IAES Internatio
n
a
l
Jo
u
rna
l
o
f
Artificia
l
Inte
llig
ence (IJ-AI)
Vol
.
1
0,
N
o.
3
,
Sept
em
ber
20
21
,
pp
.
76
4~
77
0
I
S
SN
: 225
2-8
9
3
8
, D
O
I
:
10.115
91
/ij
ai.v
10.i3
.p
p76
4-7
70
7
64
Jo
urn
a
l
h
o
me
pa
ge
: h
ttp
://ijai.ia
esco
re.com
CLG clu
s
tering f
or dropout predi
c
tion usi
ng log-d
at
a clust
erin
g
method
Ag
un
g T
r
i
a
yu
di
1
, W
ah
yu
O
ktri Wid
yart
o
2
, Lia K
amelia
3
, Iks
al
4
, S
umiati
5
1
Department of I
n
form
atic and C
o
mmunication
Technolog
y
,
Univ
e
rsi
tas Nasion
al, Indonesia
2
Department of I
ndustrial Engin
e
er
ing, Universitas Serang Ra
y
a
,
I
ndonesia
3
Department of Electrical Eng
in
eering
, UIN Sun
an Gunung Djati, I
ndonesia
4
Departm
e
nt
o
f
Ele
c
tri
cal
Eng
in
eering
, Univ
e
rs
it
as
F
ale
t
ehan
,
In
d
onesia
5
Department of I
n
form
atic, Univ
ersitas Se
r
a
ng R
a
y
a
, Indonesia
Article Info
A
B
STRAC
T
Article histo
r
y:
Received Ma
r 2, 2021
Rev
i
sed
May 5, 202
1
Accepted May 22, 2021
Im
plem
entation
of
d
at
a
m
i
ning
,
m
achine
learning,
a
nd
s
tatistica
l
d
a
ta
from
educational departme
nt commonly
known as e
ducation
a
l data minin
g. Most of
school
s
y
s
tems
r
equire
a
t
eacher
t
o
teach
a
n
u
m
ber
of
s
tudents
at
one
t
im
e.
Exam
a
re
r
egu
l
arly
b
eing
use
as
a
m
ethod
to
m
easure
student’s
a
chiev
e
m
e
nt,
which
is
d
iffi
cul
t
t
o
underst
a
nd
because
e
xam
i
n
a
tion
cannot
b
e
done
easily
.
The
other
h
a
nd,
p
rogramming
cla
sse
s
ma
ke
s
so
urc
e
c
o
de
e
diting
a
nd
UNIX
com
m
a
nds
a
ble
to
eas
i
l
y
d
e
t
e
c
t
and
s
t
ore
autom
a
ti
cal
l
y
a
s
log
-
d
ata
.
H
enc
e
,
rather
t
ha
t
es
tim
ating
the
perfor
m
ance
of
t
hos
e
s
t
udent
b
as
ed
o
n
this
l
og-d
a
ta
,
this
s
tud
y
b
eing
more
f
ocused
on
detecting
th
em
w
ho
experien
ced
a
d
i
f
f
i
c
u
l
t
y
or
unable
to
t
ak
e
programming
cla
sses.
We
pro
pose
CLG
clusterin
g
methods
that
c
an
p
redict
a
r
isk
of
b
eing
dropped
out
from
school
u
sing
cluster
data
f
or
outlier
de
te
ction
.
K
eyw
ords
:
Dro
pou
t p
r
ed
ictio
n
Edu
catio
n
a
l d
a
ta
m
in
in
g
k-m
eans
Ou
tlier d
e
tectio
n
UN
IX
com
m
a
nds
This is an open
access article un
der the
CC BY
-S
A
l
i
cens
e
.
Co
rresp
ond
i
ng
Autho
r
:
A
gun
g Tr
iayudi
Depa
rt
m
e
nt
of
In
fo
rm
at
i
c
and C
o
m
m
uni
cat
ion
Tec
h
n
o
l
o
gy
Uni
v
ersitas Na
sional
Jl. Saw
o Ma
nila, RT.
14/R
W
.
3
, Ps.
Minggu,
K
ec.
P
s. Mi
n
ggu
Kota Jaka
r
ta Selatan
, Ja
k
arta, Indonesia
Em
a
il: ag
u
n
g
t
riayu
d
i
@civ
itas.un
as.ac.id
1.
INTRODUCTION
Ed
ucat
i
onal
d
a
t
a
m
i
n
i
ng
i
s
a
d
at
a
m
i
ni
ng
im
pl
em
ent
e
d
t
echni
que
i
n
a
n
ef
fo
rt
t
o
d
e
vel
o
p
dat
a
expl
orat
i
o
ns
o
f
vari
ous
e
d
u
ca
t
i
onal
i
n
f
o
rm
ati
on
sy
st
em
s
recom
m
e
nd
t
h
e
si
ngl
e
l
i
n
kage
(
SL
G)
d
i
ssi
m
ilari
t
y
i
n
crem
ent
di
st
ri
b
u
t
i
on
m
e
t
hod
,
gl
o
b
al
c
u
m
ul
at
i
v
e
scor
e
st
andar
d
(
SLG),
and
ave
r
age
linkage
(
AL
G)
d
i
ssimilarity
i
n
c
rem
e
n
t
d
istrib
u
tio
n,
g
lob
a
l
cu
m
u
lativ
e
sco
r
e
s
t
a
ndar
d
(
A
L
G)
w
hi
c
h
u
se
d
t
o
a
nal
y
ze
st
ude
nt
learn
i
ng
o
n
lin
e
in
teractio
n
d
a
t
a
.
Th
e
end
resu
lt
is
a
g
rou
p
i
ng
m
o
d
e
l
o
f
b
ehav
ior
p
a
ttern
s
an
d
i
n
terp
erson
a
lity
p
a
ttern
s
o
f
s
tud
e
n
t
s
[1
],
[
2
]
.
Th
e
i
n
itial
p
r
ocess
starts
f
rom
co
llect
in
g
data,
b
e
fore
it
i
s
c
on
tinu
e
d
with
d
ata
tran
sform
a
t
i
o
n
,
a
nd
i
s
term
in
ated
b
y
d
a
ta
a
n
a
lysis
[3
].
