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.
8
, No
.
6
,
Decem
ber
201
8
, p
p.
5449
~
5456
IS
S
N: 20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v8
i
6
.
pp
5449
-
54
56
5449
Journ
al h
om
e
page
:
http:
//
ia
es
core
.c
om/
journa
ls
/i
ndex.
ph
p/IJECE
Test C
ase Op
timiza
tion
an
d Redun
dancy Reducti
on
Using GA
and Neu
ra
l
Networks
Itt
i
Hood
a,
R
.
S. Chhill
ar
Depa
rtment
o
f
C
om
pute
r
Scie
n
ce a
nd
Appl
icati
on
s,
Maha
rishi
Da
y
an
and
Univ
ersi
t
y
,
Ind
ia
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Dec
2
6
, 201
7
Re
vised
Ju
n
2
9
, 201
8
Accepte
d
J
ul
22
, 2
01
8
More
tha
n
50%
of
software
dev
e
lopment
eff
ort
is
spent
in
t
esti
ng
phase
in
a
t
y
p
ical
software
deve
lopment
pr
oje
c
t.
T
est
ca
s
e
design
as
well
a
s
exe
cut
i
o
n
consum
e
a
lot
of
ti
m
e.
Henc
e
,
a
utomate
d
gene
r
a
ti
on
of
te
st
ca
se
s
is
highl
y
req
uire
d
.
Here
a
novel
t
esti
n
g
m
et
hodolog
y
is
be
ing
pre
sent
ed
to
te
st
obj
ec
t
-
orie
nt
ed
softwar
e
base
d
on
UM
L
state
cha
r
t
di
agr
ams
.
In
thi
s
appr
oac
h
,
func
ti
on
m
ini
m
i
za
t
ion
techniqu
e
is
bei
ng
app
li
e
d
and
gene
r
ate
te
st
c
ase
s
aut
om
at
i
ca
l
l
y
fr
om
UM
L
stat
e
cha
rt
dia
gr
ams
.
Software
te
stin
g
fo
rm
s
an
int
egr
al
par
t
of
the
software
d
e
vel
opm
ent
li
f
e
c
y
c
le.
Since
the
obje
c
ti
ve
of
te
sting
is
to
ensu
re
the
conf
orm
ity
of
an
app
li
c
ati
on
to
it
s
spec
ific
at
ion
,
a
te
st
“
ora
cl
e
”
is
n
ee
d
ed
to
determ
ine
whethe
r
a
giv
en
te
st
ca
se
expos
es
a
f
aul
t
or
not.
An
aut
o
m
a
te
d
or
ac
l
e
to
support
the
a
ct
iv
it
ie
s
of
hum
an
te
sters
ca
n
red
uce
the
actu
al
cost
of
th
e
t
esti
ng
proc
ess
a
nd
the
r
el
a
te
d
m
ai
nte
nan
ce
costs.
In
thi
s
p
a
per
,
a
n
ew
con
c
ept
is
bei
ng
pre
sente
d
using
an
UM
L
stat
e
cha
rt
dia
gr
am
an
d
ta
bl
es
for
the
t
est
c
ase
gen
er
a
tion,
ar
ti
fi
cial
n
eu
ral
n
et
work
as
an
opti
m
iz
at
i
on
tool
for
red
uci
ng
the
red
undan
c
y
in
th
e
te
st
ca
s
e
gene
ra
ted
using
the
g
ene
t
ic
a
lgori
thm.
A
neur
al
net
wor
k
is
traine
d
b
y
the
b
ac
k
-
propa
gation
a
lgo
rit
hm
on
a
set
of
te
st
ca
ses
app
li
e
d
to
th
e
origi
n
al
ver
sio
n
of
the
s
y
stem
.
Ke
yw
or
d:
Au
t
om
ation
tes
ti
ng
So
ft
war
e
test
in
g
li
fe cycl
e
Test
drive
n de
velo
pm
ent
Test
optim
iz
at
i
on
Ver
ific
at
io
n
Copyright
©
201
8
Instit
ut
e
o
f Ad
vanc
ed
Engi
n
ee
r
ing
and
S
cienc
e
.
Al
l
rights re
serv
ed
.
Corres
pond
in
g
Aut
h
or
:
Itti
H
oo
da
,
Dep
a
rtm
ent o
f C
om
pu
te
r
Scie
nce a
nd Ap
plica
ti
on
s,
Ma
har
is
hi D
ay
anand
U
niv
e
rsity
, Ro
hta
k,
Pin C
od
e
–
12
4001, Ha
ryana,
Ind
ia
.
Em
a
il
:
i
tt
iho
oda01@
gm
ail.co
m
1.
INTROD
U
CTION
Test
ing
is
def
i
ned
as
a
proc
ess
of
eval
uation
t
hat
ei
ther
the
sp
eci
fic
syst
e
m
m
ee
ts
it
s
or
i
gin
al
ly
sp
eci
fied
re
qu
i
rem
ents
or
not
.
It
is
fun
dam
e
ntall
y
a
pr
oce
dure
i
nclu
ding
appr
ov
al
a
nd
c
onfirm
at
ion
process
that
w
hethe
r
t
he
c
reated
f
ra
m
ewo
r
k
m
eet
s
the
necessit
ie
s
cha
racteri
ze
d
by
cl
ie
nt.
S
ubseq
uen
tl
y,
this
act
ion
br
i
ng
s
a
bout
a
con
trast
am
on
gs
t
real
an
d
exp
ect
e
d
outc
om
e.
Pr
ogram
m
ing
Test
ing
al
lud
es
to
disc
ov
e
rin
g
bugs
,
m
ist
akes
or
m
issi
ng
ne
cessi
ti
es
in
the
create
d
f
ram
e
work
or
pro
gr
a
m
m
ing
.
Along
these
li
nes,
thi
s
is
an
exam
inati
on
th
at
f
urnis
hes
t
he
p
a
rtners
with t
he
c
orrect l
ear
ning a
bout th
e
natu
re
of
t
he
it
e
m
.
So
ft
war
e
Te
sti
ng
ca
n
al
so
be
con
si
der
e
d
as
a
risk
-
base
d
ac
ti
vity
.
The
i
m
po
rta
nt
thin
g
w
hile
te
sti
ng
process
the
pr
oduct
a
naly
zer
s
m
us
t
com
pr
ehend
t
hat
how
to
li
m
it
an
ext
ensive
num
ber
of
te
sts
i
nto
s
ensible
te
sts
set
an
d
s
et
tl
e
on
i
ns
i
gh
tful
c
ho
ic
es
ab
ou
t
t
he
dange
r
s
that
are
im
per
at
ive
to
te
st
or
w
hat
are
not.
[1
]
Figure
1
dem
on
st
rates
the
t
est
ing
e
xpen
se
an
d
blun
der
s
fou
nd
a
relat
ion
s
hi
p.
The
Fi
gure
1
unm
istak
ably
dem
on
strat
es
that
cost
goes
up
drast
ic
al
ly
in
te
sti
ng
the
tw
o
sorts
i.e
.
util
it
arian
an
d
no
nfun
ct
io
nal.
T
he
basic
le
ader
s
hip
for
wh
at
to
te
st
or
dim
inish
te
sts
then
it
can
ca
use
to
m
iss
m
any
bugs
.
The
via
ble
te
sti
ng
ob
je
ct
ive
is t
o do that i
de
al
m
easur
e
of
te
sts wit
h
the
goal that
a
ddit
ion
al
test
ing exe
r
ti
on
ca
n be lim
it
ed.
A
cco
r
ding
to
Figure
1,
S
of
t
war
e
te
sti
ng
is
an
im
po
rtant
com
po
ne
nt
of
so
ft
war
e
qual
it
y
assur
a
nce.
The
im
po
rtanc
e
of
te
sti
ng
ca
n
be
c
on
si
der
e
d
from
li
fe
-
crit
ical
so
ftw
are
(e.
g.
,
flig
ht
co
ntr
ol)
te
sti
ng
w
hi
ch
ca
n
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.
8
, N
o.
6
,
Dece
m
ber
2
01
8
:
5449
-
5456
5450
be hig
hly ex
pe
ns
ive
beca
us
e
of
risk re
gardi
ng s
che
dule
d
e
la
ys, cost
ov
e
rrun
s
, or o
utright
cancell
at
ion
[
2], and
m
or
e about thi
s [3], [
4].
Figure
1. Eve
r
y Sof
t
war
e
P
roject ha
s
Op
ti
m
al
Test Ef
fort
[
1].
