i
zatio
n
Ant colony o
p
timization i
s
a multi-a
gen
t appro
a
ch fo
r difficult com
b
inatori
a
l opti
m
ization
probl
em
s, wh
ich ha
s b
een
applied to v
a
riou
s
NP
ha
rd proble
m
s.
Here, ant col
ony optimization
algorith
m
is a
pplied to
opti
m
ize th
e p
a
rameters
of RBF neu
ral
net
work. T
he
se
arching
proce
ss
of sel
e
ctin
g t
he p
a
ra
meters of
RBF
neu
ral n
e
two
r
k b
y
ant colony
optimizatio
n
algorith
m
can
b
e
descri
bed a
s
follows:
Step 1: Initially, a set of ants are init
iali
zed, the a
n
ts solution
co
n
s
ist
s
of
n
nu
mber of
feature
s
ea
ch
by using an i
n
itialization
ru
le.
Step 2: Each
of the
r
ant
s
con
s
t
r
u
c
t
r
dif
f
erent
solutio
n
s, RBF
neu
ral network ev
aluate
s
each su
bset by determini
n
g
the error in
predi
ction by
usin
g that su
bset of
n
feature
s
.
Step 3: A local u
pdatin
g
rule i
s
ap
plie
d to
the rest
of the ants.
Re
cord the l
o
cal
be
st
sub
s
et of feature.
Step 4: A global upd
atin
g rule is ap
p
lied to the solution set. Re
cord the global be
st
sub
s
et of feature.
4. Testing
and An
aly
s
is for Infor
m
ation Sec
u
rit
y
b
y
Improv
ed RBF
Neural Netw
o
r
k
Metho
d
The info
rmati
on intrusi
on t
y
pes
are
mai
n
ly “u
n
autho
ri
zed
a
c
cess”,
“una
utho
rized
acce
ss
to a system
reso
urce
”, “da
t
a l
eakage
”, “denial of
se
rvice”,
“u
n
auth
o
rized m
odifi
cation
data a
nd
softwa
r
e
”
, “system
cra
s
h
”
,whi
ch
have
a g
r
eat
infl
uen
ce
on i
n
formatio
n
se
curity. Thu
s
,
we
employ “una
uthori
z
ed a
ccess”, “un
auth
o
rized a
c
cess to a syste
m
reso
urce
”, “data leakag
e”,
“deni
al of se
rvice
”
, “un
a
u
t
horized m
o
d
i
ficati
on d
a
ta
and
softwa
r
e”, “system
cra
s
h
”
a
s
th
e
feature
s
of i
n
formatio
n secu
rity evalu
a
tion,
whi
c
h
are d
enote
d
as
“1
~6
” resp
ectively. The
experim
ental
data a
r
e
sh
o
w
n i
n
T
able
1, amon
g
wh
i
c
h th
e trainin
g
data
a
r
e
sh
own
in
Table
2,
and the
testi
ng data
a
r
e
sho
w
n i
n
Ta
ble 3. T
he in
formation
se
curity eval
uat
ion results of
the
experim
ental
data a
r
e
sho
w
n i
n
Fig
u
re
1. In this
exp
e
rime
nt, the i
m
prove
d
RBF ne
ural
net
work
is ap
plied to
informatio
n secu
rity evalu
a
tion,
the
RBF neu
ral n
e
twork
with
six input no
de
s, six
hidde
n no
de
s and
one
out
put no
de i
s
u
s
ed,
and
ant
co
lo
ny optimi
z
ation
algo
rit
h
m is ap
plied
to
optimize the
para
m
eters
o
f
RBF neural netwo
rk.
