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,
pp.
5253
~
5259
IS
S
N: 20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v8
i
6
.
pp
5253
-
52
59
5253
Journ
al h
om
e
page
:
http:
//
ia
es
core
.c
om/
journa
ls
/i
ndex.
ph
p/IJECE
Using D
ata Mini
ng to
Ide
ntify CO
SMIC F
un
ctio
n
Point
Measur
em
ent Co
mp
ete
nc
e
Selami B
agri
yanik
1
,
Ad
e
m
Ka
r
ahoca
2
1
Digit
al Learn
in
g
Soluti
ons
Te
ch
nolog
y
Depa
r
tment, T
urk
ce
l
l Te
chnol
og
y
,
Turk
e
y
2
Depa
rtment of
Software
Eng
ineeri
ng,
Bah
ce
sehi
r
Univer
sit
y
,
Tur
ke
y
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Feb
27
, 201
8
Re
vised
Jun
18
201
8
Accepte
d
J
ul
29
, 2
01
8
Cosm
ic
Functi
on
Point
(CFP
)
me
asure
m
ent
err
or
s le
ads
budget
,
s
che
dul
e
an
d
qual
ity
proble
m
s
in
software
pr
oje
c
ts.
Th
ere
for
e,
i
t’s
important
to
ide
n
t
i
f
y
and
pl
an
r
equi
re
m
ent
s
engi
ne
ers’
CF
P
tra
ini
ng
nee
d
qu
ic
kl
y
an
d
cor
re
c
t
l
y
.
The
purpose
of
thi
s
pape
r
is
to
ide
nti
f
y
softw
are
req
uir
ement
s
engi
nee
rs’
COS
MIC
Functi
on
Point
m
ea
s
ure
m
ent
compet
enc
e
dev
el
opm
e
nt
nee
d
b
y
using
m
ac
hine
le
arn
ing
al
gor
ithm
s
and
req
uire
m
ent
s
art
if
acts
cre
a
te
d
b
y
engi
ne
ers.
Us
ed
art
ifacts
have
bee
n
provide
d
b
y
a
la
rg
e
service
an
d
te
chno
lo
g
y
compan
y
ec
os
y
stem
in
Telco
.
First,
f
ea
tur
e
set
h
as
be
en
ext
r
acte
d
from
the
req
u
ir
ements
m
odel
at
h
and.
To
do
the
data
pr
ep
ara
t
ion
for
educ
a
ti
ona
l
dat
a
m
ini
ng,
req
uirem
ent
s
and
CO
S
MIC
Functi
on
P
oint
(CFP
)
audi
t
do
cument
s
have
b
ee
n
c
onver
te
d
int
o
CF
P
dat
a
set
b
ase
d
on
th
e
designe
d
fe
at
ure
set.
Thi
s
da
ta
se
t
has
bee
n
used
t
o
tra
in
and
te
st
t
he
m
ac
hi
n
e
le
arn
ing
m
odel
s
b
y
design
ing
t
wo
diffe
ren
t
ex
per
iment
se
tt
ing
s
to
re
ac
h
stat
isti
ca
l
l
y
signifi
c
ant
result
s.
Te
n
diffe
r
ent
m
ac
hin
e
le
a
rning
al
gorit
hm
s
have
be
en
used.
Final
l
y
,
al
gori
th
m
per
form
anc
es
have
be
en
compare
d
with
a
base
li
n
e
and
eac
h
othe
r
to
find
th
e
best
per
form
in
g
m
odel
s on
thi
s
dat
a
set
.
In
conc
lusion
,
RE
PTree
,
OneR
,
a
nd
Support
Vec
tor
Mac
hine
s
(
SV
M)
with
Sequent
i
al
Mi
nimal
Optimiz
at
ion
(SM
O)
al
gorit
hm
s
a
chi
ev
ed
top
per
form
anc
e
in
f
ore
ca
st
ing
r
equi
r
ements
engi
n
ee
r
s’ CFP
tra
ini
ng
nee
d.
Ke
yw
or
d:
COSMIC
funti
on point
Ed
ucati
on
al
da
ta
m
ining
Ma
chine
le
a
rn
i
ng alg
or
it
hm
s
Re
qu
irem
ents
analy
st
Re
qu
irem
ents
arti
facts
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
:
Ad
em
K
ara
hoc
a
,
Dep
a
rtm
ent o
f Sof
t
war
e
E
ng
i
neer
i
ng,
Ba
hceseh
i
r Uni
ver
sit
y,
Faculty
of E
ngineerin
g, Be
sik
ta
s,
Ista
nbul, 3
4349, T
urkey.
Em
a
il
:
ade
m
.k
arahoca@
en
g.bau.ed
u.
t
r;
aka
rahoca@
gm
ai
l.
com
1.
INTROD
U
CTION
Re
qu
irem
ents
eng
i
neer
s
a
re
on
e
of
the
ke
y
pr
ofi
le
s
within
softwa
re
dev
el
op
m
ent
t
ea
m
s.
They
balances
al
l
pr
oj
ect
sta
kehold
ers’
e
xpect
at
ion
s
f
r
om
idea
to
post
pro
du
ct
i
on
phases
.
Re
qu
irem
ens
eng
ine
erin
g
is
no
t
s
olely
a
te
chn
ic
al
di
sci
pline.
Addi
ti
on
al
ly
,
it
al
s
o
has
a
n
inte
r
-
discipli
nar
y
natu
re
that
co
ncerns
Cognit
ive
Psyc
ho
l
og
y,
A
nthro
po
l
og
y,
S
ociol
og
y,
Lin
gu
ist
ic
s
an
d
P
hilos
ophy
asp
ect
s of
t
he
s
ubj
ect
[
1
]
. Thus,
they
hav
e
a
dram
atical
i
m
p
act
on
the
su
c
cess
of
the
softw
are
product
s
and
their
co
ntinuo
us
com
petence
dev
el
op
m
ent
is crit
ic
al
.
Functi
on
al
siz
e
m
easur
em
ent
(F
SM)
is
a
n
i
m
po
rtant
ta
s
k
that
is
us
ed
for
sc
op
i
ng,
budget
ing,
m
anag
in
g
outsourci
ng
c
ontra
ct
s,
ef
fort
est
i
m
at
ion
,
et
c.
T
his
ta
s
k
is
ge
ner
al
ly
un
der
the
re
spo
ns
ibil
it
y
of
syst
e
m
analy
st
s
or
re
quirem
e
nts
en
gin
ee
rs
(
REs).
CFP
is
on
e
of
t
he
rec
ent
FSM
m
et
h
od
s
.
F
un
ct
io
n
Po
int
var
ia
nts
are
m
ai
nly
us
ed
in
s
of
t
war
e
co
st
est
i
m
ation
[
2
]
and
pro
du
ct
ivit
y
of
the
de
velop
m
ent
organ
i
sat
ion
s
[
3
]
.
Functi
on
Po
int
is
al
so
a
good
i
nd
ic
a
tor
i
n
ide
ntif
yi
ng
bu
si
n
ess
com
plexity
of
t
he
s
of
t
wa
re
[
4
]
.
Additi
on
al
ly
It
’s
CF
P
var
ia
nt
is
al
so
a
str
ong
too
l
f
or
re
qu
i
r
e
m
ent
qual
it
y
and
pr
ocess
im
pr
ovem
ent
[5
]
.