The
pri
m
ary
tar
get
of
te
sti
ng
a
so
ftwa
re
m
od
ule
is
to
deci
de
how
well
an
assesse
d
ap
pl
ic
at
ion
fi
ts
i
n
with
it
s
requir
e
m
ents.
Tw
o
norm
al
ways
t
o
deal
with
program
m
ing
te
sti
ng
are
blac
k
-
box
a
nd
w
hite
-
box
te
sti
ng
.
Wh
il
e
the
wh
it
e
-
bo
x
app
r
oac
h
util
iz
es
the
wr
it
te
n
co
de
of
the
pro
gr
am
un
de
r
te
st
to
play
ou
t
it
s
te
sti
ng
,
the
bl
ack
box
a
ppr
oa
ch
c
heck
s
the
p
r
ogram
yi
eld
ag
ai
ns
t
the
con
t
rib
ution
w
it
ho
ut
c
onside
r
ing
it
s
inwa
rd
wor
kings.
S
oft
war
e
te
sti
ng
is
well
e
xp
la
ine
d
int
o
three
phases:
ge
ner
at
io
n
of
te
st
data,
ap
plica
ti
on
of
the
data
to
t
he
softwa
re
bein
g
te
ste
d,
a
nd
e
valuati
on
of
th
e
res
ults.
In
pa
st
,
pro
g
ram
m
i
ng
te
sti
ng
wa
s
done
ph
ysi
cal
ly
by
a
hu
m
an
analy
zer
w
ho
pick
e
d
the
te
st
case
s
and
i
nv
e
sti
ga
te
d
the
outc
om
es
of
the
s
oft
war
e
m
od
ule.
N
ow
days,
beca
us
e
of
the
e
xpansi
on
in
t
he
num
ber
a
nd
siz
e
of
the
pro
j
ect
s
bein
g
trie
d
in
pr
ese
nt
days,
the
bur
de
n
on
t
he
hu
m
an
analy
zer
is
m
or
e,
a
nd
a
lt
ern
at
ively
,
som
e
autom
atic
pro
gr
am
m
ing
t
est
ing
strat
egies
are
r
equ
i
red.
Wh
il
e
autom
at
ic
strat
egies
seem
to
con
si
der
the
c
on
t
ro
l
over
t
he
par
t
of
the
hu
m
an
analy
zer,
the
i
ssu
es
of
reli
ab
il
ity,
qu
al
it
y
a
nd
the
ca
pacit
y
of
the
pro
du
ct
te
sti
ng
te
chn
iq
ues
sti
ll
sh
ou
l
d
be
ver
ifie
d.
Alon
g t
hese lines
, tes
ti
ng
is a
criti
cal
v
ie
w
po
i
nt in
the
ou
tl
ine
of a
so
ft
war
e
prod
uc
t [5
]
.
Au
t
om
at
ed
te
st
case
ge
ne
rati
on
is
process
of
ge
ner
at
i
ng
the
te
st
data
on
the
basis
of
usa
ge
of
th
e
app
li
cat
io
n
an
d
real
ti
m
e
sce
nar
i
o.
Wh
e
n
t
he
pr
ocess
is
a
uto
m
at
ed
then
it
is
qu
it
e
po
s
s
ible
that
the
pr
ocess
m
igh
t
produce
the
case,
w
hich
are
sam
e
in
so
m
e
m
ann
er
f
or
wh
ic
h
the
outp
ut
is
sa
m
e
fo
r
al
l
instance
s,
th
e
si
m
il
ar
ty
pes
of
te
st
cases
wil
l
no
t
af
fect
the
eff
i
ci
e
ncy
of
t
he
s
of
t
war
e
m
odule
but
will
aff
ect
t
he
e
ff
ic
ie
ncy
of
the
te
sti
ng
proces
s
in
te
rm
s
of
the
r
unni
ng
tim
e.
The
two
functi
on
s
of
r
edun
dan
cy
are
passive
redu
ndancy
and
act
ive
redunda
ncy
.
B
oth
f
un
ct
io
ns
pre
ven
t
pe
rfor
m
ance
decli
ne
from
exceedin
g
sp
eci
ficat
io
n
lim
it
s
without
hu
m
an
inter
ven
ti
on
usi
ng ex
tra
cap
a
ci
ty
.
Passive
re
dund
ancy
us
es
exce
ss
capaci
ty
to
red
uce
the
im
pact
of
com
po
ne
nt
fail
ur
es
.
O
ne
com
m
on
form
of
pa
ssiv
e
redunda
ncy
is
the
extra
str
eng
t
h
of
cabli
ng
a
nd
str
uts
us
e
d
in
br
i
dg
e
s.
This
extra
s
tren
gth
al
lows
s
om
e
st
ru
ct
ur
al
com
po
ne
nts
to
fail
without
bri
dge
colla
ps
e.
T
he
extra
stre
ngth
us
e
d
in
the
de
sign
is
cal
le
d
the m
arg
in
of sa
fety
.
Acti
ve
re
dund
ancy
el
i
m
inates
perform
ance
decli
nes
by
m
on
it
or
i
ng
t
he
pe
rfor
m
anc
e
of
in
div
i
dual
dev
ic
es
,
an
d
th
is
m
on
it
or
in
g
i
s
us
e
d
in v
otin
g
lo
gic.
T
he
vo
ti
ng
lo
gic
is
li
nk
ed
t
o
switc
hing
that
aut
om
at
i
cal
ly
reconfi
gures
t
he
com
ponen
t
s.
Er
ror
detect
ion
a
nd
c
orrec
ti
on
an
d
t
he
G
l
ob
al
P
os
it
ion
i
ng
Syst
em
(G
PS)
a
re
two
e
xam
ples o
f
acti
ve
r
ed
un
dan
cy
.
Ther
e
a
re
so
m
e
of
the
chall
e
ng
e
s
relat
ed
to
the
con
ce
pt,
te
st
case
gen
era
ti
on
is
hav
i
ng
so
m
e
of
the
def
i
ned chall
en
ges
s
om
e o
f
t
hem
w
hich
are
c
on
si
der
e
d
i
n
th
e wor
k
a
re:
C1:
Def
i
ning
the
com
plete
req
ui
rem
ents
clear
ly
and
c
omplet
e,
C2:
la
ck
of
a
bili
ty
to
i
den
ti
fy
t
he
crit
ic
al
do
m
ain
re
quirem
ents,
C3:
F
un
ct
i
on
al
requirem
ents
de
finiti
on
gap,
f
or
t
he
eff
ic
ie
nt
te
st
case
gen
e
rati
on
bette
r
el
ab
or
at
io
n
of
t
he
require
m
ent
is
bein
g
require
d.
I
n
th
e
pro
posed
w
ork
UML
diag
r
a
m
is
bein
g
us
e
d
f
or
def
ini
ng
t
he
f
un
ct
io
nal
an
d
oth
e
r
relat
ed
r
equ
i
rem
ent
of
the
syst
e
m
m
o
du
le
under
te
st
.
C4:
Re
dundancy
in
te
st
case
ge
ne
rati
on
proces
s,
in
the
case
w
he
n
the
c
om
plete
functi
onal
an
d
ot
her
relat
ed
par
ts
of
the
syst
em
is
being
el
ab
or
at
e
d
for
bet
te
r
cov
e
rag
e
t
hen
the
gen
e
r
at
ion
process
so
m
et
i
m
e
gen
erates
duplica
te
te
st
case
for
w
hich
in
the
pro
po
s
ed
w
ork
A
N
N
is
being
us
ed
fo
r
reducin
g
t
he
re
dundancy
in
the
gen
e
rati
on
of
t
he
te
st case
s
.
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
Test C
as
e
Opti
miza
ti
on
and R
edun
dancy
Red
uction U
sin
g G
A an
d
Ne
ural
Ne
tw
or
ks
(
Itti
H
oo
da
)
5451
2.
RELATE
D
W
ORK
UML
act
ivit
y
di
ag
ram
-
based
te
st
case
gen
er
at
ion
has
bee
n
inv
est
igate
d
in [
6]
by
Liz
ha
ng
et
al
.
They
hav
e
pro
duced
te
st
cases
util
iz
ing
a
grey
bo
x
strat
e
gy.
I
n
t
heir
a
ppr
oach,
te
st
sit
uations
are
strai
ghtf
orwardly
go
tt
en
from
th
e
act
ivit
y
char
ts
dem
on
strat
ing
a
n
operati
on
.