The informati
on se
cu
rity evaluation val
ues of the i
m
prove
d
RB
F neural net
work a
r
e
given in
Figu
re 2, an
d the
i
n
formatio
n
se
curity
eval
uat
ion value
s
of
the RBF
neu
ral net
work
are
given in Fig
u
re 3. In o
r
der to
sho
w
the su
peri
o
rity of the improve
d
RB
F neu
ral n
e
twork
comp
ared
wi
th the
RBF
neu
ral
net
work, th
e inf
o
rmatio
n
se
curity evaluati
on e
r
ror of
the
improve
d
RBF neu
ral
net
work
and
th
e RBF
ne
ural net
work i
s
give
n. Fig
u
re
4 give
s
the
informatio
n secu
rity evaluation
error of
the improve
d
RBF neu
ral
network, an
d Figure 5 gi
ves
the informatio
n se
curity evaluation
e
r
ror of the RBF neural n
e
two
r
k.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 4, April 2014: 2936 – 2
940
2938
Table 1. The
Experimental
Data
U
1
U
2
U
3
U
4
U
5
U
6
Evaluation value
s
1
0.4 0.3 0.4
0.6 0.5 0.3
0.38
2
0.3 0.5 0.8
0.4 0.2 0.5
0.45
3
0.5 0.6 0.2
0.8 0.7 0.5
0.62
4
0.4 0.3 0.4
0.2 0.3 0.4
0.38
5
0.3 0.2 0.3
0.2 0.2 0.4
0.26
6
0.4 0.3 0.4
0.6 0.5 0.4
0.47
7
0.7 0.6 0.8
0.7 0.8 0.6
0.75
8
0.3 0.2 0.3
0.2 0.4 0.2
0.24
9
0.5 0.4 0.5
0.6 0.4 0.4
0.48
10
0.5 0.5 0.6
0.8 0.4 0.5
0.56
11
0.6 0.7 0.8
0.5 0.7 0.7
0.72
12
0.3 0.2 0.4
0.5 0.3 0.2
0.35
Table 2. The
Traini
ng Data
U
1
U
2
U
3
U
4
U
5
U
6
Evaluation value
s
1 0.4
0.3 0.4 0.6 0.5 0.3
0.38
2 0.3
0.5 0.8 0.4 0.2 0.5
0.45
3 0.5
0.6 0.2 0.8 0.7 0.5
0.62
4 0.4
0.3 0.4 0.2 0.3 0.4
0.38
5 0.3
0.2 0.3 0.2 0.2 0.4
0.26
6 0.4
0.3 0.4 0.6 0.5 0.4
0.47
7 0.7
0.6 0.8 0.7 0.8 0.6
0.75
8 0.3
0.2 0.3 0.2 0.4 0.2
0.24
Table 3. The
Testing
Data
U
1
U
2
U
3
U
4
U
5
U
6
Evaluation value
s
9
0.5 0.4 0.5
0.6 0.4 0.4
0.48
10
0.5 0.5 0.6
0.8 0.4 0.5
0.56
11
0.6 0.7 0.8
0.5 0.7 0.7
0.72
12
0.3 0.2 0.4
0.5 0.3 0.2
0.35
Figure 1.
The Information
Sec
u
rity
Evaluation Result
s of the Experimental
Data
Figure 2.
The
Information
Secu
rity Evaluation
Values of the
Improved
RBF Neu
r
al Network
0
2
4
6
8
10
12
0.
2
0.
3
0.
4
0.
5
0.
6
0.
7
0.
8
0.
9
E
v
al
uat
i
o
n v
a
l
u
es
No
.
9
10
11
12
0.
3
5
0.
4
0.
4
5
0.
5
0.
5
5
0.
6
0.
6
5
0.
7
0.
7
5
E
v
al
uat
i
o
n v
a
l
u
es
No
.
AC
T
U
A
L
IR
B
F
N
N
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
An Im
proved
RBF Ne
ural
Network Met
hod for Info
rm
ation Securi
ty Evaluatio
n (Liu Yinfen
g)
2939
Figure 3.
The
Information
Secu
rity Evaluation
Values of the
RBF Ne
ural
Network
Figure 4.
The
Information
Secu
rity Evaluation
Erro
r of the Improve
d
RB
F Neu
r
al Network
Figure 5.