CFP
m
easur
em
ent
error
s
,
m
ade
by
re
quirem
ents
en
gin
eer
s,
l
eads
budget
,
sche
du
le
a
nd
q
ualit
y
pro
ble
m
s
in
so
ft
war
e
proje
ct
s.
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:
5253
-
52
59
5254
Ther
e
f
or
e,
it
’s
cru
ci
al
to
fores
ee
and
pla
n
re
qu
i
rem
ents
eng
inee
rs’
CFP
trai
ning
nee
d
in
a
qu
ic
k
an
d
correct
m
ann
er.
A
r
ecent
pa
per
po
i
nts
out
that
CFP
trai
ning
nee
d
s
hould
be
re
pr
ese
nted
m
or
e
in
higher
edu
cat
io
n
[
6]
.
W
e
t
hink
trai
ning
is
al
s
o
c
r
it
ic
al
in
the
w
ork
place
set
ti
ng
a
nd
REs
s
houl
d
be
c
on
ti
nual
ly
dev
el
op
e
d
i
n
CFP
com
peten
ce
w
hen
nee
d
arises.
T
raini
ng
is
t
he
do
m
inati
ng
facto
r
f
or
qual
it
y
i
m
pr
ovem
ent
of
F
SM
[7
]
.
Fa
ct
or
s
that
cau
s
e
inco
ns
ist
ent
and
i
naccurat
e
CF
P
m
e
asur
em
ents
m
igh
t
be
i
m
pr
ov
e
d
by
trai
ni
ng
[8]
.
I
n
this
stu
dy,
requirem
ents
en
gin
ee
rs
C
FP
trai
ning
ne
ed
has
bee
n
forecast
ed
by
us
i
ng
the
arti
facts
they
pro
du
ce
d
i
n
th
e wor
kpla
ce an
d
m
achine lear
ning alg
ori
thm
s.
Data
m
ining
or
s
softwa
re
a
naly
ti
cs
stud
ie
s
that
use
re
quirem
ents
en
gi
neer
i
ng
arti
fac
ts
are
sca
rce
[9]
.
F
or
e
xam
ple,
on
e
of
t
hes
e
rar
e
stu
dies
a
i
m
s
early
te
s
t
e
ffor
t
pr
e
dicti
on
by
us
in
g
UM
L
diag
ram
s
[
10
]
.
On
the
oth
er
hand,
softwa
re
data
m
ining
st
ud
ie
s
w
hich
are
ba
s
ed
on
s
ofwa
re
cod
e
a
nd
co
de
chang
e
arti
fact
s
are
com
m
on
[9
]
.
Fo
r
insta
nce,
cod
e
sm
el
ls
in
the
s
ource
co
de
hav
e
been
inv
est
igate
d
usi
ng
Ne
ur
al
Ne
twork
Mod
el
s
i
n
a
re
cent
stu
dy
[
11]
.
W
e
obse
rv
e
that
us
i
ng
CFP
data
an
d
data
m
ining
for
E
ducat
ion
al
pur
poses
is
even
m
or
e
ra
re
in
t
he
li
te
ratu
r
e.
A
s
far
as
w
e
know,
this
re
search
is
t
he
fi
rst
stu
dy
in
the
li
te
ratur
e
t
hat
us
es
CFP
data
and
edu
cat
io
nal
dat
a
m
ining
to
i
m
pro
ve
REs’
CF
P
m
easur
em
ent
capab
il
it
ie
s.
The
rest
of
the
pap
e
r
is
orga
nized
a
s
fo
ll
ow
s:
i
n
the
2nd
sect
io
n,
a
bac
kgr
ound
on
Data
Mi
ni
ng,
Ma
c
hin
e
Lear
ning
Al
gorithm
s,
CFP an
d
stu
dy
d
et
ai
ls are p
rovide
d.
Result
s ar
e p
rese
nte
d
in the 3
r
d
sect
ion
a
nd
is fo
ll
owed by con
cl
usi
on
s in
the 4
t
h
sect
io
n.
2.
RESEA
R
CH MET
HO
D
In
this
se
ct
io
n,
first
of
al
l,
m
achine
le
ar
ning
a
nd
CFP
m
et
ho
ds
a
re
ex
plaine
d
br
i
efly
in
t
he
su
bse
ct
ions
2.1
an
d
2.2
.
Sec
ond,
CF
P
trai
ni
ng
need
pr
e
di
ct
ion
us
ec
ase,
featur
e
set
desi
gn
a
nd
data
ga
therin
g
and
pr
e
pa
rati
on
phases
of
t
he
stud
y
is
pre
sented
in
2.3,
2.4
an
d
2.5
.
F
inall
y,
m
od
el
s
trai
ning
detai
ls
and
evalua
ti
on
res
ul
ts are g
i
ven in
2
.
6.
2.1.
Data
Mining
an
d
Machine
Le
arnin
g Alg
orithm
s
Data
Mi
ning
is
def
i
ned
as
“t
he
process
of
di
sco
ver
in
g
patte
rn
s
,
aut
om
atical
ly
or
sem
i
-
autom
atical
ly
,
in
la
r
ge
quantit
ie
s
of
data”
[
12
]
.
Kno
wled
ge
disc
ov
e
ry
f
r
om
data
(K
D
D
)
is
an
oth
e
r
c
omm
on
t
erm
us
ed
in
t
he
li
te
ratur
e
[
13
]
. Foll
owin
g
al
go
rithm
s w
hich were
im
ple
m
e
nted
i
n Wek
a
[
14
]
a
re
us
e
d
in
this stu
dy:
Rando
m
Fore
st
(RF):
This
is
an
ensem
ble
le
arn
in
g
m
et
ho
d
c
on
sis
ti
ng
of
a
set
of
decisi
on
tr
ee
cl
assifi
ers.
Ea
c
h
tree
in
t
he fo
rest is trig
ge
re
d by an
in
dep
e
nd
e
ntly
created
r
a
ndom
n
um
ber
vector
[
15
]
.
Na
ïv
e
Bayes
(N
B)
:
This
m
et
hod
us
es
B
ay
es’
ru
le
to
do
the
cl
ass
ific
at
ion
by
com
pu
ti
ng
cl
as
s
pro
bab
il
it
ie
s
and
us
i
ng
obse
r
ved
at
trib
ute
va
lues.
T
he
m
eth
od
is
cal
le
d
“
naïve”
si
nce
it
has
t
wo
basic
assum
ption
s:
a
tt
ribu
te
s
are
c
onditi
on
al
ly
in
dep
e
ndent
a
nd
no
hi
dd
e
n
fac
tor
im
pacts
on
the
pre
dicti
on
process
[
16
]
.
RE
PTree:
Thi
s
is
a
fast
deci
sion
tree
al
gor
it
h
m
that
gen
erates
a
decisi
on
tree
us
in
g
in
form
ation
gai
n
m
et
ho
d
t
o
s
plit
[
17
]
. Missi
ng
values
are
m
an
aged as i
n
C4
.
5
al
go
rithm
[
18
]
.
J48:
It is
a Ja
va
i
m
ple
m
entat
i
on of a
sli
gh
tl
y dif
fer
e
nt
ver
si
on of C
4.5
[
17
]
.