T
his
te
chn
i
qu
e
m
anag
es
the
log
ic
al
co
ve
rage
crit
eria
of
w
hit
e
box
strat
egy
and
disc
overs
al
l
the
co
nceiv
able
ways
fro
m
the
desig
n
m
od
el
wh
ic
h
e
xp
la
in
s
the
norm
al
con
duct
of
an
operati
on.
Alon
g
these
li
nes,
al
l
the
data
for
ge
ner
at
io
n
of
the
te
st
case
(i.
e.
input/
ou
t
pu
t
gro
upin
g
pa
ra
m
et
ers,
the
con
st
raints
co
ndit
ion
s
a
nd
e
xp
ect
e
d
ob
j
ec
t
m
et
ho
d
stra
te
gy)
is
extricat
ed
fro
m
each
te
st
situ
at
ions.
At
lo
ng
la
st,
t
hey
pr
oduce
the
possi
ble
val
ues
of
al
l
the
in
form
at
i
on
/y
ie
ld
par
am
et
ers
by
ap
plyi
ng
cat
e
gory
par
ti
t
ion
strat
egy.
It
c
reates
te
st
cas
es
w
hich
ca
n
accom
plish
th
e
pat
h
cov
e
ra
ge.
In
a
ny
case,
t
his
te
chn
i
qu
e
dis
regards
data
a
bout
the
c
onditi
on
of
t
he
it
em
s
inside
the
fr
am
ework
at
the tim
e o
f
e
xe
cution.
The
ap
proac
h
i
ntr
oduce
d
by
Andrews
et
al
.
[7
]
rec
ognized
thin
strin
gs
f
r
om
top
-
le
vel
U
ML
act
ivity
diag
ram
.
A
thin
threa
d
is
a
ba
se
us
e
sit
uation
in
a
s
of
t
wa
re
fr
am
ework.
It
sign
ifie
s
a
total
sit
uation
f
ro
m
the
end
cl
ie
nt'
s
per
sp
ect
ive
.
T
hat
is,
the
f
ram
e
work
ta
kes
i
nput
in
f
or
m
at
ion
,
play
s
out
a
few
cal
c
ulati
ons
,
a
nd
yi
el
ds
the
ou
t
com
e.
They
pr
op
os
ed
a
nove
l
m
et
ho
do
l
ogy
to
produce
thin
th
reads
from
activity
di
agr
am
,
wh
ic
h
incl
ud
e
d
pr
e
processi
ng
of
t
he
syst
em
le
vel
act
ivity
char
ts,
c
ha
ngin
g
ov
e
r
the
m
into
act
ion
hype
r
diag
ram
s
and
after
that
getti
ng
al
l
exec
utio
n
pat
hs
f
ro
m
t
he
dia
gr
am
.
Their
te
ch
nique
do
e
s
not
co
ntain
an
y
sta
te
data
f
or
the
obj
ect
s
of
t
he
fr
am
ewo
r
k.
Che
n
Mi
ngs
ong
et
al
.
[8
]
di
sp
la
ye
d
a
pla
n
to
get
the
dec
reased
te
st
su
it
e
f
or
a
us
a
ge
util
iz
ing
act
ivit
y
grap
hs
.
They
c
onsi
der
e
d
t
he
ar
bit
rar
y
e
ra
of
e
xperim
ents
for
Java
pro
gr
am
s.
Runnin
g
the
pro
j
ec
ts
with
ap
plyi
ng
the
e
xperim
e
nts,
they
go
t
t
he
traces
of
the
pro
gr
am
execut
ion
.
At
lo
ng
la
st,
a
dim
inished
te
st
su
it
e
is
gott
en
by
c
on
t
rasti
ng
the
ba
sic
ways
a
nd
pro
gr
am
exe
cution
fo
ll
ows.
Sim
ple
pat
h
c
ov
e
r
a
ge
c
rite
rio
n
he
lps
to
av
oid
the
pa
th
e
xplo
sion
due
t
o
t
he
presen
ce
of
loops
and co
nc
urren
c
y.
Kan
s
om
keat
and
Ri
vep
i
boon
[9
]
have
pr
opos
e
d
a
te
ch
nique
f
or
creati
ng
te
st
cases
util
iz
ing
UM
L
sta
te
char
ts
diagr
am
.
They
change
the
sta
te
char
t
gr
a
ph
into
a
sm
oo
thed
var
i
ous
le
veled
struct
ur
e
of
sta
te
s
cal
le
d
te
sti
ng
f
low
dia
gr
am
(TFG).
T
he
TF
G
is
then
cr
os
s
ed
from
the
root
node
to
t
he
le
af
node
t
o
cre
at
e
te
st
cases.
F
ro
m
the
TFG,
they
li
s
t
con
cei
vab
le
e
ven
t
se
quence
wh
ic
h
they
c
onside
r
as
te
st
s
equ
e
nce.
Th
e
te
sti
ng
basis they
us
ed
to
direct t
he
er
a o
f
te
st se
qu
e
nce is the s
c
op
e o
f
t
he
sta
te
s
and
c
ha
nges of
a TFG. T
his strategy
m
anag
es
a
s
pe
ci
fic
sta
te
char
t
diag
ram
.
Howe
ver,
in
a
n
exec
utio
n
of
a
us
e
case,
m
or
e
than
one
obj
ect
regularly
takes
an
i
nterest.
Such c
ond
uct wo
uld
be ha
rd to t
est
u
ti
li
zi
ng
thi
s approac
h.
Kim
e
t
al
.
[1
0]
propose
d
a
te
chn
i
qu
e
t
o
pro
duce
te
st
cases
f
or
cl
ass
le
vel
t
est
ing
util
iz
ing
UML
sta
te
gr
a
ph
cha
rts.
T
hey
cha
nged
sta
te
diagr
am
s
to
e
xten
ded
fini
te
sta
te
m
achine
(EFS
Ms
)
to
infe
r
te
st
cases
.
Th
e
var
i
ou
s
le
vele
d
an
d
sim
ultan
eo
us
str
uctu
r
e
of
sta
te
s
is
sm
oo
thed
an
d
com
m
un
ic
at
e
d
co
rr
es
ponde
nces
are
disp
e
ns
e
d
with
in
the
s
ubseq
uen
t
E
FSMs.
The
n,
data
flo
w
is
ide
ntifie
d
by
trans
f
or
m
i
ng
EFSMs
i
nto
f
l
ow
gr
a
phs
to
w
hi
ch
co
nventio
na
l
data
flow
a
naly
si
s
te
chn
iq
ues
are
a
pp
li
e
d.
H
artm
ann
et
al
.
[1
1]
enla
r
ge
the
UML
dep
ic
ti
on
with
par
ti
c
ul
ar
doc
um
entat
ion
s
t
o
m
ake
a
desig
n
-
base
d
te
sti
ng
c
ondi
ti
on
.
T
he
en
gi
neer
s
init
ia
ll
y
char
a
ct
erize
the
dynam
ic
con
du
ct
of
eve
ry
fr
a
m
ewo
r
k
pa
rt
util
iz
ing
a
sta
te
char
t
dia
gr
a
m
.
The
associat
ions
be
tween
m
od
ules
are
then
dete
r
m
ined
by
exp
l
ai
nin
g
th
e
sta
te
char
t
grap
hs
,
and
the
s
ub
se
qu
e
nt
global FSM
th
at
r
el
at
es to t
he
incor
porated
f
ram
ewo
rk co
nduct is
u
ti
li
zed
to cr
eat
e t
he
te
sts.
Gn
esi
et
al
.
[12]
gav
e
a
m
at
he
m
at
ic
al
way
t
o
de
al
with
c
onf
or
m
ance
te
sti
ng
a
nd
a
uto
m
at
ed
te
st
case
gen
e
rati
on
f
or
UML
sta
te
di
agr
am
s.
They
pro
po
se
d
a
f
or
m
al
con
f
orm
ance
te
sti
ng
connecti
on
f
or
in
pu
t
-
e
m
po
we
re
d
tr
ansiti
on
fram
ewor
ks
with
ad
van
ces
nam
ed
by
in
put/
ou
t
pu
t
s
et
s
(IOLT
Ss).
Test
ing
pro
gr
am
m
ing
so
as
to
set
up
the
sat
isfact
io
n
of
the
pr
e
de
te
rm
ined
pr
e
re
qu
isi
te
s
is
known
as
c
onf
orm
ance
te
sti
ng
.