The
Information
Secu
rity Evaluation Error o
f
the RBF Ne
ural Netwo
r
k
It is in
dicated
that
the
information
se
cu
ri
ty
evaluation error of the improve
d
RB
F neural
netwo
rk is
smaller th
an th
at of the RBF
neu
ral
n
e
two
r
k. T
h
u
s
, the i
m
prove
d
RB
F neu
ral
network
is very suitabl
e for informati
on se
cu
rity evaluation.
5. Conclusio
n
This p
ape
r
pre
s
ent
s an
improved
RBF neu
ral
netwo
rk m
e
thod for i
n
formation
evaluation.
RBF n
eural
netwo
rk
ca
n app
roxi
ma
te co
ntinuou
s fun
c
tion
mappin
g
with the
excelle
nt accura
cy. Ant co
lony optimi
z
a
t
ion is
a
multi
-
age
nt ap
pro
a
ch fo
r difficu
lt combi
natori
a
l
optimizatio
n
probl
em
s, wh
ich ha
s b
een
applied to
v
a
riou
s
NP ha
rd proble
m
s.
Here, ant col
ony
optimizatio
n algorith
m
is
applie
d to optimiz
e
the para
m
eters of
RB
F neural netwo
rk.
The
informatio
n in
trusio
n types inclu
d
ing
“u
nautho
riz
ed
acce
ss”, “u
na
uthori
z
ed
access to a
syst
em
resou
r
ce”, “d
ata lea
k
ag
e”,
“deni
al of servic
e
”
,
“u
na
uthori
z
ed mo
dification dat
a
and softwa
r
e”,
“sy
s
tem
crash” h
a
ve a
g
r
e
a
t influen
ce
o
n
inform
ation
se
cu
rity. Thu
s
, we e
m
ploy
“unauth
o
ri
zed
acce
ss”, “un
authori
z
e
d
a
c
cess to
a
system
re
so
urce”, “d
ata leakage
”,
“d
enial of
serv
ice”,
“una
utho
rized
modificatio
n
data an
d so
ftware
”
, “sys
t
e
m cra
s
h”
as the feature
s
of informatio
n
se
curity eval
uation. It is i
n
dicate
d that t
he info
rmatio
n security ev
aluation
erro
r of the im
pro
v
ed
RBF ne
ural
netwo
rk is
smaller th
an t
hat of t
he
RBF neu
ral n
e
t
work. The
r
ef
ore, the i
m
proved
RBF neu
ral n
e
twork is ve
ry suitable fo
r i
n
formatio
n se
curity evaluati
on.
9
10
11
12
0.
3
5
0.
4
0.
4
5
0.
5
0.
5
5
0.
6
0.
6
5
0.
7
0.
7
5
E
v
al
uat
i
on v
a
lues
No
.
AC
T
U
A
L
RB
F
N
N
9
10
11
12
-0
.
0
8
-0
.
0
6
-0
.
0
4
-0
.
0
2
0
0.
02
0.
04
E
v
aluat
ion er
r
o
r
No
.
9
10
11
12
-0
.
0
8
-0
.
0
6
-0
.
0
4
-0
.
0
2
0
0.
02
0.
04
0.
06
0.
08
0.
1
E
v
alua
t
i
on
e
r
r
o
r
No
.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 4, April 2014: 2936 – 2
940
2940
Referen
ces
[1]
LIU Ju
n
x
iu,
XI
AO Lon
g, LU
L
i
u
x
iu, S
U
Xi
ao
w
e
i. F
a
st Infor
m
ation
Secur
i
ty V
e
rificati
on
Method
an
d its
Appl
icatio
n in
Electric Ener
g
y
Manag
eme
n
t T
e
rminal.
JCIT
. 2012; 7(2): 21
7-22
4.