LMT
:
Lo
gisti
c
Mod
el
Trees
are
sta
nd
a
r
d
de
ci
sion
trees
w
hich
us
e
lo
gist
ic
reg
ressi
on
f
un
ct
io
ns
at
their
le
aves
[
19
]
.
Mul
ti
layer
Perceptro
n
(ML
P):
MLP
is
a
feed
-
f
orward
arti
fici
al
neu
ral
netw
ork
w
hich
use
s
bac
k
pro
pag
at
io
n
tr
ai
nin
g
al
gorith
m
.
It
is
a
sys
tem
of
interc
onne
ct
ed
node
s
or
neur
on
s
w
hich
m
aps
an
in
put
vecto
r
int
o
a
n
outp
ut
vector
to
m
ai
ntain
a
nonlin
e
ar
rel
at
ion
[
20
]
.
T
he
ne
uro
ns
a
re
co
nn
ect
e
d
vi
a
weig
hts a
nd ou
tpu
t si
g
nals
[
20
]
.
Supp
or
t
Ve
ct
or
Machines
(
SVM
)
and
Se
qu
enti
al
Mi
ni
ma
l
Op
ti
mi
z
ation
(
SMO
):
I
n
li
nea
r
case,
an
SV
M
is
a
hype
rp
la
ne
that
set
a
boun
dar
y
between
so
m
e
posit
ive
instances
an
d
ne
gative
i
ns
ta
nces
[
21
]
.
It
can
al
so
be
fu
rt
her
exte
nd
e
d
to
no
n
-
l
inear
cases
[
21
]
.
Tr
ai
ning
an
SV
M
re
qu
i
res
quad
rat
ic
pro
gr
am
m
ing
(Q
P
)
optim
iz
a
ti
on
pro
blem
so
lvi
ng
w
hich
is
a
ver
y
ti
m
e
and
m
e
m
or
y
c
onsu
m
in
g
op
e
rati
on a
nd
SMO is a
subst
antia
l im
pr
ove
m
ent o
n t
he
original trai
ning a
lgorit
hm
[
21
]
.
K
-
neare
st
Nei
ghbour
Cl
as
si
fier
(IB
k):
It
c
la
ssifie
s
a
data
point
based
on
it
s
k
m
os
t
si
m
il
ar
oth
er
dat
a
po
i
nts
[
22
]
.
ZeroR
:
It
pre
di
ct
s
the
m
ajo
rit
y
cl
ass
of
no
m
inal
te
st
data
wh
il
e
it
pr
e
dic
ts
the
aver
a
ge
value
if
num
eri
c
cl
ass
is
the
cas
e
[
12
]
.
I
n
t
his
s
tud
y,
it
will
be
us
e
d
as
a
basel
ine
f
or
t
he
pe
rfor
m
ance
of
m
a
chine
le
ar
ning
al
gorithm
s.
On
eR
:
T
his
m
et
ho
d
cl
assi
f
ie
s
instances
base
d
on
a
one
r
ule
w
hic
h
is
extracte
d
from
a
sing
le
at
tribu
te
[
23
]
.
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
Using
Da
t
a
Mi
ning to
Ide
ntif
y COSMIC
Fu
nc
ti
on
P
oin
t
Me
as
ure
me
nt C
ompete
nce
(
Sel
ami B
ag
riy
an
ik
)
5255
2.2.
COSMI
C
F
un
ction
Po
in
t (C
FP)
So
ft
war
e
f
unct
ion
al
siz
e
m
ea
su
rem
ent
(FS
M)
ha
s
been
in
us
e
f
or
m
or
e
t
han
f
ort
y
ye
ars
[
24
]
.
Th
ere
are
m
any
FSM
m
et
ho
ds
[
25
]
.
C
OS
MIC
Functi
on
Po
i
nt
is
a
ne
w
gen
e
rati
on
s
oft
war
e
f
unct
io
na
l
siz
e
m
easur
em
ent
m
et
ho
d.
First
ver
si
on
of
t
he
m
et
ho
d
was
publis
hed
in
19
88
[
26
]
.
It’s
one
of
f
our
ISO
certi
fied
FSM
m
et
ho
ds
w
hich
are
do
m
inati
ng
in
th
e
in
du
st
ry:
IFPU
G,
C
OS
MI
C,
NE
SMA
a
nd
Ma
r
k
II
[
25
]
.
CF
P
m
easur
em
ent i
s a thr
e
e
-
ste
p
proces
s: m
easur
e
m
ent strate
gy,
m
app
ing
a
nd
m
easur
em
ent steps.
P
urp
os
e,
sco
pe,
and
le
vel
of
gran
ularit
y
of
t
he
m
easur
em
e
nt
are
dete
rm
i
ned
i
n
the
fir
s
t
ste
p;
in
m
app
in
g
phase,
functi
on
al
processes
a
nd
data
gro
ups
in
the
re
quire
m
ents
are
det
erm
ined;
in
th
e
final
sta
ge,
data
m
ov
em
ents
are
sp
eci
fied
a
nd
counted
,
f
or
al
l
fu
nctio
nal
pr
ocesses
[
26
]
.
CFP
is
the
fun
ct
ion
al
siz
ing
m
easur
em
ent
m
et
ho
d
that i
s u
se
d
i
n
t
he
c
om
pan
y u
nder
stu
dy
[
3,7,27
]
.
2.3.
CFP Req
uire
ment On
to
l
ogy
Re
qu
irem
ent
artefact
s
that
w
il
l
be
us
ed
in
trai
ning
the
m
achine
le
ar
ning
m
od
el
s
are
i
ns
ta
nces
of
requirem
ent
and
CFP
onto
log
ie
s
desig
ne
d
in
[
3
]
.
C
urren
tl
y,
this
re
qu
irem
ent
ontolo
gy
a
nd
CF
P
m
easur
em
ents
are
sta
nd
a
rd
m
et
ho
ds
use
d
by
requirem
ents
eng
inee
rs
w
it
hin
the
sa
m
e
te
le
co
m
m
un
icati
ons
com
pan
y
in
w
hich
t
his
stu
dy
is
cond
ucted.
P
erio
dical
ly
,
a
su
bse
t
of
al
l
requirem
ents
do
c
um
en
ts
are
ra
ndom
ly
sel
ect
ed
and
e
xam
ined
by
internal
au
dit
te
a
m
m
anu
al
ly
to
identify
error
s
in
CFP
m
easur
em
ents.
Af
te
r
each
aud
it
,
pr
ob
le
m
at
ic
CFP
m
eas
ur
em
ents
are
identifie
d,
rec
orde
d
an
d
pote
ntial
le
arn
ing
needs
are
re
ported
to
requirem
ents
eng
i
neer
i
ng
m
a
na
gem
ent.
By
this
data
m
ining
researc
h,
t
he
m
anu
al
exam
inati
on
proc
ess
by
aud
it
te
am
is
i
nten
ded
to
be
sem
i
-
autom
at
e
d
an
d
le
arn
i
ng
opport
un
it
ie
s
will
autom
atic
al
ly
be
extracte
d
from
requirem
ents docu
m
ents.
2.4.
Feature Se
t
E
xt
r
act
i
on
fr
om
C
F
P
Requ
i
reme
nt
On
to
l
og
y
CFP
O
nto
l
og
y
co
ncep
ts
a
re
sh
ow
n
in
the
seco
nd
c
olu
m
n
of
t
he
Ta
ble
1.