A
co
nfor
m
ance
co
nn
ect
io
n
cha
r
act
erizes
the
eff
ect
ive
ness
of
the
exe
cuti
on
as
f
or
the
fo
rm
al
determ
inati
on
. I
OL
TSs g
ive
a
reasona
ble
sem
antic
m
od
el
t
o
a
co
ndu
ct
s
poke
t
o
by
a
s
ubset
of
sta
te
dia
gr
am
s.
They a
dd
it
io
na
ll
y give an al
gorithm
w
hic
h p
rod
uces a
te
st s
uite f
or
a
g
i
ve
n st
at
e char
t
m
od
el
.
Ali
et
al
.
[13]
hav
e
pro
posed
an
ap
proac
h
f
or
sta
te
-
base
d
in
te
gr
at
io
n
te
sti
ng.
T
heir
work
const
ru
ct
s
a
transiti
on
al
t
es
t
m
od
el
cal
le
d
SCOTEM
(
Stat
e
Coll
abo
rati
on
Test
Mo
del)
from
UML
colla
borati
on
diag
ram
and
the
c
orres
pondin
g
sta
te
c
ha
rts.
SC
OTEM
m
od
el
s
ever
y con
cei
vab
le
way
for
ob
j
ect
st
at
e
cov
e
ra
ge
cr
it
eria
that
a
m
essage
patte
rn
m
ay
t
rig
ger
.
SCOT
EM
at
that
po
int
crea
te
s
te
st
ways
in
vie
w
of
diff
e
re
nt
co
ver
a
ge
crit
eria.
T
heir
pro
du
ce
d
te
st
c
ases
inten
d
t
o
r
eveal
sta
te
de
pe
nd
e
nt
inte
racti
on
f
a
ults.
T
heir
w
ork
co
ns
ide
rs
t
he
sco
pe
of
e
ver
y
sing
le
c
on
cei
va
ble
co
ndit
ion
of
c
ollab
or
at
io
n
am
on
g
cl
asse
s
in
a
com
m
unic
at
ion
.
Bri
a
nd
et
al
.
[14]
hav
e
c
on
s
idere
d
com
m
u
nicat
ion
s
am
ong
obj
ect
s
in
th
ei
r
work,
howe
ver
their
at
te
nt
ion
is
again
on
cl
ass
-
le
vel
te
sti
ng
.
Their
w
ork
de
li
ver
s
an
e
xp
e
rim
ent
detail
c
om
pr
isi
ng
of
a
feasible
seq
ue
nce
of
transit
ion
.
In
their
wor
k,
to
cat
ch
the
conne
ct
ions
am
on
g
sta
te
dep
en
de
nt
obj
ect
s,
an
inv
ocati
on
se
qu
e
nce
tree
is
bu
il
t
wh
ic
h
is t
hen use
d
t
o deriv
e
te
st co
ns
trai
nts
f
or the tra
ns
it
io
n
se
quences
to be test
ed
.
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.
8
, N
o.
6
,
Dece
m
ber
2
01
8
:
5449
-
5456
5452
3.
PROP
OSE
D MET
HO
DOL
OGY
Pr
og
ram
m
ing
te
sti
ng
play
s
a crit
ic
al
par
t
in
so
ft
war
e d
evel
op
m
ent
since
it
can
lim
i
t
the
dev
el
opm
ent
cost. Pr
ogram
m
ing
test
ing
a
ppr
oach
es
are parti
ti
on
e
d
i
nto t
hr
ee s
ect
ion
s
i.e. code
-
base
d t
est
ing
, speci
fi
cat
ion
-
base
d
te
sti
ng
and
m
od
el
-
bas
ed
te
sti
ng.
I
n
m
od
el
-
based
te
sti
ng
,
t
he
te
sti
ng
be
gin
s
at
desig
n
ph
a
se.
Th
us
ly
,
early
r
evelat
io
n
of lac
ks
can be r
efi
ned
by usi
ng
this app
roach m
or
eov
e
r
r
edu
ci
ng
ti
m
e,
cost and endea
vors
of
the
pro
gr
am
m
er
to
an
exte
nsi
ve
de
gr
ee.
In
this
pap
er
a
proce
dure
is
propose
d
f
or
m
a
king
te
st
cases
us
in
g
UML
diag
ram
,
g
e
netic
alg
or
it
hm
an
d
the
AN
N
(
A
rtific
ia
l N
eur
al
Netw
ork
)
.
In
t
he
init
ia
l
s
ta
ge
a
ric
h
a
nd
fi
nish
UML
gr
a
ph
is
im
po
rte
d
w
hich
w
il
l
be
parsed
t
o
e
xtricat
e
fun
dam
ental
m
et
a
-
data
a
nd
s
ha
pe
data
to
po
pu
la
te
a
sta
te
char
t
ta
b
le
w
hic
h
will
com
pr
ise
of
s
ubtl
e
el
em
ents
of
t
he
in
div
i
dual
gr
a
ph.
T
his
will
add
it
io
nally
help
us
to
de
fine
the
te
st
ca
ses
util
iz
ing
th
e
appr
oach
exa
m
ined
unde
r
the
h
ea
di
ng
Test
Case
Gen
e
rato
r [15]
.
In
t
he
sec
ond
sta
ge
we
c
ha
nge
ove
r
a
m
odel
gr
a
ph
by
e
xt
r
ic
at
ing
the
da
ta
by
parsi
ng
m
et
ho
d
a
nd
popula
ti
ng
this
inform
at
ion
in
a
decisi
on
ta
bl
e.
Test
cases
are
obta
ine
d
us
i
ng
al
l
com
bin
a
ti
on
s
usi
ng
Ge
netic
Algorithm
an
d
furthe
r op
ti
m
ized
us
in
g Ar
ti
fi
ci
al
N
eu
ral Net
work.
In
t
he
first
sta
ge
we
im
po
rt
a
m
od
el
char
t
of
t
he
m
od
ule
of
softwa
re
under
te
st.
A
gain
,
we
will
be
con
ce
ntrati
ng
just
on
the
act
iv
it
y
and
sta
te
ch
art
gra
ph
s
m
ade
acco
rd
i
ng
to
the
UML
2.0
pri
nciples.
We
a
ccept
the ch
a
rt to be
finish
e
d
a
nd r
i
ch
ye
t t
her
e is
a chance f
or
t
he
cli
ent to i
m
po
rt a
n
UML
gr
aph
w
hich
m
ay
no
t
be
in si
m
il
arity w
it
h
UML 2.0 be
nch
m
ark
s. In
t
he
pa
per
[16] V.
Mary S
um
a
la
tha an
d Dr. G.
S.
V.
P
. Raju tal
ke
d
about
a
sim
ple
appro
ac
h
to
e
xp
el
am
big
uiti
es
from
activity
char
ts.
T
his
can
be
c
onnec
te
d
to
infl
uen
c
e
th
e
gr
a
ph to
free f
r
om
i
m
p
erf
ect
ion
s.
The
gr
a
ph
im
p
or
te
d
will
be
pa
rsed
t
o
ob
ta
in
m
et
a
-
data.
I
n
the
eve
nt
of
a
n
act
ivit
y
gr
ap
h
we
rec
ove
r
the
swim
-
la
ne
act
or
s
,
swim
-
la
ne
s
pecific
act
ion
s,
deci
sion
node
,
f
ork
node
,
join
node,
co
nne
ct
or
s
,
dep
e
ndencies
a
nd
al
so
it
s
rela
ti
on
s
hip
with
t
he
pr
e
vious
an
d
nex
t
sh
a
pe
.
Transf
or
m
at
ion
of
any
grap
h
to
te
st
cases
inclu
des
par
si
ng
as
it
s
f
irst
phase.
Alt
erati
on
of
sta
te
char
t
gr
a
ph
t
o
sta
te
char
t
ta
ble
is
strai
ghtf
orward.
As
for
our
first
ap
proac
h,
w
e fou
nd
t
hat
c
ha
ng
i
ng
ove
r
a
ny
gr
a
ph
to
a
sta
t
echa
rt g
ra
ph wou
l
d
prof
it
u
s
i
ns
te
ad
of fram
ing
a d
i
ff
e
ren
t
proce
dure fo
r
eac
h kin
d of g
raphs
.
Pr
ese
ntly
con
c
ern
i
ng
t
he
sta
te
char
t
gra
ph,
we
extricat
e
data,
f
or
e
xam
ple,
the
sta
te
s,
com
po
sit
e
sta
te
s,
subm
ac
hin
e
sta
te
s,
dec
isi
on
n
odes
,
co
nn
ect
or
s
,
c
ondi
ti
on
s
a
nd
f
ur
t
he
rm
or
e
it
s
asso
ci
at
ion
with
th
e
past
and
ne
xt
s
ha
pe
.