E
du
cati
onal
dat
a
m
i
n
i
ng
i
s
i
m
p
l
e
m
e
nt
ed
i
n
o
r
der
t
o
ach
iev
e
t
h
e
goal
o
f
f
u
l
filling
th
e
u
s
efu
l
i
n
f
orm
a
t
i
o
n
n
e
ed
s
o
f
large
am
ounts
o
f
electronic
data
r
ecorde
d
i
n
the
educat
or
sy
s
tem
[4]
-[6]
.
R
e
ferri
ng
t
o
t
h
e
m
ai
n
t
opi
c
of
t
he
d
i
s
cussi
o
n
t
h
i
s
t
im
e,
t
he
a
c
t
u
al
e
ducat
i
o
n
sy
st
em
.
In
m
ost
sch
ool
s
,
m
o
st
t
eachers
will
teach
a
num
ber
of
s
tude
nts
in
a
c
lass
a
t
a
t
i
m
e
[7
],
[
8
]
.
Of
c
ou
rse,
t
h
i
s
will
co
m
p
lic
ate
th
e
teacher
i
n
displaying
the
m
a
teri
al
i
n
detail
on
eac
h
of
its
s
tu
de
nt
s
[9]
.
O
n
t
h
e
ot
he
r
ha
nd
,
t
eac
hers
n
eed
t
o
k
now
t
h
e
s
k
ill
lev
e
l
o
f
s
tuden
t
s
to
g
u
i
d
e
t
h
e
m
an
d
prov
id
e
a
h
i
g
h
qu
ality
e
d
u
cation
.
T
h
e
refore,
t
e
stin
g
peri
odi
cal
l
y
n
eeds
t
o
b
e
d
o
n
e
t
o
s
ee
i
f
t
he
s
t
ude
nt
s
ha
ve
t
he
s
k
ills
h
e
n
e
ed
s
[1
0
]
,
[11
]
.
Howev
e
r,
it
is
q
u
ite
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J Artif In
t
el
l
I
S
SN
: 225
2-8
9
3
8
CLG
cl
u
s
tering
for
d
r
opo
u
t
pred
ictio
n u
s
i
n
g lo
g-d
a
t
a
clu
s
t
e
rin
g
m
e
t
h
od
(
A
gu
ng
Tria
yu
di
)
76
5
diffic
u
lt
to
u
nderstand
t
h
e
st
ud
ents
'
a
bility
from
each
s
ubj
e
c
t,
b
ecau
s
e
th
e
ex
am
s
h
a
v
e
d
ep
leted
a
lo
t
o
f
t
i
m
e
as wel
l
as i
ncri
m
i
nat
e
d cl
ass
m
e
m
b
ers [
12]
.
The
ot
he
r
han
d
, i
t
is co
m
m
o
n
to
track
th
e
activ
ities an
d b
e
hav
i
ors
in
pro
grammin
g
class, as sou
r
ce
cod
e
ed
itin
g and
UNIX comman
d
s
save
t
hem
as l
og-
dat
a
[
13]
,
[
14]
.
W
e
f
o
u
n
d
s
e
v
e
r
a
l
s
t
u
d
i
e
s
r
e
l
a
t
e
d
t
o
l
o
g
-
d
a
t
a
u
s
a
g
e
s
.
L
i
k
e
s
,
t
h
e
re
i
s
a
research
t
h
a
t
pred
icts
s
tu
den
t
sk
ills
b
ased
o
n
th
e
lo
g
-
d
a
ta
[
1
5
]
,
[16
]
.
Altho
ugh
t
h
e
l
ev
el
o
f
acc
uracy
o
f
t
h
e
m
e
t
hod
i
n
t
hi
s
pre
d
i
c
t
i
on
i
s
not
so
h
igh
,
a
n
d
still
a
lack
o
f
co
n
s
i
d
eration
wh
eth
e
r
th
e
ev
alu
a
ti
o
n
is
d
one
b
a
s
ed
o
n
th
e
p
r
e
s
cr
ib
ed
a
sp
e
c
t
s
.
Mo
reo
v
e
r,
i
n
so
m
e
r
esearch
a
p
p
a
ren
tly
f
o
und
d
ifficu
lties
in
e
v
a
l
u
atin
g
t
h
e
stu
d
e
n
t
s’
a
cqu
i
red
sk
ills
b
ased
on
log-data.
T
h
e
r
efore,
t
his
res
earch
i
s
aim
e
d
at
obtaining
d
ata
o
n
stud
ents
w
ho
canno
t
k
eep
i
n
g
up
w
ith
t
h
e
pr
o
g
ram
m
i
ng
cl
ass,
r
at
her
t
h
an
t
o
est
i
m
a
t
e
t
he
achi
e
vem
e
nt
o
f
s
t
u
de
nt
s
base
d
o
n
d
at
a-
l
ogs
.
W
e
s
peci
fi
cal
l
y
pr
o
pose
a
m
e
tho
d
t
h
at
c
a
n
b
e
ap
pl
i
e
d
t
o
p
redi
ct
d
r
o
po
ut
b
y
usi
n
g
ou
tlier
d
e
tection
with
ou
t
an
y
learn
i
ng
su
perv
ision
.
2.
R
E
SEARC
H M
ETHOD
2.
1.
Pr
obl
em
setti
n
g
f
or
dr
o
pou
t
pre
di
cti
o
n
Monitoring
a
nd
s
uppo
rti
n
g
the
st
ude
nts
highly
n
eces
sary
accordi
n
g
t
o
t
h
e
de
part
m
e
nt
o
f
e
ducat
i
o
n
.
w
h
e
n
t
h
e
t
e
a
c
h
e
r
i
s
a
b
l
e
t
o
t
r
a
c
k
a
s
t
u
d
e
n
t
w
i
t
h
h
i
g
h
r
i
s
k
b
e
i
n
g
d
r
o
p
out
f
r
o
m
t
h
e
begi
n
n
i
n
g,
t
hey
can
t
ake
act
i
on
i
m
m
e
diat
el
y
and
m
a
ke
s
u
r
e
t
o
h
el
p
t
h
at
s
t
u
dent
s
o
he
o
r
sh
e
will
n
o
t
b
e
exp
e
lled
fro
m
t
h
e
s
ch
oo
l
.
Hen
c
e,
it
is
i
m
p
o
r
tan
t
t
o
p
r
ed
ict
th
o
s
e
risk
y
stud
en
ts
over
th
e
c
lass,
so
t
h
a
t
t
h
e
teach
e
rs
w
ou
ld
g
iv
e
th
em
speci
al
g
ui
da
n
ce.