Re
la
te
d
c
on
ce
pts
a
re
cat
egorised
int
o
co
nce
pt
cat
egories
to
s
pe
ci
fy
data
ind
i
cat
or
s
that
wil
l
be
us
e
d
in
data
m
ining
proces
s.
On
t
ology
co
nc
ept
cat
eg
or
ie
s
are
s
how
n
in
the f
irst
c
olu
m
n
o
f
Ta
ble
1. As
a
res
ult, f
eat
ur
es
of
the
data
a
nd
the
pr
e
dicte
d
outc
om
e
(Cla
ss)
of
the
cl
assifi
cat
i
on
process
is
s
how
n
Table
2.
In
Ta
ble
2,
the
first
seve
n
at
tribu
te
s
are in
put at
trib
utes a
nd the las
t on
e
, “CF
P T
r
ai
nin
g Nee
d”, i
s class
or
Cl
ass
ific
at
ion
resu
lt
.
Table
1
.
CF
P
On
t
ology C
onc
ept Cat
eg
or
ie
s
On
to
lo
g
y
Co
n
cept
Categ
o
ry
On
to
lo
g
y
Co
n
cept
Use case
Use case
Use case
Ap
p
licatio
n
I
n
tera
ctio
n
Diagra
m
Interaction
Interaction
Evo
lu
tio
n
T
y
p
e
Ad
d
E
v
o
lu
tio
n
T
y
p
e
Evo
lu
tio
n
T
y
p
e
Mod
if
y
E
v
o
lu
tio
n
T
y
p
e
Evo
lu
tio
n
T
y
p
e
Delete
Evo
lu
tio
n
T
y
p
e
Ap
p
licatio
n
Ap
p
licatio
n
Bu
siness Mod
u
le
Ap
p
licatio
n
Ap
p
licatio
n
Database Mod
u
le
Ap
p
licatio
n
Ap
p
licatio
n
Ser
v
ice
Ap
p
licatio
n
Ap
p
licatio
n
Ser
v
ice Bo
u
n
d
ary
Use case
Use case
Acto
r
Use case
Use case
Even
t
Inf
o
r
m
atio
n
Inf
o
r
m
atio
n
Asset
Interaction
Integ
ration
E
n
try I
n
terac
tio
n
Interaction
Integ
ration
E
x
it I
n
t
erac
tio
n
Interaction
User I
n
te
rf
ace
Ent
r
y
I
n
te
raction
Interaction
User I
n
te
rf
ace
Exit
I
n
terac
tio
n
Interaction
Databas
e
W
rite
I
n
t
erac
tio
n
Interaction
Databas
e
Read
Interaction
Sco
p
e
Project Sco
p
e
Sco
p
e
Ap
p
licatio
n
Ser
v
ice Scop
e
No
t App
licab
le
Prod
u
ctiv
ity
M
eas
u
re
m
en
t
Bu
sin
ess
L
o
g
ic
Use case Bu
sin
ess
Log
ic
Bu
sin
ess
L
o
g
ic
Interaction
Bu
sin
ess
L
o
g
ic
Table
2
.
CF
P
On
t
ology I
ndic
at
or
set a
nd T
he
ir P
os
sible
Va
lues
On
to
lo
g
y
Co
n
cept
Categ
o
ry
Valu
e
Use case
Yes, No,
Par
ti
al
Interaction
Yes, No,
Par
ti
al
Inf
o
r
m
atio
n
Yes, No,
Par
ti
al
Evo
lu
tio
n
T
y
p
e
Yes, No,
Par
ti
al
Bu
sin
ess
L
o
g
ic
Yes, No,
Par
ti
al
Ap
p
licatio
n
Yes,
No
,
Par
ti
al
Sco
p
e
Yes, No,
Par
ti
al
CFP Tr
ain
in
g
Nee
d
Yes, No
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:
5253
-
52
59
5256
2.5.
Data G
atheri
ng
and Pre
parati
on
First
se
ven
data
at
tribu
te
s
s
hown
in
Ta
ble
2
are
ob
ta
in
ed
f
r
om
req
ui
rem
ents
docum
ents
by
chec
king
wh
et
her
each
c
on
ce
pt
cat
egor
y
value
is
exist
ing
or
not.
If
their
value
s
are
par
ti
al
ly
existin
g
the
n
the
co
ncep
t
cat
egory
value
is
recorde
d
as
“Parti
al
”.
The
l
ast
at
tribu
te
is
captu
red
fro
m
aud
it
exam
i
nation
res
ults.
If
th
e
diff
e
re
nce
bet
ween
the
act
ua
l
m
easur
em
ent
done
by
re
quirem
ents
eng
in
eer
an
d
c
orrec
t
m
easur
em
ent
resu
l
t
identifie
d
by
a
ud
it
te
am
is
gr
eat
er
tha
n
5
%
then
t
his
case
is
recorde
d
as
a
le
arn
i
ng
op
port
un
it
y
by
r
ec
ordin
g
“C
FP
Traini
ng
Need
”
val
ue
a
s
“Yes”.
10
1
da
ta
po
ints
ha
ve
been
colle
ct
ed
and
res
ults
ha
ve
bee
n
rec
ord
ed
in
a
com
m
a
-
delim
it
ed
values
(.c
sv
)
file
.
Ne
xt,
this
file
has
be
en
co
nverted
into
the
at
trib
ut
e
-
relat
ion
file
form
a
t
(.
ar
ff)
w
hich
is
the
sta
nd
a
r
d
fi
le
form
at
us
ed
by
T
he
Waikat
o
E
nviro
nm
ent
f
or
K
nowled
ge
A
naly
sis
(
W
EKA)
data m
ining
s
oft
war
e
[
14
]
. T
he
conv
e
rsion t
ool u
sed
for .cs
v t
o
.a
rff is a
n o
nline
web to
ol
[
28
]
.
2.6.
Model
Tr
aini
ng
and Ev
alu
at
i
on
To
trai
n
a
nd
e
valuate
the
m
a
chine
le
a
rn
i
ng
m
od
el
s,
W
e
ka
Ex
per
im
enter
[
29
]
has
bee
n
us
ed
.
T
wo
exp
e
rim
ents
hav
e
bee
n
do
ne
in
Wek
a
.
In
the
first
exp
e
rim
ent,
“Data
set
s
first”
para
m
et
er
check
e
d
an
d
nu
m
ber
of r
e
pe
ti
ti
on
s h
as b
ee
n
set
as 1
00 in i
te
rati
on
contr
ol
p
aram
et
ers
pan
el
. Exper
im
ent ty
pe
is sel
ec
t
ed
as
Cros
s
validat
io
n.
N
um
ber
of
fo
l
ds
at
tribu
te
is
set
to
10
.
D
at
aset
has
bee
n
sel
ect
e
d
as
the
.ar
ff
file
w
hich
is
create
d
a
s
des
cribe
d
in
sect
ion
2.5
.
All
al
gorithm
s
wh
ic
h
hav
e
been
e
xp
la
ine
d
in
se
ct
ion
2.1
hav
e
been
sel
ect
ed
in
Algorithm
s
panel
.
Nex
t,
the
ex
pe
rim
ent
with
this
co
nf
i
gurati
on
has
bee
n
r
un
on
an
I
ntel
Co
re
i7
-
5600U
CPU
, 2
.6
G
H
z,
8 GB
RAM an
d 6
4
-
bi
t W
i
ndows
O
pe
rati
ng Syste
m
m
achine.