A
sta
te
cha
rt
ta
ble
is
a
n
op
ti
on
al
m
et
hod
f
or
desc
ribi
ng
s
equ
e
ntial
m
od
al
log
ic
.
Ra
the
r
tha
n
dr
a
wing
sta
te
s
and
tra
ns
it
ion
s
grap
hical
ly
in
a
sta
te
char
t
gr
a
ph,
the
m
od
al
log
ic
is
com
m
un
ic
at
ed
i
n
a
ta
bu
la
r
orga
nizat
ion
.
We
will
be
c
hangin
g
over
sta
te
char
t
gra
ph
int
o
sta
te
ch
art
ta
ble
since
it
is
a
br
is
k,
fr
es
h
config
ur
at
io
n
f
or
a
sta
te
char
t
gr
a
ph.
They
a
dd
it
io
nally
le
s
sen
the
upke
ep
of
grap
hical
it
e
m
s.
No
t
at
all
li
ke
sta
te
char
t
grap
hs
,
will
exp
a
nsi
o
n
a
nd
e
ras
ur
e
of
sta
te
s
into
a
sta
te
cha
rt
ta
ble
discar
d
the
ove
r
-
he
ad
of
m
od
ify
ing
stat
es, a
dv
a
nces a
nd intersect
i
ons.
Af
te
r
the
act
ivi
ty
gr
ap
h
is
parsed
an
d
sig
nifi
cant
sh
a
pe
dat
a
is
extricat
ed
our
fram
ewo
rk
will
isolate
act
ivit
ie
s
in
l
igh
t
of
on
-
scree
n
c
har
act
ers
in
t
wo
cl
asses,
cl
i
ent
act
ivit
ie
s
a
nd
f
ram
ewo
r
k
pro
du
ce
d
act
iv
it
ie
s.
Cl
ie
nt
act
ivit
ies
in
the
s
wim
-
l
ane
dia
gr
am
s
excep
ti
onal
ly
w
el
l
po
rtray
t
he
inf
o
co
ndit
ion
s
to
be
inc
orp
orat
ed
into
the
decisi
on
ta
ble.
A
de
ci
sion
hub
li
ke
wise
de
picts
the
co
ndit
ion
s
on
w
hich
t
he
app
li
cat
io
n
un
der
te
s
t
will
d
epe
nd. I
n o
ur ap
proac
h cl
ie
nt acti
viti
es an
d decisi
on
hubs wil
l act
li
ke
cond
it
io
ns
t
o t
he
decisi
on table.
The
fr
am
ewo
r
k
create
d
act
iv
it
ie
s
rep
rese
nts
outp
ut
act
ivit
ie
s
to
be
c
om
pl
et
ed
in
t
he
dec
isi
on
ta
ble.
Condit
ion
opti
on
s
or
ge
ne
rall
y
cal
le
d
com
bi
nations
are
c
re
at
ed
util
iz
ing
Ca
rtesi
an
pro
duct
.
T
his
is
li
kew
ise
cal
le
d
as
ex
ha
us
ti
ve
te
sti
ng.
These
c
onditi
on
opti
ons
will
be
in
T
r
ue/fals
e
sh
a
pe.
T
o
ge
t
the
est
i
m
at
io
n
of
the
norm
al
ou
tp
ut
we
fo
ll
ow
the
UML
Dia
gram
con
side
rin
g
it
as
a
tree
with
the
beg
i
nn
i
ng
node
goin
g
ab
ou
t
as
the
r
oo
t.
Be
gi
nnin
g
with
the
r
oo
t
node
we
go
th
rou
gh
the
t
ree
to
disc
ov
e
r
the
out
pu
t
act
ivit
y
whose
norm
a
l
resu
lt
is
to
be
c
om
pu
te
d.
Ba
se
d
on
t
he
c
om
bin
at
ion
col
um
n
values
we
deci
de
the
d
irect
i
on o
f
t
he
tra
ver
s
al
.
At
wh
at
e
ver
po
i
nt
there
is
an
act
ivit
y
and
a
tru
e
ou
t
pu
t
is
gott
en
from
the
gr
ou
ps
f
or
the
i
nd
i
vidual
act
ivit
y
we
m
ov
e
fo
r
ward.
In
the
eve
nt
that
a
false
ou
t
pu
t
is
gott
en
we
stop
a
nd
r
est
or
e
a
false
value
to
the
norm
al
ou
tc
om
e
se
g
m
ent.
At
w
hatev
er
point
t
her
e
i
s
a
decisi
on
hub
i
n
the
way
to
the
ou
t
pu
t
a
ct
ivit
y
the
choi
ce
of
wh
ic
h
child
to
m
ov
e
reli
es
up
on
the
m
ix
values
for
that
cho
ic
e.
At
wh
at
ever
poi
nt
ther
e
is
a
fo
rk
hu
b
in
the
way
to
t
he
outp
ut
act
ivit
y
al
l
the
offs
pr
i
ng
of
th
e
f
ork
hub
a
re
na
vig
at
ed
one
by
on
e
to
sea
r
ch
f
or
the outp
ut stat
e
.
The
sta
te
ch
art
ta
ble
as
of
no
w
c
on
ta
in
s
al
l
the
data
im
po
rt
ant
to
m
ake
a
t
est
case.
Wh
il
e
m
aking
te
st
cases
from
s
ta
t
echart
grap
h
c
are
ought
to
be
ta
ken
that
eve
ry
on
e
of
the
tr
ansiti
on
s
a
re
pract
ic
ed
in
any
even
t
on
ce
. T
his strat
egy f
or test
ing
gu
a
ra
ntees ide
al
scope
without creat
in
g
e
xp
ansive
num
ber
te
sts [17].
Til
l
the
tim
e,
ou
tp
ut
of
the
m
e
thodo
l
og
y
is
th
e
set
of
ce
rtai
n
te
st
cases
and
h
ence so
as
to p
r
ovide
th
e
m
axi
m
u
m
co
de
covera
ge,
di
f
fer
e
nt co
m
bin
a
ti
on
s
of the test
cases a
re
gen
e
rated
us
in
g
t
he Geneti
c al
gorithm
.
Gen
et
ic
al
gorithm
s u
se the
f
ollow
in
g
t
hr
ee
operati
ons
on it
s
populat
ion.
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
Test C
as
e
Opti
miza
ti
on
and R
edun
dancy
Red
uction U
sin
g G
A an
d
Ne
ural
Ne
tw
or
ks
(
Itti
H
oo
da
)
5453
a.
Sele
ct
ion
:
A
sel
ect
ion
process
is
ap
plied
to
dete
rm
ine
a
way
in
wh
i
ch
in
div
id
uals
are
ch
os
en
for
m
a
ti
ng
from
a
popula
ti
on
bas
ed on t
heir
f
it
ne
ss.
b.
Cros
s
ove
r:
Af
te
r
the
sel
e
ct
ion
process
,
the
cro
ss
over
op
e
rati
on
is
app
li
ed
to
the
chrom
os
om
es
sel
ect
ed
from
the
popula
ti
on
.
c.
Muta
ti
on
:
Af
te
r
the
c
rosso
ver
proces
s,
the
m
utate
operati
on
is
ap
plied
to
a
rand
om
l
y
sel
ect
su
bse
t
of
the
popula
ti
on
[
18]
.
Af
te
r
ap
plyi
ng
the
Ge
netic
Algo
rithm
there
a
re
different
te
st
cas
es
obta
ine
d
by
diff
e
re
nt
com
bin
at
ion
s
of
e
xisti
ng
te
st
cases
that
m
e
ans
the
re
dund
ancy
of
the
te
s
t
cases
increas
es
so
optim
iz
ation
of
the
te
st
cases
i
s
bein
g
do
ne
us
in
g
the
A
rtific
ia
l
Neu
ral
Netw
ork.
A
N
N
is
bei
ng
use
d
li
ke
if
t
he
new
ly
gen
e
rated
te
st
case
is
m
a
tc
hin
g
with
t
he
al
r
eady
existi
ng
t
est
case
in
the
li
st.
If
ye
s,
t
he
n
it
will
just
re
m
ov
e
the g
e
ne
rated t
est
case an
d
el
s
e the
gen
e
rated
test
case
is a
dded
to
t
he
te
st
pool.
Ar
ti
fici
al
ne
ural
netw
orks
(
ANNs
)
are
m
ade
ou
t
of
strai
ghtfo
r
ward
com
po
ne
nts
w
hich
w
ork
si
m
ultaneou
sly
.