I
t
i
s
pot
e
n
t
i
a
l
t
o
c
o
n
t
r
ol
s
t
u
dent
act
i
o
n
usi
n
g
l
og
-
d
at
a
i
n
t
he
p
r
o
gram
m
i
ng
cl
ass.
B
y
devel
opi
ng
a
l
og
gi
n
g
s
y
s
t
e
m
t
h
at
can
r
ec
o
r
d
appl
i
cat
i
o
n
t
r
a
ces
o
f
s
ou
rce
co
d
e
e
d
iting
and
UNIX
co
mman
d
,
a
dat
a
set
w
h
i
c
h
obt
ai
ne
d
fr
om
o
u
r
p
r
o
g
r
am
m
i
ng
l
e
ss
o
n
’s
s
t
u
dent
s
f
r
o
m
39
st
ude
nt
s.
M
an
ufact
uri
n
g
e
v
al
uat
o
r
s
wi
t
h
l
ear
ni
n
g
s
up
er
vi
si
o
n
a
r
e
c
om
m
onl
y
used
t
o
p
r
edi
c
t
a
per
s
o
n
'
s
d
r
o
po
ut
pot
e
n
t
i
a
l
.
H
owe
v
er
,
t
h
e
dat
a
i
s
di
ffi
c
u
l
t
t
o
u
nd
erst
an
d,
a
s
t
h
e
si
ze
gi
ve
n
f
r
o
m
our
n
ar
ro
w
dat
a
set.
A
lso,
t
he
f
eatures
i
n
t
h
e
l
o
g-data
d
e
p
ends
on
t
h
e
elem
ents
f
rom
every
class,
s
uc
h
as
i
f
there
is
a
l
ot
o
f
training
o
r
a
plenty
o
f
expla
n
ations.
As
a
r
esult,
we
u
se
a
n
un
atten
d
e
d
learn
i
ng
m
eth
o
d
w
ith
o
u
tlier
d
e
tecti
o
n,
a
s
su
m
i
n
g
t
h
at
s
tud
e
n
t
s
as
p
art
of
a
n
outlier
cl
ust
e
r
ca
n
be
c
om
pared
base
d
on
s
t
u
de
nt
s’
achi
e
vem
e
nt
s,
e
i
t
h
er
s
u
p
e
r
i
o
r
o
r
i
n
f
e
r
i
o
r
s
t
u
d
e
n
t
s
[
1
6
]
,
[
1
7
]
.
T
h
e
appl
i
cat
i
o
n
of
t
hi
s
k-m
eans
cl
ust
e
ri
n
g
t
ech
ni
que
i
s
ad
ju
st
ed
w
i
th
E
u
c
lid
ean
d
i
stan
ce,
in
o
rd
er
t
o
do
c
lu
stering
by
u
si
n
g
t
he
d
y
n
am
i
c
tim
e
warpi
ng
an
d
be
n
c
hm
arki
ng
a
g
a
i
nst
act
i
ve
t
ime
beha
vi
or
.
Th
eref
ore
,
i
t
i
s
possi
bl
e
for u
s
t
o
co
m
p
are th
e flow of
activ
ities to
th
e
ex
c
ep
tio
n of ti
m
e
-seri
e
s de
vi
at
i
ons
[
1
8
]
.
2.2. Dynamic time wrapping
Dy
nam
i
c
t
i
m
e
w
ar
pi
n
g
(
D
T
W)
i
s
an
a
l
g
ori
t
hm
t
hat
used
t
o
m
easure
t
h
e
s
i
m
i
l
a
r
i
t
i
e
s
b
e
t
w
e
e
n
t
h
e
t
w
o
seq
u
ences
w
i
t
h
di
ffe
re
nt
l
en
gt
hs
o
r
am
oun
t
s
o
f
dat
a
.
D
T
W
m
a
t
c
he
s
two
se
quences
by
calculating
t
e
m
poral
i
n
f
o
rm
at
i
on
so
t
hat
bot
h
o
f
t
hem
can
b
e
ali
gne
d.
A
l
i
g
nm
ent
i
s
t
he
s
m
a
llest
m
easured
c
um
ulative
distance
bet
w
ee
n t
w
o s
y
nced sam
pl
es.
If i
t
the
n
a
ss
umed
t
hat
there are
t
wo
s
eq
u
e
ntial
d
a
ta,
Q
an
d
C,
w
ith
t
h
e
r
an
g
e
of
n
an
d m
sev
e
r
a
lly as sh
own
in (1
)
and
(
2)
[
19], [
20
].
Q = q1
, q2
,
…
,
qi
,
…
,
qn
(1
)
C =
c1
,
c2
, …,
c
j
,
…,
cm
(2
)
Th
en
,
to
a
lig
n
th
ese
two
sequ
en
ces
using
d
y
n
a
m
i
c
ti
me
w
arp
i
ng
,
a
m
a
t
r
i
x
i
s
f
o
r
m
e
d
m
×
n
w
i
t
h
matrix
ele
m
e
n
t
(i,j
)
in
th
e form o
f d
i
stan
ce v
a
lu
e
d(
q
i
,c
j
)
between two
q
i
poi
nt
s, an
d
dec
l
a
red as
d(
q
i
,c
j
)
= (
q
i
–
c
j
)
2
.
Each
o
f
matrix
e
le
m
e
n
t
(
i,j
)
rel
a
t
e
s
to
a
l
i
gn
bet
w
ee
n
q
i
a
n
d
c
j
poi
nt
s.
W
a
r
pi
ng
pat
h
W
i
s
a
g
r
o
u
p
o
f
ad
jo
in
i
n
g
m
a
tr
ix
e
lem
e
n
t
s
th
at
d
efin
e
m
a
p
p
i
n
g
b
e
tween
Q
a
n
d
C
.
T
h
e
k
e
l
e
m
e
n
t
o
f
W
i
s
fo
rm
ulated
a
s
w
k
=
(
i,j
)
k
, s
o
we
g
ot
(
3).
,
,…,
,…,
(
3
)
with
:
,
–
1
.
Wh
ile
t
h
e
p
ath
is
d
efin
ed
a
s
th
e
cu
m
u
lativ
e
d
i
stan
ce
D(i,j),
th
at’s
d
istance
d
(qi,cj)
for
the
ele
m
ents
a
dde
d
with
th
e
m
in
i
m
u
m
cu
m
u
l
ativ
e d
i
st
ance
from
adjace
nt elem
e
nts, as
show
n i
n
(
4)
.