The
total
exec
ution
ti
m
e
wa
s
194
seco
nds
and
MLP
ha
d
the
slo
west
run
ning
tim
e.
Finall
y,
in
analy
se
ta
b
of
Wek
a
E
xp
e
rim
enter
us
e
r
inte
rf
ace,
al
l
al
gor
it
h
m
s
hav
e
be
en
sel
ect
ed
as
te
st
base
se
parat
el
y
and
te
st
is
perform
ed
fo
r
ea
ch
al
gorithm
.
Test
has
been
rep
eat
ed
f
or
t
hr
ee
ev
al
uatio
n
m
e
tric
s:
Accu
racy
(Num
ber
of
Correct
Cl
assifi
cat
ion
s
),
F
-
Me
asur
e
(F
M
)
,
an
d
Kappa
sta
ti
sti
c.
In
th
e
seco
nd
ex
pe
rim
ent,
exp
e
rim
ent
typ
e
was
set
t
o
Train/Test
Pe
r
centage
Sp
l
it
(
data
ra
ndom
ized
)
an
d
trai
n
per
ce
ntage
wa
s
set
t
o
66%.
All
ot
her
config
ur
at
io
n
rem
ai
ned
the
sa
m
e
as
in
Exp
e
rim
ent
1.
In
thi
s
case,
the
total
execu
ti
on
ti
m
e
wa
s
74 sec
onds
a
nd
MLP
had the
s
lowest
run
ning
tim
e, ag
ai
n.
3.
RESU
LT
S
A
ND AN
ALYSIS
Table
3
de
sig
nates
the
al
gor
it
h
m
per
f
orm
a
nces
i
n
te
rm
s
of
ac
cu
racy,
F
M,
an
d
Kappa
m
e
tric
s
for
bo
t
h
ex
per
im
e
nts.
Me
tric
values
are
show
n
as
aver
a
ges
with
sta
nd
a
r
d
de
viati
on
s
.
All
al
go
rithm
s
see
m
m
eaningfu
ll
y
bette
r
than
Ze
r
oR
baseli
ne
pe
rfor
m
ance.
Suppo
rt
Vecto
r
Ma
chine
s
a
nd
On
eR
al
go
rith
m
s
hav
e
the
la
rg
est
a
ve
rag
e
acc
ur
acy
values
.
H
owev
er
eval
uating
t
he
al
gorithm
s
so
le
ly
based
on
the
a
ver
a
ge
values
and
sta
ndar
d
de
viati
on
s
w
ouldn’t
be
s
uffici
ent
since
di
ff
e
ren
ces
bet
wee
n
resu
lt
s
m
ight
no
t
be
sta
ti
sti
cal
ly
sign
ific
a
nt.
T
h
eref
or
e
,
in
W
eka
E
xperim
e
nter
A
naly
se
interface
,
Sig
nificance
has
be
en
set
t
o
0.0
5,
al
l
al
gorithm
s
have
bee
n
sel
ect
e
d
as
te
st
base
s
epar
at
el
y
an
d
t
est
s
ha
ve
been
perf
or
m
ed.
St
at
ist
ic
ally
sign
ific
ant
diff
e
re
nces
ha
ve
bee
n
rec
ord
ed
duri
ng
te
sts.
Stat
ist
ic
al
sign
ific
ance
is
de
no
te
d
by
“
v”
a
nd
“
*”
sym
bo
ls
in
the
Wek
a
i
nter
fac
e.
Form
er
m
ea
ns
sta
ti
sti
cal
ly
sign
ific
a
nt
be
tt
er
perform
ance
wh
il
e
la
tt
er
i
m
plies
sta
ti
st
ic
al
l
y
sign
ific
a
nt wo
r
se p
e
rfor
m
ance [
29
]
.
Stat
ist
ic
ally
si
gn
i
ficant
s
upe
rior
it
ie
s
betwe
en
al
go
rithm
s
are
s
how
n
i
n
Table
4.
For
i
ns
ta
nce
,
in
Ex
per
im
ent
1,
Naïve
Ba
ye
s
perform
s
bette
r
tha
n
IB
k
a
nd
Ze
ro
R
w
hen
Acc
ur
acy
a
nd
Ka
pp
a
m
et
rics
are
con
ce
r
ned.
Be
st
perform
ing
al
gorithm
s
have
bee
n
determ
i
ned
by
c
om
par
in
g
the
num
ber
of
al
l
sta
ti
sti
cal
l
y
sign
ific
a
nt
su
pe
rior
it
ie
s.
We
s
how
this
num
ber
as
“
Nu
m
ber
of
Wins
”
in
Ta
ble
4.
As
a
res
ult;
REPTree,
On
e
R
and
S
VM
with
SMO
al
gorith
m
s
hav
e
the
m
axim
u
m
“Nu
m
ber
of
W
i
ns
”
va
lues
a
nd
are
de
te
rm
ined
to
be
the
top
t
hr
ee
al
gor
it
h
m
s
perform
i
ng
best
i
n
CF
P
dataset
of
t
his
stud
y.
A
s
fa
r
as
we
know,
this
is
the
first
stud
y
that
us
e
data
m
ining
on
re
quirem
ents
and
CFP
m
easur
em
ent
data.
Th
eref
or
e
,
we
c
ould
n’
t
com
pare
the
perform
ance
of
ou
r
stu
dy
with
oth
e
r
si
m
il
ar
research
directl
y.
Howev
e
r
if
we
ben
c
hm
ark
with
so
m
e
edu
cat
io
nal
da
ta
m
ining
stu
di
es
in
ge
ne
ral,
we
see
our
top
pe
rfor
m
ing
m
od
el
s
are
ve
ry
good
i
n
te
rm
s
of
accuracy
[
30
–
33
]
.
Table
3
.
Algori
thm
Per
form
ances for C
FP Da
ta
set
Alg
o
rith
m
Exp
eri
m
en
t 1: C
ros
s
-
v
alid
atio
n
Exp
eri
m
en
t 2: 6
6
% Sp
lit T
est
Accurac
y
(
%)
FM
Kap
p
a
Accurac
y
FM
Kap
p
a
Ran
d
o
m
For
est (R
F)
7
8
.79
(
1
1
.72
)
0
.82
(
0
.11
)
0
.56
(
0
.25
)
7
9
.20
(
5
.54
)
0
.83
(
0
.04
)
0
.56
(
0
.12
)
Naiv
e Bay
es (
NB)
8
2
.97
(
1
1
.13
)
0
.86
(
0
.09
)
0
.63
(
0
.24
)
8
1
.29
(
5
.46
)
0
.85
(0.0
4
)
0
.60
(
0
.12
)
REPTr
ee
8
4
.02
(
1
1
.25
)
0
.86
(
0
.11
)
0
.67
(
0
.23
)
8
4
.27
(
4
.81
)
0
.86
(
0
.04
)
0
.68
(
0
.10
)
J4
8
(
W
ek
a C 4.5
I
m
p
l
e
m
en
tatio
n
)
8
2
.59
(
1
1
.39
)
0
.84
(
0
.11
)
0
.65
(
0
.23
)
8
3
.38
(
4
.58
)
0
.85
(
0
.04
)
0
.66
(
0
.09
)
Log
istic Mod
el T
r
ees (LMT
)
8
3
.84
(
1
1
.34
)
0
.86
(
0
.11
)
0
.67
(
0
.23
)
8
3
.52
(
5
.21
)
0
.86
(
0
.05
)
0
.66
(
0
.11
)
Multilaye
r
Pe
rcept
ron
(
ML
P)
7
6
.74
(
1
2
.50
)
0
.80
(
0
.12
)
0
.52
(
0
.26
)
7
5
.94
(
6
.55
)
0
.80
(
0
.05
)
0
.49
(
0
.15
)
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
Using
Da
t
a
Mi
ning to
Ide
ntif
y COSMIC
Fu
nc
ti
on
P
oin
t
Me
as
ure
me
nt C
ompete
nce
(
Sel
ami B
ag
riy
an
ik
)
5257
Table
3
.