[2]
Seun
g-Ju
ng S
h
in, J
u
n
g
-H
w
a
Kan
g
, Ka
e-D
a
l K
w
a
ck. A
Stud
y of
a Pr
otocol
Mod
e
l
for Informati
o
n
Securit
y
of the RFID Base I
ndi
vidu
al Ide
n
ti
fication Device.
RN
IS
. 2011; 7: 1
0
-12.
[3]
F
e
ixia
n Su
n,
Gai
w
e
n
Guo.
Res
earch
of
I
mmunit
y
-b
as
ed A
noma
l
y I
n
trusio
n D
e
te
ction
and
Its
Appl
icatio
n for Securit
y
Eval
u
a
tion
of E-g
o
ve
rnment Affair Systems.
JDCT
A
. 2012; 6(2
0
): 429- 43
7.
[4]
Pei
x
i
n
Qu, Z
h
en L
i
u. A D
i
g
i
tal Si
gn
ature A
l
gorit
hm in
Informatio
n
a
nd
Net
w
ork S
e
cur
i
t
y
B
a
sed
O
n
Discrete L
ogar
i
t
hm and Has
h
F
unctions.
JCI
T
. 2012; 7(20):
557- 56
4.
[5]
Abdu
lla
h J
aaf
ar, Azman
Sa
msudi
n. A Su
rve
y
of
Bl
ack-
and-W
h
ite Vis
ual
Cr
ypto
gra
p
h
y
Mo
de
ls.
JDCTA
. 2012;
6(15): 23
7- 24
9.
[6]
Juan Ign
a
cio
Mulero-M
artíne
z
. Anal
y
s
is of the erro
rs in th
e mode
lli
ng of mani
pul
ators
w
i
th Gaussia
n
RBF
neura
l
net
w
o
rks.
Neur
oc
omputi
n
g
. 20
0
9
; 72(7–
9): 196
9-19
78.
[7]
Juan
Ign
a
cio
Mulero-M
artíne
z
. Boun
de
dne
ss of the
nom
inal
co
efficient
s in G
aussi
an
RBF
n
eura
l
net
w
o
rks.
Ne
ur
oco
m
p
u
ting
. 2
007; 71(
1–
3): 197-2
20.
[8]
RF T
a
vares Neto, M Godinho Filh
o. Literat
u
re revi
e
w
r
e
g
a
rdi
ng Ant C
o
l
o
n
y
Optimiz
a
ti
on a
ppl
ie
d to
sched
uli
n
g
pro
b
lems: Gu
ide
l
i
nes for
imp
l
e
m
entatio
n
and
dir
e
ctions
for
future res
earc
h
.
En
gi
ne
e
r
ing
Appl
icatio
ns of Artificial Intel
lig
ence
. 20
13; 26
(1): 150-1
61.
[9]
Shah
la N
e
mati
, Mohamma
d
Ehsan B
a
siri.
T
e
xt-i
nde
pe
nd
ent spe
a
ker v
e
rificati
on us
in
g ant co
lo
n
y
optimiz
ation-
ba
sed sel
e
cted fe
atures.
Expert Systems w
i
th Appl
icatio
ns
. 2
011; 38(
1): 620
-630.
[10]
Marc Re
ima
nn,
Marco
La
uma
nns. Sav
i
n
g
s b
a
sed
a
n
t col
o
n
y
optim
izatio
n f
o
r the
ca
pacita
t
ed mi
nim
u
m
span
nin
g
tree
prob
lem.
Co
mputers & Opera
t
ions Res
earch
. 2006; 33(
6): 1794-
182
2.
[11]
Betul Y
agm
ah
an. Mi
xed-mo
d
e
l
assemb
l
y
lin
e b
a
la
nc
i
n
g
usi
ng
a m
u
lti-o
b
je
ctive a
n
t co
lon
y
optim
izati
o
n
appr
oach.
Exp
e
rt Systems w
i
th Appl
icatio
ns
. 201
1; 38(1
0
): 1245
3-12
46
1.
Evaluation Warning : The document was created with Spire.PDF for Python.