D(
i,j)
= d(
q
i
,c
j
)
+
min
{
D(
i-
1,j-
1)
,D(
i
-1,j)
,
D(
I,j-1)
}
(4
)
Once
o
bt
ai
ne
d
t
h
e o
p
t
i
m
al
warpi
ng
pat
h, t
he
d
i
s
t
a
nce
or
wa
r
pi
n
g
c
ost
i
s
cal
cul
a
t
e
d
base
d
on
(
5)
.
DTW
,
m
i
n
∑
(
5
)
Evaluation Warning : The document was created with Spire.PDF for Python.
ISS
N
:
2252-
8
938
I
n
t
J
A
r
tif
In
t
ell,
V
o
l
.
10
,
No
.
3
,
S
ep
tem
b
er
2
021
:
7
64
-
770
76
6
2.3. CL
G clus
tering
In
t
hi
s
st
u
d
y
,
a
p
r
o
pose
d
m
odi
fi
cat
i
on
m
e
t
hod
i
n
t
he
c
l
u
st
eri
n
g
a
lgo
r
ith
m
is
c
o
m
p
l
ete
lin
k
a
g
e
di
ssi
m
i
l
a
ri
t
y
i
ncrem
e
nt
d
i
s
t
r
i
but
i
o
n-
gl
o
b
al
c
um
ul
at
i
v
e
score
st
an
dar
d
(
C
L
G)
,
t
h
i
s
a
l
gori
t
h
m
i
s
a
c
om
bi
ned
al
go
ri
t
h
m
bet
w
een
t
he
c
om
pl
et
e
l
i
nka
ge
(
C
L
)
al
g
o
ri
t
h
m
[2
0]
,
t
h
e
d
issimilarit
y
i
n
c
remen
t
d
istribu
tio
n
(DID)
al
go
ri
t
h
m
[20]
,
gl
obal
c
u
m
u
l
a
t
i
v
e
score
st
anda
r
d
(
GC
SS
)
al
g
o
ri
t
h
m
[
2
1
]
.
T
h
e
C
L
G
a
l
g
o
r
i
t
h
m
w
o
r
k
s
b
y
com
b
i
n
i
ng
el
e
m
ent
s
o
f
free
gra
p
h-
base
d
p
a
ram
e
t
e
rs
a
nd
m
odel
-
bas
ed
a
pp
r
o
aches
(
w
h
i
c
h
a
r
e
de
fi
n
e
d
by
co
m
b
in
in
g
criteria b
y
ch
a
racterizin
g clu
s
ters in
p
r
ob
ab
ilistic
ter
m
s)
f
or
g
ro
up
ing
.
CL=
,
,
,
(
6
)
DI
D=
;
(
7
)
√
X
√
√
GCSS=
,
,
,
,
,
,
,
,
,
,
,
,
,
Υ
,
Ψ
,
,
,
,
,
,
,
,
(8)
Th
e
CLG
algorith
m
p
r
ov
id
es
d
ifferen
t
t
reatmen
t
t
o
s
m
a
ll
clu
s
ter
can
di
da
t
e
g
ro
up
s.
E
ac
h
can
di
dat
e
gr
o
ups
w
h
o
se
s
i
ze
i
s
l
ower
t
han
YM
I
N
i
s
not
r
e
q
ui
re
d
t
o
e
xpl
ai
n
t
he
m
ergi
ng
c
r
i
t
e
ri
a.
I
n
fact
,
t
h
e
m
e
rg
e
r
b
e
t
w
e
e
n
C
i
a
n
d
C
j
a
l
w
a
y
s
o
c
c
u
r
s
i
n
t
h
e
c
a
s
e
o
f
t
h
e
t
w
o
g
r
o
u
p
s
o
f
can
d
i
dates
less
th
an
t
h
e
v
alu
e
o
f
th
e
YMIN
ob
ject
.
R
e
gar
d
i
ng
t
h
e
cl
ust
e
r
si
ze
t
h
res
hol
d,
i
t
i
s
i
m
port
a
nt
t
o
n
o
t
e
t
he
d
i
f
fere
nce
bet
w
ee
n
t
h
e
H
a
nd
Y
M
I
N
param
e
ters;
because
b
oth
values
r
e
f
er
t
o
group
size,
p
a
r
a
m
eter
H
i
s
t
h
e
r
e
a
l
v
a
l
u
e
u
s
e
d
i
n
t
h
e
c
a
l
c
u
l
a
t
i
o
n
o
f
t
h
e
dy
nam
i
c
m
e
rge
t
h
r
e
sh
ol
d
,
w
hi
l
e
Y
M
I
N
i
s
t
he
i
nt
ege
r
t
h
r
esh
o
l
d
val
u
e u
s
ed w
hen
di
rec
t
i
ng
t
h
e
c
o
m
p
ari
s
o
n
wi
t
h
t
he
re
q
ui
r
e
d cl
u
s
t
e
r si
ze.
2.
4.
k-
mea
ns
++
Algo
rith
m
in
k
-m
ean
s
o
f
ten
app
lied
i
n
c
l
u
stering
techn
i
q
u
e
s
th
a
t
aim
to
m
in
i
m
izes
t
h
e
s
qu
ared
distance that has been levele
d
bet
ween
p
oi
nt
s i
n
t
he sam
e cl
ust
er. B
ut
t
he
a
l
g
o
r
i
t
h
m
of k-
m
eans al
go
ri
t
h
m
has
a
disadva
n
tage
t
hat
cannot
p
rovi
de
p
recise
accuracy
e
ve
n
using
si
m
p
le
a
n
d
f
ast
cal
cul
a
t
i
ons
[
22]
,
[2
3]
.
If
k-m
ean
s
ad
d
e
d
with
r
ando
m
i
zed
s
eed
in
g
tech
n
i
q
u
e
w
ill
i
m
p
r
o
v
e
t
he
accuracy
f
rom
the
a
l
gorithm
of
k-m
eans.
T
he
accuracy
f
rom
the
algorit
h
m
of
k
-m
eans
heavily
d
ep
e
n
ds
o
n
a
val
u
e
o
f
cent
r
oi
d
(C
)
at
t
he
begi
nni
ng
o
f
t
h
e
cal
cul
a
t
i
on, t
hen i
f
usi
ng
a
di
ffe
re
nt
C
val
ue
will g
i
v
e
d
ifferen
t
resu
lt
even
if requ
ires a lo
t
o
f
iteratio
n
s
t
o
d
e
termin
e
th
e
m
e
m
b
er
o
f
a
cl
u
s
ter
if
t
h
e
v
al
u
e
C
in
app
r
op
r
i
ate.