Algori
thm
Per
form
ances for C
FP Da
ta
set
Alg
o
rith
m
Exp
eri
m
en
t 1: C
ros
s
-
v
alid
atio
n
Exp
eri
m
en
t 2: 6
6
% Sp
lit T
est
Accurac
y
(
%)
FM
Kap
p
a
Accurac
y
FM
Kap
p
a
Su
p
p
o
rt
Vector M
achi
n
es with
SM
O
8
4
.16
(
1
1
.25
)
0
.86
(
0
.11
)
0
.68
(
0
.23
)
8
3
.70
(
4
.91
)
0
.86
(0.0
4
)
0
.67
(
0
.10
)
K
-
n
eare
st
Neigh
b
o
u
r
Clas
sif
ier
(
IBK)
7
5
.61
(
1
1
.56
)
0
.80
(
0
.10
)
0
.48
(
0
.25
)
7
5
.38
(
5
.16
)
0
.81
(
0
.04
)
0
.47
(
0
.12
)
On
eR
8
4
.16
(
1
1
.25
)
0
.86
(
0
.11
)
0
.68
(
0
.23
)
8
4
.27
(
4
.81
)
0
.86
(
0
.04
)
0
.68
(
0
.10
)
Zer
o
R
5
9
.45
(
0
1
.64
)
0
.75
(
0
.01
)
0
.00
(0.0
0
)
5
9
.37
(
0
.63
)
0
.75
(
0
.04
)
0
.00
(
0
.00
)
Table
4
.
Stat
ist
ic
al
ly
Sign
ific
ant S
up
e
rio
riti
es of A
l
gorithm
s f
or CFP
D
at
as
et
Alg
o
rith
m
Exp
eri
m
en
t 1: C
ros
s
-
v
alid
atio
n
Exp
eri
m
en
t 2: 6
6
% Sp
lit
Nu
m
b
e
r
o
f
W
in
s
Accurac
y
FM
Kap
p
a
Accurac
y
FM
Kap
p
a
Ran
d
o
m
Fo
r
est (R
F)
Zer
o
R
Zer
o
R
Zer
o
R
Zer
o
R
Zer
o
R
Zer
o
R
6
Naiv
e Bay
es (
NB)
IBk
,
Ze
roR
ML
P,
I
Bk
,
Zer
o
R
IBk
,
Zer
o
R
Zer
o
R
Zer
o
R
Zer
o
R
10
REPTr
ee
RF, M
LP,
IBk
,
Ze
roR
ML
P,
I
Bk
,
Zer
o
R
RF,
ML
P,
IBk
,
Zer
o
R
IBk
,
Zer
o
R
IBk
,
Zer
o
R
IBk
,
Zer
o
R
17
J4
8
(
W
ek
a C 4.5
I
m
p
le
m
en
tatio
n
)
IBk
,
Ze
roR
Zer
o
R
IBk
,
Zer
o
R
IBk
,
Zer
o
R
Zer
o
R
IBk
,
Zer
o
R
10
Log
istic Mod
el T
r
ees (LMT
)
ML
P,
I
Bk
,
Zer
o
R
ML
P,
I
Bk
,
Zer
o
R
ML
P,
IBk
,
Zer
o
R
IBk
,
Zer
o
R
Zer
o
R
IBk
,
Zer
o
R
14
Multilaye
r
Pe
rcept
ron
(
ML
P)
Zer
o
R
Zer
o
R
Zer
o
R
Zer
o
R
Zer
o
R
Zer
o
R
6
Su
p
p
o
rt
Vector M
achi
n
es with
SMO
RF, M
LP,
IBk
,
Ze
roR
ML
P,
I
Bk
,
Zer
o
R
RF,
ML
P,
IBk
,
Zer
o
R
IBk
,
Zer
o
R
Zer
o
R
IBk
,
Zer
o
R
16
K
-
n
eare
st
Neigh
b
o
u
r
Clas
sif
ier
(I
Bk
)
Zer
o
R
Zer
o
R
Zer
o
R
Zer
o
R
Zer
o
R
Zer
o
R
6
On
eR
RF, M
LP,
IBk
,
Ze
roR
ML
P,
I
Bk
,
Zer
o
R
RF,
ML
P,
IBk
,
Zer
o
R
IBk
,
Zer
o
R
IBk
,
Zer
o
R
IBk
,
Zer
o
R
17
Zer
o
R
No
n
e
No
n
e
No
n
e
No
n
e
No
n
e
No
n
e
0
4.
CONCL
US
I
O
N
In
this
st
ud
y,
w
e
c
onduct
ed
an
e
du
c
at
ion
al
d
at
a
m
ining
r
e
search
. I
n
the s
cop
e
of
t
his u
s
e
case,
a
CFP
dataset
w
hich
was
c
ollec
te
d
from
a
la
rg
e
t
el
ecom
m
un
ic
ation
s
ser
vices
a
nd
te
ch
no
l
og
y
com
pan
y
ha
s
bee
n
analy
sed
us
i
ng
10
m
achine
l
earn
i
ng
al
gorit
hm
s
to
identif
y
CFP
le
arn
in
g
nee
d
of
Re
quirem
ents
Engineers.
Af
te
r
t
wo
ex
pe
rim
ents,
m
od
el
perform
ances
are
eval
uated
a
nd
t
op
pe
rfor
m
er
al
gorithm
s
ha
ve
be
en
i
den
ti
fied.
REPT
ree,
On
e
R
an
d
SV
M
with
SMO
al
gorithm
s
perf
orm
ed
bette
r
tha
n
oth
e
r
al
gorithm
s
in
a
sta
ti
sti
cal
ly
sign
ific
a
nt
m
a
nn
e
r.
T
op
pe
r
form
ing
m
od
e
l
pr
edict
io
n
pe
rfor
m
ances
are
suffici
ent
to
be
us
e
d
in
the
pro
du
ct
io
n
e
nv
iro
nm
ent in th
e
co
m
pan
y.
In the
fu
t
ur
e,
foll
owin
g researc
h
i
s p
la
nn
e
d:
Do
m
inati
ng
in
dicat
or
s
i
n
CFP
m
easur
em
ent
will
be
identifie
d
by
us
i
ng
featur
e
sel
ect
ion
al
go
rithm
s.
So
m
e n
ew
i
nd
i
cat
or
s
from
the r
e
qu
irem
ents
arti
facts m
ay
a
rise in t
his
pro
cess.