A
N
N
syst
em
can
be
pr
e
pa
red
to
play
ou
t
a
sp
eci
fic
pa
rt
by
changin
g
the
weig
hts
betwee
n
com
po
ne
nts.
S
yst
e
m
wo
r
k
is
con
t
ro
ll
e
d
by
the
ass
ociat
io
ns
bet
ween
com
po
ne
nts.
T
her
e
is
act
ivati
on
f
unct
io
n
us
e
d
to
create
ap
plica
ble
outc
om
e.
In
put
proces
s
[
19]
wi
th
N
N
orga
niz
e
that
incl
ud
i
ng
weig
hts
delivere
d
ou
tc
om
e.
The
ou
t
pu
t
is
c
on
t
r
ast
ed
with
t
he
exp
ect
e
d
ou
tc
om
e,
if
the
de
li
ver
ed
out
co
m
e
per
fectl
y
m
at
ches
with
outp
ut
th
en
the
in
form
at
ion
is
rig
ht
el
se
m
ake
an
adjustm
ent
in
the
weig
ht.
N
N
syst
em
esse
ntial
ly
worked
with
w
ei
gh
ts.
The
b
ac
k prop
a
gatio
n neural
net
w
ork
u
se
d
in
this e
xperim
ent is cap
able of
gen
e
ral
iz
ing
the
trai
ning
da
ta
;
ho
w
e
ve
r,
t
he
net
wor
k
ca
nnot
be
a
ppli
ed
to
the
raw
data,
as
the
at
tribu
te
s
do
not
hav
e
un
i
form
pr
esen
ta
ti
on
.
S
om
e
inp
ut
at
trib
utes
are
not
nu
m
eric,
and
in
order
f
or
the
ne
ur
al
ne
twork
to
be
tr
ai
ne
d
on
the
te
st
data
,
the
ra
w
data
hav
e
to
be
pr
e
processe
d,
an
d,
a
s
t
he
value
s
and
ty
pes
dif
fe
r
f
or
eac
h
at
tri
bu
te
,
it
beco
m
es
neces
sary
that
the
i
nput
data
are
no
rm
alized.
The
values
of
a
c
onti
nu
ous
at
tri
bu
t
e
var
y
over
a
r
ang
e
,
wh
il
e
the
re
are
on
ly
tw
o
poss
ible
values
for
a
bin
ary
at
trib
ute
(
0
or
1).
T
hu
s
,
f
or
t
he
ne
two
rk
to
be
a
bl
e
to
process
the
dat
a
unif
or
m
ly
,
the
co
ntinuo
us
i
nput
at
trib
utes
hav
e
to
be
nor
m
al
iz
ed
to
a
num
ber
bet
wee
n
0
a
nd
1.
T
he
outp
ut
at
tribu
te
s
wer
e
processe
d
i
n
a
dif
fer
e
nt
m
a
nn
e
r.
If
the
ou
tpu
t
at
trib
ute
was
bin
a
ry,
th
e
tw
o
po
s
sible
netw
ork
outp
uts
a
re
0
a
nd
1.
On
th
e
oth
e
r
ha
nd,
con
ti
nu
ous
ou
t
pu
ts
can
not
be
treat
ed
i
n
th
e
sam
e
way, si
nce th
e
y ca
n
ta
ke
a
n u
nli
m
it
ed
nu
m
ber
of
value
s [2
0].
The
sim
plest f
un
ct
io
n
t
hat
does this is t
he
st
ep fu
nctio
n.
The
ste
p functi
on is
def
ine
d
a
s:
(1)
Th
us
,
cal
c
ulati
ng the
outp
ut of
our neu
r
on m
od
el
is
co
m
pr
is
ed of
tw
o
ste
ps
:
1)
Ca
lc
ulate
theinte
grat
io
n.
The
integ
rati
on,
as
de
fine
d
above,
is
the
s
um
for
vecto
rs
,w
,
x
an
d
b
scal
ar.
2)
Ca
lc
ulate
theo
utput
.
The
outp
ut
is
the
act
ivati
on
f
unct
ion
ap
plied
to
the
res
ult
of
ste
p
1.
Si
nce
th
e
act
ivati
on
f
un
c
ti
on
in
our
m
od
el
is
the
ste
p
f
un
ct
io
n,
the
ou
tp
ut
of
the
neuron
is
,
w
hi
c
h
is1whe
n
>0
a
nd0othe
r
wise.
Ther
e
is
a
co
ntinuo
us
a
ppr
ox
im
at
ion
of
the
ste
p
f
unct
ion
cal
le
d
the
log
ist
ic
c
urv
e,
orsigm
oi
d
functi
on,
de
note
d
as.
This
f
un
ct
ion
'
s
ou
tp
ut
r
ang
e
s
over
al
l
values
betwee
n0
a
nd1a
nd
m
akes
a
transiti
on
from
values
n
ea
r0
t
o values
n
ea
r1
at
at
x
=0
, s
im
il
ar to
the
ste
p func
ti
on
H(x).
Sigm
oid
al
Fun
ct
ion
is
r
ep
rese
nted
a
s:
(2)
Step
by step
exe
cution o
f
t
he pr
opos
e
d
m
et
ho
dolo
gy
Step 1
:
UML
di
agr
am
o
f
th
e s
yst
e
m
m
od
ule
unde
r
is i
m
po
rt
ed (sta
te
and a
ct
ivit
y diagr
am
s ar
e c
onside
re
d),
Step
2:
UML
diag
ram
is
con
ver
te
d
into
t
he
decisi
on
ta
ble
and
sta
te
c
hart
ta
bl
e
us
in
g
pa
rser
for
in
for
m
at
ion
extracti
on,
Step 3
:
Test
ca
se g
e
ner
at
io
n,
Step
4:
Now
t
he
te
st
case
ge
ner
at
e
d
us
i
ng
the
decisi
on
ta
ble
are
re
ge
ne
rated
us
in
g
G
eneti
c
al
gorith
m
fo
r
bette
r
c
od
e
cov
erag
e
u
si
ng cr
osso
ver
an
d
m
utati
on
,
Step
5:
T
he
re
dundancy
of
t
est
cases
is
in
creased
beca
use
of
t
he
re
ge
ne
rati
on
of
the
te
st
cases
us
in
g
the
Gen
et
ic
Algori
thm
, h
ence
Op
t
i
m
iz
ation
is
done,
Step 6
:
Op
ti
m
i
zat
ion
of the te
st case
s
us
in
g Ar
ti
fici
al
N
e
ural
N
et
w
ork.
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.
8
, N
o.
6
,
Dece
m
ber
2
01
8
:
5449
-
5456
5454
Figure
2. Bl
oc
k diag
ram
o
f
th
e pro
posed
m
eth
od
ology
Exam
ple il
lustrati
ng
the
m
et
ho
dolo
gy:
Figure
3. Acti
vi
ty
d
ia
gr
am
o
f t
he
Co
nnect
ion m
aking
pr
oces
s.
Con
si
der
i
ng th
e exam
ple o
f
the c
onnecti
on
m
aking
proces
s of a
ny d
e
vice
w
it
h dif
fer
e
nt
po
s
sibil
it
ie
s,
the U
ML
d
ia
gra
m
o
f
t
he
m
odule s
howing t
he
conn
ect
i
on m
akin
g process:
Table
1.
Stat
e char
t t
a
ble corr
esp
onding t
o
a
ct
ivit
y diagr
am
in fig
ur
e
3.