B
y
add
i
ng
f
o
r
m
u
las
r
a
ndomized
seed
ing
techn
i
q
u
e
,
th
en
it
will
d
e
termin
e
th
e
v
a
lu
e
of
C
a
t
t
h
e
b
e
g
inning
o
f
the
calculation.
E
ac
h
m
e
m
b
er
h
a
s
th
e
op
portun
ity
b
eco
m
e
a
cen
tro
i
d
so
t
h
e
v
alu
e
o
f
opp
ortu
n
ities
o
f
eac
h
m
e
m
b
er
i
s
counte
d
t
o
found
whic
h
one
i
s t
h
e
m
o
st ap
p
r
o
pri
a
t
e
.
H
e
re i
s a
ra
nd
om
i
zed see
di
n
g
t
e
c
hn
i
que
f
orm
u
l
a
.
∑
∈
2.
5.
k-
met
hod
s
Th
e
k
-
m
e
d
o
i
ds
a
lg
orith
m
is
a
c
lassic
p
a
rtitio
n
i
ng
t
ech
n
i
qu
e
o
f
cl
ust
e
ri
n
g
t
hat
pe
rf
orm
s
c
l
u
st
eri
ng
d
a
taset
o
f
n
o
bj
ects
in
to
k
c
lusters,
kno
wn
a
s
a
p
r
i
o
ri.
Th
is
a
l
g
o
r
i
t
h
m
operat
e
s
o
n
p
ri
nci
p
l
e
t
o
m
i
nim
i
ze
t
he
am
ount
o
f
sim
ilarity
b
etween
each
o
bject
a
ppropriate
r
efe
r
ence
po
in
t.
T
he
k
-m
ed
o
i
d
s
a
lg
orith
m
can
b
e
d
o
n
e
as bei
ng as
[24]
, [25]
:
In th
e
first step, in
itialise th
e
cen
t
er o
f th
e cluster b
y
k
(t
he
a
m
ount
o
f cl
ust
e
rs)
.
In th
e
secon
d step
, co
u
n
t
each
en
tity to
a n
e
arb
y
cl
u
ster u
si
n
g
E
uclidian Distance size e
q
uations.
The
thi
r
d
step,
after
calc
u
latin
g
the
E
u
clidi
a
n
Distance,
i
niti
alize
the
center
of
t
he
n
ew
c
luste
r
eac
h
ob
ject
as
a
no
n
-
m
e
doi
ds ca
ndi
dat
e
.
Th
e
fourth
s
tep
,
m
easu
r
e
the
g
a
p
b
e
tween
each
en
tity
l
o
cated
o
n
e
ach
cl
u
s
ter
with
n
on-app
licant
m
e
doi
ds
.
The
fifth
step,
m
easure
the
t
o
tal
de
viation
(S)
by
p
roce
ss
ing
t
he
n
e
w
t
ot
al
d
i
s
t
a
nce
–
t
h
e
ol
d
t
o
t
a
l
d
i
stan
ce.
I
f
S
<
0
,
t
h
e
n
ex
chan
g
e
e
n
tity
w
ith
non
m
ed
o
i
d
s
c
lu
st
er
d
ata
to
f
orm
a
new
se
t
of
k
o
bjects
a
s
m
e
doi
ds
.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J Artif In
t
el
l
I
S
SN
: 225
2-8
9
3
8
CLG
cl
u
s
tering
for
d
r
opo
u
t
pred
ictio
n u
s
i
n
g lo
g-d
a
t
a
clu
s
t
e
rin
g
m
e
t
h
od
(
A
gu
ng
Tria
yu
di
)
76
7
Th
e
si
x
t
h
step,
rep
eat
s
teps
3
-5
u
n
til
no
m
o
r
e
ch
ang
e
s
to
t
h
e
m
edoi
d
,
t
he
n
we
a
re
a
l
r
ea
dy
g
ot
c
l
u
st
er
me
m
b
ers and t
h
eir res
p
ective cluster m
e
m
b
ers.
2.
6.
E
x
peri
ment
ati
o
n d
a
tas
e
t
UN
IX
c
om
m
a
nd
i
n
p
u
t
hi
st
o
r
y
are
use
d
d
uri
ng
e
x
am
due
t
o
p
r
og
ram
m
i
ng
cl
ass
co
nsi
s
t
of
3
9
st
ude
nt
s.
A
ss
u
m
ed
t
hat
t
h
e
l
o
g
-
dat
a
can
r
at
e
m
a
ny
a
spect
s,
s
uch
a
s
m
o
tiv
atio
n,
i
n
d
i
v
i
du
al
s
k
ills,
and
o
t
h
e
rs.
The
n
w
e
nee
d
t
o
b
r
eak
t
he
t
e
achi
n
g
si
g
n
al
s
fr
om
t
hi
s
l
og-
d
a
t
a
t
o
create
a
new
one
bi
nary
linear
c
lassifier
t
hat
separat
e
s
a l
a
r
g
e
of
st
u
dent
b
ased
o
n t
h
ei
r
l
e
vel
.
Th
en
w
e
con
s
i
d
er
a
w
ay
t
o
easily
c
lassif
y
a
b
o
u
t
t
h
e
g
r
oup
o
f
s
t
u
d
ent
s
w
i
t
h
u
nsu
p
e
r
vi
s
e
d
l
ear
ni
n
g
with
ou
t
firstly
p
rep
a
red
a
qu
an
titativ
e
ev
aluatio
n
m
ach
in
e.
T
h
e
o
u
tlier
cl
ass
con
s
id
ered
f
ro
m
o
n
e
s
ubset
o
f
th
is
g
rou
p
.
Accu
m
u
lated
ti
me-series
d
a
ta
o
f
fiv
e
-m
in
u
t
es
U
NIX
c
om
m
a
nd
s
i
n
put
w
i
t
h
k
-
m
e
doi
d
m
e
t
h
o
d
s
will
in
teg
r
ate
k
-
m
ean
s++
fo
r
in
itial
v
a
lu
e
defin
itio
n.
H
ereafte
r,
w
e
in
sp
ect
th
e
trend
o
f
t
h
e
c
lu
sters
b
e
l
o
ng
ed
th
en
set th
e
ou
tlier clu
s
ter t
o
ev
e
ry lesson
from
th
e ev
al
u
a
tio
n.