Data
points
num
ber
will
be
increase
d
a
nd
the
stu
dy
wil
l
be
re
plica
te
d
by
al
so
a
dding
so
m
e
oth
er
al
gorithm
s su
ch
as
Ada
ptive
Neur
o
F
uzzy I
nf
e
ren
ce
Syste
m
(
AN
FIS).
REFERE
NCE
S
[1]
Nus
ei
beh
B,
E
aste
rbrook
S.
“
Req
uire
m
ent
s e
ng
in
ee
ring
:
a
roa
dm
a
p
”
.
In
:
Proceedi
ngs of
th
e conf
er
enc
e
on
The
fut
ure
o
f
Sof
tware
eng
ineering
-
I
CSE
’00
.
2000.
p.
35
–
46
.
[2]
Naik
P,
Na
y
a
k
S.
“
Insights
on
Resea
rch
Techn
ique
s
towar
ds
C
ost
Esti
m
at
ion
i
n
Software
Desi
gn.
”
In
t
J
Elec
t
r
Comput
Eng
(
IJ
ECE
)
.
2017;7(5)
:2883
–
94.
[3]
Bagriy
an
ik
S,
Kara
hoc
a
A.
Aut
om
at
ed
COS
MI
C
Functi
on
Point
m
ea
surem
e
nt
using
a
req
uirem
ent
s
engi
nee
r
i
ng
ontol
og
y
.
Inf
Sof
tw
Techno
l
.
201
6;72:
189
–
203.
[4]
Faja
r
AN
,
Shofi
IM.
R
educed
Software
Com
ple
xity
for
E
-
G
over
nm
ent
App
li
c
at
ions
wi
th
ZE
F
Fram
ework.
TEL
KOMNIKA
(Tele
kommunic
a
t
ion
Computing
,
El
e
ct
ronics
and
Control)
.
2017;1
5(1):415
–
20.
[5]
Trude
l
S,
Tur
c
ott
e
A
.
“
Com
bini
ng
Qualita
ti
ve
and
Quan
ti
t
at
i
ve
Software
Proce
ss
Evaluatio
n :
A
Propos
ed
Approac
h
”
.
Col
l
Ec
on
Ana
l
Ann
.
2017;43:
135
–
54.
[6]
S
y
m
ons
C,
Abr
an
A,
Ebe
rt
C
,
Vogele
z
ang
F.
“
Mea
surem
ent
of
software
size
:
adva
nc
es
m
ade
b
y
th
e
COS
MIC
comm
unity
”
.
In
:
2016
Jo
int
C
onfe
renc
e
of
th
e
Inte
rnat
ional
Workshop
on
Soft
ware
Me
asu
rement
and
th
e
Inte
rnational
Co
nfe
renc
e
on
Softw
are
Proce
ss
an
d
Product Me
as
urement
.
IEEE
;
2016.
p
.
75
–
86
.
[7]
Salmanoğlu
M,
Öztür
k
K,
Bağr
ı
y
an
ık
S,
Ungan
E,
Dem
irörs
O.
“
Bene
fit
s
and
c
hal
l
enge
s
of
m
e
asuring
software
size
:
ea
rl
y
r
esults
in
a
la
rge
o
rg
ani
z
at
ion
”
.
In
:
2
5th
Inte
rnationa
l
Workshop
on
Soft
ware
Me
asu
rement
and
10th
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:
5253
-
52
59
5258
Inte
rnational
Co
nfe
renc
e
on
Softw
are
Proce
ss
an
d
Product Me
as
urement
.
20
15.
[8]
Ozka
n
B,
Dem
i
rors
O.
“
On
th
e
Seven
Misco
nce
pt
ions
about
Functi
on
al
Si
z
e
Mea
sur
ement
”
.
In
:
2016
Jo
i
nt
Confe
renc
e
of
th
e
Int
ernati
onal
Workshop
on
Soft
ware
Me
asur
e
ment
and
th
e
In
t
ernati
onal
Con
fer
enc
e
on
Soft
wa
re
Proce
ss
and
Product
M
easureme
nt.
I
EEE;
2016
.
p.
45
–
52
.
[9]
Bagriy
an
ik
S,
K
ara
hoc
a
A.
“
Big
dat
a
in
software
engi
nee
r
ing:
A
s
y
stematic
li
t
erature
rev
i
ew
”
.
Gl
ob
J
Inf
Technol
.
2016;6(1):
107
–
1
6.
[10]
Sahoo
P,
Mohant
y
JR
.
“
Ea
rl
y
Te
st
Eff
or
t
Pred
ic
ti
on
using
UM
L
Diagr
ams
’
.
Indone
s
J
E
le
c
t
r
Eng
Comput
Sc
i
(
IJE
EC
S)
.
2017;
5(1):220
–
8.
[11]
Kim
DK
.
“
Finding
Bad
Cod
e
Sm
el
ls wi
th
Neur
al
Network
Models
”
.
In
t J Elect
r C
omput
Eng
.
201
7;7(6):
3613
–
21.
[12]
H.
W
it
t
en
I
,
Fra
nk
E, A.
Hall M.
“
Data
m
ini
ng
:
P
rac
t
ic
a
l
m
ac
h
ine l
e
arn
ing
tool
s
a
nd
technique
s
”
.
3rd
ed. 2011
.
[13]
Han
J,
Kam
ber
Mi
che
li
n
e.
“
Dat
a
Mining
Conc
e
pts a
nd
Techni
q
ues
”
.
2nd
ed
.
20
06.
[14]
Hall
M,
Frank
E,
Holm
es
G,
Pfahringe
r
B,
R
eut
emann
P,
W
it
te
n
IH.
“
The
W
EKA
dat
a
m
i
ning
software
:
an
updat
e
”
.
ACM
SI
GKD
D E
xpl
or N
ewsl
.
2009
;11(1)
:10
–
8.
[15]
Brei
m
an
L
.
R
an
dom
fore
sts.
Ma
ch
L
ea
rn
.
2001
;4
5(1):5
–
32.
[16]
H.
John
G,
La
ngley
P.
“
Estim
at
ing
cont
inuo
us
distri
buti
ons
in
Ba
y
esi
an
class
ifi
ers
”
.
In:
P
roce
edi
ngs
of
t
he
El
e
ve
nth
con
fe
re
nce
on
Unce
rta
i
nty
in
arti
fi
c
ial i
nte
lligen
ce
.
Mor
gan
Kaufm
ann
P
ubli
shers Inc.;
1
995.
p
.
338
–
45
.
[17]
Zha
o
Y,
Zh
ang
Y.
“
Com
par
ison
of
de
ci
sion
tree
m
et
hods
for
findi
ng
a
ct
i
ve
obj
ec
ts
”
.
Ad
v
Sp
Re
s
.
200
8;
41(12):1955
–
9.
[18]
Quinla
n
JR.
C4.
5:
progra
m
s for m
ac
hine
learni
n
g.
2014
.
[19]
La
ndwehr
N
,
Ha
ll
M,
Frank
E.
“
Logi
stic m
odel t
ree
s
”
.
Mac
h
Lea
rn
.
2005;59(1
–
2
):161
–
205.
[20]
Gardne
r
M.
,
Do
rli
ng
S.
Ar
ti
fi
cial
n
eur
al
net
wo
rks
(the
m
ult
i
lay
er
per
ce
pt
ron)
—
a
rev
ie
w
of
a
ppli
c
at
ions
in
th
e
at
m
ospheric
scie
nc
es.