State
Of
f
lin
e
o
n
lin
e
Co
n
n
ectin
g
Er
ror
Of
f
lin
e
-
-
Switch
ON
-
On
lin
e
-
-
-
-
Co
n
n
ectin
g
-
-
-
Multip
le f
ailu
re
Er
ror
Switch
of
f
Co
n
n
ectio
n
estab
li
sh
ed
-
-
Test
cases
are
con
si
der
e
d
f
r
om
the
info
rm
at
ion
obta
ine
d
from
the
m
er
ged
gr
a
phs
of
the
softwa
re
m
od
ule
under
t
est
,
con
si
der
i
ng
two
te
st
case
s
for
an
exam
ple
as
11
an
d
18
,
know
obta
in
the
bin
a
ry
eq
uiv
al
ent
of the test
case
s,
Bi
nar
y eq
uiva
le
nt of
11 is
00001011
Bi
nar
y eq
uiva
le
nt of
18 is
00010010
Apply cr
os
s
ov
er (Ge
netic
A
l
gorithm
)
at
the
4
bits
of
t
he o
btained
b
i
nar
y
string o
f
t
he
te
st case
s,
New
bin
a
ry
str
ing
s
a
fter
cr
ossove
r
a
re
00
000010
an
d
0001
1011,
c
om
pu
te
the
decim
al
e
qu
i
valent
of
the obtai
ne
d bi
n
ary st
rin
gs
as:
000000
10
=
2, 0001
1011=2
7
2
a
nd
27
are
t
wo
ex
pecte
d
te
st
cases
for
t
he
syst
em
m
od
ule
un
der
te
st
he
nce
know
t
he
t
est
su
it
e
T
is
wr
it
te
n
as:
T=
{1
1,
18,
2,
27
}
,
sim
il
arly
we
ca
n
go
m
or
e
cr
os
s
over
by
choosi
ng
ot
he
r
bin
a
ry
bit
posit
ion
onto
the
bi
na
ry
bits
and
al
s
o
the
ge
netic
al
gorithm
can
al
so
be
a
ppli
ed
bet
wee
n
the
ne
wly
ob
ta
ined
te
st
cases
and
al
read
y
e
xisti
ng
te
st
cases
.
N
ow
after
ob
ta
in
ing
t
he
e
xpect
ed
te
st
cases
the
A
NN
chec
ks
wh
et
her
we
ha
ve
th
e
sam
e
te
st
case
al
read
y
in
t
he
pool o
f
t
he
te
s
t
cases
w
ritt
en
as
T.
F
or
that
t
he
ANN
assi
gns
weig
hts
t
o
th
e
input
(n
e
w
te
st
cases)
an
d
for
ward
the
sam
e
towards
the
hi
dd
e
n
la
ye
rs
of
the n
e
twork
a
nd
afte
r
processin
g
f
rom
the
hidden
la
ye
r
t
o
ou
t
pu
t
la
ye
r
a
s
0
or
1
w
her
e
0
sig
nifies
not
pr
ese
nt
an
d
1
s
ign
ifie
s
pr
ese
nc
e
of
the
te
st
c
ase
in
the
te
st
po
ol.
The
ultim
at
e
go
al
of
t
he
ANN
is
to
re
duce
the
redu
nd
a
nc
y
of
t
he
te
st
cases
ge
ne
rated
a
fter
app
ly
in
g gene
ti
c algori
thm
f
or
gen
e
rati
ng th
e ex
pected
test
cases for
b
et
te
r
test
ing
of the
s
yst
e
m
m
od
ule.
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
Test C
as
e
Opti
miza
ti
on
and R
edun
dancy
Red
uction U
sin
g G
A an
d
Ne
ural
Ne
tw
or
ks
(
Itti
H
oo
da
)
5455
4.
RESU
LT
A
N
ALYSIS
The
pro
posed
m
et
ho
dolo
gy
i
s
qual
it
at
ively
analy
zed
on
th
e
basis
certai
n
factor
as
ef
fici
ency
m
eans
the
abili
ty
of
the
m
et
ho
dolo
gy
to
detect
t
he
error
in
t
he
a
va
il
able
co
de,
num
ber
of
te
st
c
ases
ge
ne
rated
wh
ic
h
def
i
nes
li
ke
w
hen
the
te
st
ca
ses
ge
ner
at
e
d
a
re
m
or
e
then
there
is
m
or
e
c
han
ce
of
te
sti
ng
the
m
odule
wh
ic
h
is
unde
r
te
st,
c
ode
co
ver
a
ge.
Re
dundancy
of
th
e
te
st
cases
tha
t
m
eans
to
e
nsure
that
wh
et
he
r
the
ge
ner
at
e
d
te
s
t
cases
are
rep
e
at
ed
in
the
fin
al
li
st,
cod
e
cov
era
ge
m
eans
le
vel
to
wh
ic
h
the
te
st
cases
cov
e
rin
g
the
c
od
e
or
te
sti
ng
the
cod
e
:
Table
1 Qu
al
it
at
ive Analy
sis of the
relat
ed work
a
nd the
pro
posed
m
et
ho
do
l
og
y.
Para
m
eters
Ef
f
icien
cy
No
.
o
f
T
est cases
g
en
erate
d
Red
u
n
d
an
cy
Co
d
e
Co
v
erage
Metho
d
o
lo
g
ies
1.
Mediu
m
Mediu
m
Hig
h
Mod
erate
2.
Low
Mediu
m
Hig
h
Mod
erate
3.
Mediu
m
Mediu
m
Mod
erate
Mod
erate
4.
Mediu
m
Mediu
m
Hig
h
Low
5.
Mediu
m
Mediu
m
Hig
h
Low
6.
Hig
h
Increased
Low
Hig
h
Table
2
Fact
ors and
Im
pact Ra
ti
o.
Facto
rs
I
m
p
act
in
L
izh
an
g
et al
[
6
]
I
m
p
act
in
And
rews
et
al.
[
7
]
I
m
p
acts in
Prop
o
sed
M
eth
o
d
o
lo
g
y
Reg
u
lated
Deoictio
n
55%
50%
60%
Valid
and
I
n
v
alid
inp
u
ts
38%
35%
45%
Go
o
d
detaillev
el
44%
48%
50%
Variation
in test
ca
ses
75%
70%
80%
Un
d
erstand
Sy
ste
m
f
ra
m
ewo
rk
50%
55%
60%
Clear beg
in
n
in
g
positio
n
60%
55%
65%
Test des
ig
n
techn
iq
u
es
80%
70%
84%
Accurate su
p
p
o
sitio
n
s
70%
78%
80%
Test cae
s
asses
s
m
e
n
t
50%
45%
55%
Tidy
u
p
af
ter
ex
ec
u
tio
n
78%
70%
85%
The
se
factors
are
play
ing
ver
y
i
m
porta
nt
role
in
te
st
ca
ses.
B
y
a
naly
z
ing
it
finds
tha
t
el
abo
rated
fac
tors
havi
n
g
diffe
ren
t
l
eve
l
of
infl
uen
ce
on
te
s
t
ca
ses
for
d
iffe
r
ent
t
ec
hniqu
es
d
iscussed
in
the
literature
r
evi
ew
.
All
the
se
f
actors
hav
e
diffe
ren
t
impac
t
rat
io
and
the
se
ent
ir
e
ratios
desc
ribe
d
in
ta
b
le
2
.
Here
ar
e
four
sets
of
te
st
c
ase
s,
in
which
m
ax
imum
impact
of
r
egul
a
te
d
dep
ic
t
ion
is
60%
for
the
pro
posed
m
et
hodol
og
y
and
is
be
tt
e
r
tha
n
al
l
d
iscussed
in
the
rev
i
e
w.
For
diffe
ren
t
techniq
ues
conside
red
,
m
axi
m
um
impact
of
input
ana
l
y
s
is
is
38%
is
for
the
proposed
m
et
hodolog
y
.
Thi
s
stud
y
indi
c
at
es
that
29%
of
the
subjec
ts
did
not
rea
d
the
dire
c
ti
on
of
convey
an
ce
of
e
xper
iment
and
16%
did
not
fini
sh
the
whole
l
a
y
out
,
so
it'
s
impact 45% for t
he
proposed m
et
ho
dolog
y
.
5.
CONCL
US
I
O
N
Test
ing
is
the
m
os
t
cri
ti
cal
p
art
of
the
S
of
t
war
e
Dev
el
opm
ent
Lifecy
cle,
as
it
is
so
m
et
hin
g
up
on
wh
ic
h
the
final
delivery
of
t
he
pro
du
ct
is
depend
e
nt.
It
is
ti
m
e
con
s
um
ing
and
a
n
i
ntensive
process
,
the
r
efore,
enh
a
nce
d
te
ch
niques
a
nd
i
nnov
at
ive
m
et
ho
do
l
og
ie
s
a
re
re
qu
isi
te
.
F
or
w
hi
ch
the
ef
fici
en
t
te
sti
ng
m
et
ho
do
l
ogy
is
require
d
w
hich
ca
n
gen
e
rat
e
te
st
cases
as
m
any
as
possi
bl
e
with
c
onside
rin
g
the
fact
or
li
ke
re
dundanc
y
an
d
al
so
s
hould
prov
i
de
m
axi
m
um
cod
e
co
ver
a
ge.
I
n
the
pr
opose
d
m
et
h
odology
the
UM
L
al
gorithm
,
Gen
et
ic
al
gorithm
and
la
stl
y
the
trai
ned
A
N
N
is
bei
ng
us
e
d
f
or
re
du
ci
ng
t
he
re
dunda
ncy
of
t
he
gen
e
rated
te
st
cases.