3.
RESULTS
A
ND
DI
S
C
U
S
S
I
ON
3.
1.
Fe
ature
v
ector
veri
fi
cat
i
on
Figure
1
s
hows
t
he
c
ommand
input
ratio
o
f
each
s
tude
nt’s
c
lasse
s
,
w
h
e
r
e
t
h
e
p
i
c
t
u
r
e
g
i
v
e
n
p
r
e
s
e
n
t
s
the
exec
utable
f
iles
as
“
ls”,
“cd
”,
a
nd
“
g
cc
”.
L
ooki
ng
a
t
the
g
raph,
ca
n
be
s
een
t
hat
the
ratio
o
f
c
o
mmand
i
n
p
u
t
use
d
d
ep
endi
ng
o
n
t
he
s
ub
ject
s.
T
he
r
e
fo
re,
i
t
i
s
n
ot
a
pp
r
op
ri
at
e
t
o
b
e
u
s
ed
a
s
an
i
np
ut
g
ui
de
f
or
t
he
perform
a
nce of each class
.
Fi
gu
re
1
.
Eval
uat
i
o
n
f
r
om
st
u
dent
g
rade
E
3.
2.
Clus
terin
g
me
t
h
o
ds
f
or
o
utlier
detec
t
ion
For
insta
n
ce,
t
he
r
e
s
ult
by
g
rouping
eac
h
ot
her
class
can
b
e
se
en
i
n
Fi
gu
r
e
s
2
a
n
d
3.
I
n
t
h
i
s
case,
t
he
num
ber
of
c
lus
t
ers
are
arra
nged
o
n
a
scale
of
0
t
o
4,
b
eca
use
a
t
t
h
e
e
n
d
of
t
he
l
esso
n,
t
he
n
um
ber
o
f
i
np
ut
s
fr
om
t
he
c
l
a
ss
m
e
doi
d
cl
ust
e
r
i
s
l
ow
.
It
i
s
c
o
m
m
onl
y
kn
ow
n
t
h
at
i
n
th
is
c
ase,
c
lu
sters
ten
d
t
o
sim
p
ly
c
l
a
ssify
th
e
stud
en
ts
b
ased
o
n
th
e
qu
an
tity
o
f
inpu
t.
E
x
c
ep
t,
f
or
t
ho
se
user
s
w
h
o
sud
d
e
n
l
y
e
x
p
e
r
i
e
nce
a
n
i
nc
re
asi
n
g
th
e
nu
m
b
er
o
f
in
pu
ts
w
ill
create
th
eir
o
w
n
clu
s
ter.
F
igure
2
su
it
th
is
p
h
e
no
m
e
n
o
n
.
Wh
en
t
h
e
d
ata
lo
cated
out
si
de
f
r
o
m
the
cl
ust
e
r
l
e
ss
t
h
an
10%
o
f
t
h
e
am
ount
c
l
u
st
ers,
t
h
e
n
t
he
t
e
nde
ncy
o
f
t
hi
s
di
sl
ocat
ed
c
l
u
s
t
er
can
b
e
d
escri
b
ed
a
s:
i
)
less
co
mman
d
inpu
t
than
o
th
er
c
lusters
as
s
h
o
w
n
i
n
F
i
g
ure
3;
a
n
d
i
i
)
t
he
i
np
ut
i
ncrease
s
r
a
p
i
d
l
y in
a sho
r
t
p
er
i
o
d of
ti
m
e as sho
wn
i
n
Fi
g
u
r
e
2
.
3.3. Outlier clust
er interpre
tation
Th
is
i
s
th
e
characteristic
o
f
th
e
stud
en
ts
b
elon
g
e
d
t
o
t
h
e
o
u
t
l
y
i
ng
cl
ust
e
rs
u
si
n
g
t
he
c
l
u
st
e
r
i
n
g
feat
ure
s
s
uc
h
a
s
t
he
s
t
u
dent
’s fi
v
e-
g
r
ade pre
d
i
c
t
i
on (A
t
o
E
)
.
Fi
g
u
r
e
1 bel
o
n
g
t
o
a
st
ude
n
t
w
h
o
se
e
val
u
a
t
i
on
i
s
E
and
Fi
gu
re
4
b
el
o
n
g
t
o
a
st
ude
nt
w
ho
g
o
t
pre
d
i
c
t
i
on
D
as
h
i
s
i
nde
x
of
achi
e
vem
e
nt
.
These
t
y
pi
c
a
l
i
t
e
m
s
p
r
ob
ab
ly su
s
p
e
ct as supp
or
ter
f
o
r
so
lv
i
n
g th
e issu
es.
Fi
gu
re
2
a
ppl
i
e
d
t
o
a
p
l
e
nt
y
of
s
t
u
dent
s,
e
speci
al
l
y
f
or
s
t
ude
n
ts
w
ith
e
v
a
lu
ation
g
r
ade
A
an
d
C
.
Whi
l
e
i
t
i
s
pos
si
bl
e
t
h
at
t
he
p
ro
gram
m
ay
n
o
t
w
or
k
w
e
l
l
.
We
det
ect
a
h
i
gh num
ber
o
f
c
om
m
a
nd
i
n
put
e
nt
ered
d
u
r
i
ng
d
eb
ugg
in
g
wo
rk
,
or
t
heir
t
ask
w
a
s
f
i
n
i
sh
ed
e
ar
lier
w
h
en
t
h
ey
s
till
p
r
o
ceed
e
d
th
e
task
d
uring
p
e
rson
al
learnings.
Figure
3
b
elongs
to
a
c
luster
o
f
each
c
lass
on
one
t
est.
T
he
e
v
al
uat
i
o
n
i
s
a
bout
t
he
s
im
il
ari
t
y
o
f
appl
i
cat
i
o
n
t
e
c
hni
que
s
i
n
s
t
u
dent
s
t
h
at
a
re
b
al
ance
d
e
v
e
n
w
i
t
h
l
o
w
m
o
tivation.
B
ecause,
in
a
t
est
there
are
Evaluation Warning : The document was created with Spire.PDF for Python.
ISS
N
:
2252-
8
938
I
n
t
J
A
r
tif
In
t
ell,
V
o
l
.
10
,
No
.