At
m
os
Envi
ron
[Inte
rne
t]
.
1998;32(14
–
15):2627
–
36.
Avail
able
from
:
htt
p://ww
w.sci
e
nce
dir
ec
t
.
com/sc
ie
nc
e/
ar
ti
c
le/pii/
S1352231097004470
[21]
Plat
t
JC.
Sequen
ti
al
Min
imal
Optimiza
t
ion:
A
Fast
Algorit
hm
for
Tra
ini
ng
Suppo
rt
Vec
tor
Mac
h
i
nes.
Adv
ker
nel
m
et
hods
[Inte
rne
t]
.
1998;185
–
208.
Avail
ab
le
from
:
htt
p://ww
w.bra
d
bloc
k.
com/Sequ
ent
i
al
_Minimal_
Optimiza
ti
o
n_A_F
ast_Algori
th
m
_for_T
rai
ning
_Support_Vect
o
r_
Mac
hine.pdf
[22]
Aha
DW
,
Kibler
D,
Alb
ert
MK
.
I
nstanc
e
-
B
ase
d
L
ea
rning
Algorit
h
m
s.
Mac
h
L
ea
rn
.
1991;6(1):
37
–
6
6.
[23]
Holte
RC.
V
ery
Simple
Classifi
cat
ion
Ru
le
s
P
e
rform
We
ll
on
Most
Comm
only
Us
e
d
Datasets
.
Mac
h
Learn.
1993;11(1):
63
–
9
0.
[24]
Bundschuh
M,
Dekke
rs
C.
“
T
he
IT
m
ea
sure
m
ent
compendi
um
:
esti
m
at
ing
and
b
enc
hm
ar
king
succ
ess
with
func
ti
on
al
si
ze
m
ea
surem
ent
”
.
Springer
Sci
ence
&
Busin
ess Medi
a; 2008.
[25]
Jones
C.
“
Appli
ed
Software
Me
asure
m
ent
Glob
a
l
Ana
l
y
s
is
of
P
roduc
ti
vi
t
y
and
Quali
t
y
”
.
M
cGra
w
-
Hill
Educ
a
tion
Group;
2008.
[26]
Consortium
CS
MI.
“
The
COS
MIC
Functi
onal
Size
Mea
sure
m
ent
Method
Version
4.
0.
1
Me
asure
m
ent
Man
ual
“
[Inte
rne
t]
.
[
ci
t
e
d
2018
Jan
17]
.
Avail
ab
le
f
rom
:
htt
ps://
cosm
ic
-
sizi
ng.
org
/publ
i
ca
t
ions/m
ea
surem
ent
-
m
anua
l
-
401/
[27]
Bağrıy
an
ık
S,
Kara
hoca
A,
Er
so
y
E
.
”
Selection
of
a
fun
ctio
nal
siz
ing
m
et
h
odolog
y
:
A
t
el
e
comm
unic
at
ions
compan
y
c
ase
st
ud
y
”
.
Glob
J
Te
chnol
.
2015;7(7)
:98
–
108.
[28]
Il
y
a.
ku
zovki
n@
gm
ai
l.
com.
Onli
ne
co
nv
ert
er
fro
m
.
csv
to
W
EK
A
.
arf
f
[
Inte
rn
et]
.
[ci
te
d
2018
J
an
20]
.
Availabl
e
from
:
htt
p:
//
iku
z
.
eu/csv2arff/
[29]
Scuse
D,
R
eut
e
m
ann
P.
“
W
EKA E
xper
iment
er Tut
ori
al
for
Vers
ion
3
-
5
-
8
”
.
2008
.
[30]
As
if
R,
Merc
ero
n
A,
Ali
S
A,
Haide
r
NG
.
Anal
y
z
ing
under
gra
dua
te
student
s
â
€
TM
per
form
anc
e
using
educ
ationa
l
dat
a
m
ini
n
g.
Com
put
Educ
[Inte
rne
t]
.
2017;113:177
–
94.
Available
fr
om
:
htt
p://dx.doi.org/10.1016/j.c
om
pe
du.
2017.
05
.
007
[31]
Kaba
kchieva
D.
“
Predic
ti
ng
Stu
dent
Perform
ance
b
y
Us
ing
Data
Mining
Method
s
for
Cla
ss
ifi
cati
on
”
.
Cyb
ern
In
f
Technol
.
2013;1
3(1):61
–
72.
[32]
Ahm
ad
F,
Ism
ai
l
NH
,
Azi
z
AA
.
“
The
Predi
ct
ion
of
Student
s
’
Aca
demic
Perfor
m
anc
e
Us
ing
Cl
assific
a
ti
on
Da
ta
Mining
T
ec
hniq
ues
”
.
Appl Mat
h
Sci
.
2015;9(129)
:6415
–
26.
[33]
Kaur
P,
Singh
M,
Singh
G.
“
Cla
ss
ifi
cation
and
pre
diction
base
d
dat
a
m
ini
ng
a
lg
orit
hm
s
to
pre
di
ct
slow
le
a
rne
rs
in
educ
a
ti
on
sec
to
r
”
.
In:
P3rd
I
nte
rnational
Co
nfe
renc
e
on
Rece
nt
Tr
ends
in
Computing
2015(
ICRTC
-
2015
)
[Inte
rne
t]
.
El
sev
i
er
Mass
on
SA
S; 2015.
p
.
500
–
8
.
Avail
ab
le
f
rom
: http:
/
/dx.
doi
.
org/
10.
1016/j.proc
s.
2015.
07.
372
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
Using
Da
t
a
Mi
ning to
Ide
ntif
y COSMIC
Fu
nc
ti
on
P
oin
t
Me
as
ure
me
nt C
ompete
nce
(
Sel
ami B
ag
riy
an
ik
)
5259
BIOGR
AP
H
I
ES
OF
A
UTH
ORS
Dr.
Sela
m
i
B
ag
ri
y
ani
k
ho
lds
a
PhD
in
Software
Engi
ne
eri
ng
.
He
is
int
e
rest
ed
in
softwar
e
deve
lopment
,
re
quire
m
ent
s
enginee
ring
,
softwar
e
m
ea
surem
ent
,
adva
nc
ed
learni
ng
te
chno
logi
es
,
dat
a
m
ini
ng
an
d
big
data.
He
al
so
works
as
the
Digi
ta
l
Learni
ng
and
Bus
ine
ss
Soluti
ons
Te
chno
log
y
Ma
nage
r in
Turkcel
l.
Dr.
Adem
Kara
h
oca
ho
lds a PhD
in
Software
Eng
i
nee
ring
.
He
is
in
te
rest
ed
in
hum
a
n
–
computer
int
er
ac
t
ion, web based
edu
cation s
y
stems
,
da
ta m
ini
ng,
b
ig
d
at
a
,
a
nd
m
ana
gement
informati
on
s
y
stems
.
He
has
publi
shed
art
i
cle
s a
t
pr
esti
gious
j
ourna
ls a
bou
t
us
e
and
da
ta m
ini
n
g
applications
of
business i
nfor
m
at
ion
s
y
s
te
m
s i
n
he
a
lt
h
,
tour
ism
,
and
edu
cation.
Evaluation Warning : The document was created with Spire.PDF for Python.