The
UML
al
gorithm
in
the
form
of
the
Sequ
e
nce
diagra
m
and
the
sta
te
diagr
am
is
us
ed
for
the
bette
r
represe
ntati
on
of
the
s
of
t
ware
m
od
ule
w
orkin
g
so
that
t
he
te
st
case
giv
es
the
m
axi
m
u
m
cod
e
co
ver
a
ge
.
Gen
et
ic
al
gorit
hm
is
being
use
d
f
or
ge
ner
at
i
ng
al
l
po
s
sible
te
st
cases
for
the
m
od
ule
un
der
te
st.
As
pe
r
the
resu
lt
s
the
al
gorithm
prov
i
de
s
the
m
axi
m
um
eff
ic
ie
ncy
an
d
co
de
c
overa
ge
al
ong
with
deali
ng
with
the
redu
nd
a
ncy
of
the g
e
ne
rated t
est
cases
.
REFERE
NCE
S
[1]
P.
Ron,
“
Software
t
esti
ng
,
”
In
dia
napoli
s: Sam’s
,
v
ol.
2
,
2001
.
[2]
S.
Am
la
nd,
"Ris
k
-
base
d
t
esti
ng
,
"
Journal
o
f
S
ystem
s and
Soft
ware
,
vol
.
53
(
3
)
,
pp
.
287
–
295,
Sep
.
2
000.
[3]
Redmill
and
Fel
ix,
“
The
or
y
and
Prac
ti
c
e
of
Risk
-
base
d
Te
sting
,
”
Soft
ware
Testing,
Ve
rif
ic
at
ion
and
Re
li
ab
il
i
ty
,
vol.
15
(
1
)
,
M
ar 2005.
[4]
B.
Agarwa
l, e
t
a
l
.
,
“
Software
eng
i
nee
ring
and
te
st
i
ng
,
”
Jone
s
&
Ba
rtle
tt Learning
,
2010.
[5]
K.
Bogdan
,
“
Autom
at
ed
softwar
e
t
est
da
ta
gen
e
rat
ion
,
”
Soft
war
e
Eng
ine
ering
,
I
EE
E
Tr
ansacti
o
ns
,
pp.
870
-
879,
vol.
16
(
8
),
1990
.
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.
8
, N
o.
6
,
Dece
m
ber
2
01
8
:
5449
-
5456
5456
[6]
Jac
obson, et al.,
“
The
unified
sof
twar
e
d
evelopm
ent
pro
ce
ss
,
”
R
e
ading:
Add
ison
-
We
sley
,
vo
l. 1, 1
999.
[7]
Eve
re
tt
,
e
t
a
l.,
“
Software
t
esti
ng
:
te
st
ing
a
cro
ss
t
he
en
ti
re
softwa
re
dev
el
opm
ent
li
fe
c
y
c
le
,
”
John
Wil
e
y
&
Son
s,
2007.
[8]
Verm
a,
“
Automa
te
d
T
est
c
ase
gene
ra
ti
on
usin
g
UM
L
dia
gra
m
s
base
d
on
b
eha
vior
,
”
Int
ern
ati
onal
Journal
of
Innov
ati
ons
in
E
ngine
ering
and
Technol
ogy
(
IJI
ET)
,
vol.
4
,
issue
1
,
pp.
31
-
39,
Jun 2014.
[9]
M.
V
anmali
,
M.
La
st,
A.
K
ande
l
,
“
Us
ing
a
Neura
l
Network
in
the
Software
Te
st
in
g
Proce
ss
,
”
Inte
r
nati
onal
Journa
l
Of
Intelli
g
ent Sys
te
ms
,
vol
.
17
,
p
p.
45
-
62
,
2002
.
[10]
J.
John,
“
A
Perf
orm
anc
e
Based
Stud
y
of
Software
Te
sting
using
Artifi
c
ia
l
Neura
l
Network
,
”
Inte
r
nat
ional
Journa
l
of
Logic
Based
I
nte
lligent Sy
st
ems
,
vol. 1
(
1
)
,
pp
.
45
-
60,
2011
.
[11]
Sharm
a,
R.
Pat
ani
and
A.
Agg
arwa
l,
“
Softwar
e
Te
sting
Us
ing
Gene
tic
Algori
thms
,
”
Inte
rnati
onal
Journal
of
Computer
Scien
ce
&
Engi
nee
ring S
urve
y
(
IJCSES)
,
vol.
7
(
2
)
,
pp
.
21
-
33,
Apr
2016
.
[12]
S.
Keshava
rz
an
d
R.
Javida
n,
“
Software
Qualit
y
Control
Based
on
Gene
ti
c
Algor
it
hm
,
”
Inte
rnat
i
onal
Journal
of
Computer
Theory
and
Engi
n
ee
ri
ng
,
vol
.
3
,
Iss
ue
4,
2011
.
[13]
Soni
y
a
Ma
li
k,
“
Software
Te
sti
ng
Us
ing
Gene
ti
c
Algori
thm
,
”
Inte
rnational
Confe
renc
e
on
Technol
ogi
es
for
Sustainabi
lity
-
E
ngine
ering
,
In
fo
rm
ati
on
Technology
,
Manag
eme
nt
,
pp
.
1011
-
101
6,
Nov 2015
.
[14]
M.
La
st
,
S. E
y
al,
and
A.
K
ande
l
,
“
Eff
ec
ti
v
e
Bl
ac
k
-
Box
Te
sting
wit
h
Gene
tic
Algor
i
thms
,
”
Department
of
Computer
Sci
en
ce and
Eng
ine
ering
,
B
en
-
Gurion
Univer
si
t
y
of
th
e
Neg
ev, B
ee
r
-
Sheva
,
Isr
ae
l
,
2005
[15]
Ranj
it
a
Sw
ai
n
,
Vikas
Panthi
,
“
Autom
at
ic
T
est
ca
se
Gene
ration
from
UM
L
Stat
e
Char
t
Diagr
a
m
”,
Inte
rnation
al
Journal
of
Computer
App
li
ca
ti
on
s (
0975
–
8887)
,
Vol.
42
–
No.7, p
p.
26
-
36,
Marc
h
2012.
[16]
V.Mar
y
Sum
al
a
t
h
a
et
al
.
,
“
UM
L
base
d
Autom
at
ed
Te
st
Case
Gene
ration
tec
hnique
using
Acti
vi
t
y
-
Sequen
ce
dia
gra
m
”,
The
I
nte
rnational
Jou
rnal
of
Comput
er
Scienc
e
&
A
ppli
cations
(
TIJCSA)
,
Vol.
1
,
No.
9,
pp.
58
-
7
1,
Novem
ber
2012.
[17]
M.
Phili
ps,
N.
P
awa
r
,
N.
Jos
hi
,
S.Khande
bha
rad
,
S.
Deshm
ukh,
K.T
ambe
,
“
Automa
te
d
Te
st
C
ase
Gene
ration
Us
ing
Multi
ple
Modelling
T
ec
hniqu
es
”
,
Int
ernati
onal
J
ournal
of
S
ci
en
c
e
and
Re
search
(
IJS
R)
,
Vol.
3
Is
sue
3,
pp
.
722
-
725,
Mar
ch
201
4.
[18]
P.R.
Srivasta
v
aand
Ta
i
-
hoon
Ki
m
,
“
Applic
at
ion
of
Gene
tic
Algo
rit
hm
in
Software
Te
sting
”,
In
te
r
nati
onal
Journal
of
Soft
war
e
Eng
i
nee
ring a
nd
Its
Appl
ic
a
ti
ons
,
Vol.
3
,
No
.
4,
pp.
87
-
96,
Octob
er
200
9.
[19]
Mee
nakshi
Van
m
al
i,
Mark
La
st
,
Abraha
m
Kande
l,
“
Us
ing
a
Neura
l
Network
i
n
the
Software
Te
sting
Proce
ss
”,
Inte
rnational
Jo
urnal
o
f
In
te
l
li
g
e
nt
Syst
ems
,
pp.
4
5
-
62,
Janu
ar
y
20
02.
[20]
Gall
an
t, “S
.
I.
Ne
ura
l
Ne
twork
L
e
arn
ing and Expe
rt
S
y
st
ems
”
,
Ma
ss
achuse
tt
s Insti
t
ute
o
f
Te
chnol
og
y
,
US
A,
1994
.
Evaluation Warning : The document was created with Spire.PDF for Python.