3
,
S
ep
tem
b
er
2
021
:
7
64
-
770
76
8
two
m
a
in
p
rob
l
em
s,
n
am
e
l
y
reg
a
rd
ing
o
n
p
rog
r
ammin
g
an
d
writin
g,
w
he
re
m
ost
o
f
t
he
t
i
m
e
drai
ned
by
writing
.
Fig
u
re 2
. Ou
tpu
t
1 fro
m
ti
m
e
-
s
eries clu
s
tering
Fig
u
re 3
. Ou
tpu
t
2 fro
m
ti
m
e
-
s
eries clu
s
tering
Fi
gu
re
4
.
Eval
uat
i
o
n
f
r
om
st
u
dent
g
rade
D
3.4.
Resolution c
oncerning
the amount of clusters
As
f
ar
a
s
has
been
p
re
dicted,
th
e
nu
m
b
er
o
f
f
i
n
a
l
inpu
ts
i
s
m
o
r
e
dom
i
n
ant
i
n
t
hes
e
m
et
hods
t
ha
n
t
h
e
way
num
ber
o
f
c
om
m
a
nd
i
nput
i
nc
rease
d
u
ri
ng
t
h
e
a
n
al
y
s
i
s
.
Thi
s
i
s
e
xpect
e
d
t
o
be
t
he
b
asi
s
i
n
m
oni
t
o
ri
ng
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J Artif In
t
el
l
I
S
SN
: 225
2-8
9
3
8
CLG
cl
u
s
tering
for
d
r
opo
u
t
pred
ictio
n u
s
i
n
g lo
g-d
a
t
a
clu
s
t
e
rin
g
m
e
t
h
od
(
A
gu
ng
Tria
yu
di
)
76
9
and
grouping
these
fi
ve
g
roups
i
n
all
a
n
al
yses.
It
i
s
just
t
hat
,
fo
r
t
h
e
st
ude
nt
s
i
n
t
he
F
i
g
ure
3,
t
he
r
e
i
s
a
t
e
nde
ncy
t
h
at
m
ore
st
udent
s
are
i
n
t
he
g
r
o
u
p
w
i
t
h
l
ess
i
n
p
u
t
.
C
l
u
st
eri
n
g
i
s
unat
t
e
n
d
e
d
l
earni
ng
but
l
o
o
k
i
n
g
at
th
e resu
lting
data th
e
re will be n
o
righ
t an
swer o
f t
h
e cluster
s is nee
ded.
3.
5.
T
r
a
n
si
ti
o
n
reg
ardi
n
g
t
he num
ber
of
c
lusters
Si
nce
di
scuss
e
d
i
t
earl
i
e
r,
t
he
r
es
ul
t
s
o
f
t
h
e
gr
o
upi
ng
o
n
F
i
g
u
re
3
a
re
c
l
a
s
s
i
f
i
e
d
base
d
o
n
t
h
e
num
ber
of
c
om
m
a
nd
i
n
p
u
t
d
u
ri
ng
t
h
e
cl
asses,
e
xce
p
t
whe
n
t
hey
expe
ri
enc
e
with
i
ncreas
es
unexpecte
d
ly
i
n
a
short
peri
od
.
I
n
t
hi
s
st
udy
,
we
can
s
ai
d
t
h
at
C
l
u
st
er
0
c
on
si
st
s
o
f
t
he
s
t
u
de
nt
w
h
o
o
nl
y
ha
ve
a
f
e
w
i
np
ut
s,
w
hi
l
e
C
l
ust
e
r
4
c
o
nsi
s
t
s
o
f
t
h
e
st
u
d
e
n
t
w
ho
ha
ve
a
l
ot
o
f
i
n
p
u
t
s
.
S
t
u
de
nt
s
can
b
e
pre
d
i
c
t
e
d
as
a
l
i
n
e
of
n
um
bers.
I
n
th
is
case,
s
tu
d
e
n
t
s
o
n
th
e
left
o
f
th
e
cen
ter
po
in
t
ten
d
to
h
av
e
a
few
com
m
ands
i
np
ut
,
whi
l
e
t
hose
on
t
he
r
i
ght
w
ill
h
a
v
e
m
an
y
co
mman
d
inp
u
t
s.
T
h
e
s
tudy
fo
cu
sed
on
t
h
e
d
etecti
o
n
of
o
u
tliers
a
p
p
l
i
e
d
to
s
p
ecify
t
rend
s
acknowledge t
h
e fre
que
ntative be
ha
vi
ors
o
f
m
any
cl
asses, t
hen
i
ns
pecting the tre
nds
of ea
ch less
o
n.
3.
6.
A
tti
tude
inves
tig
a
ti
on
b
y q
u
esti
onn
a
ir
es
Al
l
of
t
hi
s
t
i
m
e
,
we
t
ry
t
o
m
a
nage
a
quest
i
o
n
n
ai
re
a
bo
ut
t
o
ens
ure
ho
w
m
u
ch
t
he
o
ut
l
y
i
ng
cl
ust
e
r
m
eans.
T
he
re
a
re
s
eve
r
al
c
on
si
derat
i
o
ns
w
e
r
e
m
a
de,
t
w
o
of
t
hem
ar
e
lik
es:
“H
ow
m
u
c
h
d
i
d
you
under
s
tand
the
today’s
content
?
”,
a
nd
“
How
m
u
ch
d
i
d
y
ou
p
ai
d
at
t
e
n
t
i
on
fo
r
t
oday
’
s
lesson
?
”
The
form
at
o
f
t
h
e
solution
we
c
reat
e
d
i
s
sel
f-e
val
u
at
i
o
n
and
bei
n
g
or
g
a
ni
zed
i
nt
o
f
o
ur
p
oi
n
t
s
ystem
s
,
as
t
h
e
s
tu
d
e
n
t
s
will
g
e
t
it
o
n
c
e
d
u
ring
t
h
e
c
lasses
a
nd
e
x
a
m
.
T
h
e
n
th
e
resu
lt
th
ere
will
b
e
n
o
s
ignificant
diffe
rence
c
o
m
p
ared
t
o
t
h
e
va
riation
an
d
ou
tlier
cl
usters
s
in
ce
t
h
e
s
co
res
fro
m
th
e
an
swer
e
ith
er
t
h
at
l
ow.
For
this
r
eason,
we
cannot
argue
d
t
hat
ou
r
hy
p
o
t
h
esi
s
i
ncre
di
bl
y
val
i
d,
b
eca
use
a
n
ot
he
r
m
e
t
hod
of
e
val
u
a
tion
i
s
s
till
req
u
i
red
.
O
v
e
rall,
o
u
r
s
u
g
g
e
st
plays a m
a
jor part as a vis
u
ali
zer of student m
o
tivation when u
nsupervise
d learning sta
rted.
3.
7.
U
N
I
X
c
o
mman
d
l
o
g
m
anu
a
l
veri
fi
ca
ti
on
M
a
nual
l
y
,
we
i
nvest
i
g
at
e
d
d
i
ffe
rent
t
y
p
e
o
f
s
t
ude
nt
’s
b
e
h
avi
o
r
d
u
ri
ng
l
e
sso
ns
u
si
ng
l
o
g-
dat
a
f
r
o
m
clu
s
ters
o
u
tliers.
Th
e
resu
lts
o
f
t
h
is
i
n
v
e
stigatio
n
h
a
v
e
b
een
co
ncl
u
de
d:
i
)
St
ude
nt
s
wh
o
i
n
creasi
n
gl
y
p
r
essed
the
keyboard
in
a
p
eri
od
of
t
im
e
resulting
“gcc”
c
o
mmand
the
n
r
un
t
h
e
p
r
ogr
am.
So
m
e
s
tu
d
e
nts
m
a
y
expe
ri
ence
p
r
o
bl
em
s
wi
t
h
i
n
c
om
pi
l
a
t
i
on
er
ro
rs
o
r
pr
og
ra
m
bugs
b
ase
d
o
n
o
u
r
i
n
vest
i
g
at
i
on.
O
n
t
h
e
ot
her
h
a
nd
,
cod
i
ng
g
o
e
s
w
e
ll
w
ith
ou
t
an
y
pr
oble
m
s.
B
y
lo
o
k
in
g
at
t
h
i
s
phen
o
m
en
on
,
w
e
cann
o
t
c
lassi
f
y
t
he
p
r
ed
icate
of
t
h
o
s
e
stud
en
ts
b
ased
o
n
how
m
u
c
h
th
e
num
b
e
r
o
f
i
np
u
t
a
nd
d
u
r
ation
wh
ile
p
r
e
ssi
ng
t
h
e
key
b
o
ar
d. i
i
)
S
t
ude
nt
s w
h
o
s
e onl
y
pr
essed
the
k
e
ybo
ard
a
few
ti
m
e
s
m
a
y
not
be abl
e
t
o com
p
l
e
t
e
t
he t
ask an
d
h
a
d
a
h
i
g
h
risk
b
ei
n
g
d
ropped
ou
t
as
w
e
p
r
ed
icted
earlier.
iii)
S
t
u
de
n
t
s
wh
o
s
u
d
d
e
n
l
y
e
xp
eri
e
nce
d
t
he
in
creasing
n
u
m
b
er
o
f
in
pu
t
wh
ile
p
ressi
n
g
k
ey
b
o
ar
d
an
d
UN
IX
c
om
m
a
nds
w
hen
past
ed
s
o
u
rc
e
co
de
i
nt
o
t
h
e
com
m
a
nd
l
i
n
e
m
a
de
t
hi
s
i
n
fo
r
m
at
i
on
po
werl
ess
i
f
onl
y
bei
ng
i
n
ve
stigate
d
b
y
the
num
ber
of
U
NIX
com
m
and
i
ssues.
It
i
s
nec
e
ssary
t
o
ad
d i
n
f
o
rm
at
i
on s
u
c
h
as
com
m
and
val
u
es or im
p
l
e
m
en
tatio
n
resu
lts.
4.
CO
NCL
USI
O
N
Thi
s
s
t
u
dy
p
r
o
pos
es
w
ay
s
o
r
m
et
hods
t
o
e
v
al
uat
e
t
hose
st
ude
nt
s
w
h
o
ar
e
bei
n
g
ri
sk
o
f
dr
op
pe
d
o
u
t
fr
om
s
chool
,
b
y
gro
u
p
i
n
g
t
h
e
m
w
i
t
h
uns
upe
rvi
s
e
d
s
t
u
dy
u
s
i
ng
o
u
t
l
i
e
r
det
ect
i
on.
W
e
use
t
h
e
dat
a
d
epe
nd
o
n
th
e
lesson
’
s
purpo
se,
m
a
k
e
s
it
d
i
fficu
lt
wh
ile
c
reated
b
y
ev
alu
at
i
o
n
en
gi
ne.
F
o
r
t
h
i
s
r
e
a
so
n,
w
e
i
nve
s
t
i
g
at
ed
th
e
g
r
o
up
of
o
u
tliers
b
y
d
i
v
i
d
e
t
h
e
m
in
to
t
h
r
ee
t
ren
d
s
w
ith
a
p
redi
ct
abl
e
cause
,
s
o
t
ha
t
st
ude
nt
s
w
h
o
ha
v
e
learning
p
roble
m
s
can
b
e
detected
a
s
s
o
o
n
a
s
p
o
s
s
i
b
l
e
.
H
o
w
e
v
e
r
,
a
s
o
u
r
p
r
op
ose
d
p
re
di
ct
i
on
m
e
t
h
o
d
s
a
r
e
st
i
ll
ne
e
d
f
u
r
t
h
e
r
de
v
e
l
o
p
m
e
nt
,
t
h
is
r
esearc
h
n
ee
d
anothe
r
proper
m
et
hod suc
h
as vi
sual
i
zat
i
on of st
u
de
nt
beha
v
i
o
r
base
d i
n
l
o
g
-
d
a
t
a.
REFERE
NC
ES
[1]
A
.
T
r
i
a
y
u
d
i
a
n
d
I
.
F
i
t
r
i
,
“
A
L
G
C
l
ustering
to
A
naly
ze
t
he
B
eh
avi
o
r
al
P
atterns
of
O
nline
Learning
S
t
udent,”
Journal
of Theoretical a
nd Applied
Information Techno
lo
gy
, vo
l
. 96, no. 1
6, pp
. 5327-533
7, 2018
.
[2]
A
.
T
r
i
a
y
u
d
i
a
n
d
I
.
F
i
t
r
i
,
“
A
n
e
w
agglom
erat
ive
hi
erarch
ic
al
c
lu
stering
to
m
odel
student
a
ctivity
i
n
online
learning,”
TELKOMNIKA Telecommunication Computin
g Electronics and Control
,
vol.
17,
no.
3
,
pp.
1226-1235,
doi: 10
.12928/telkom
nika.v17i3
.
9425.
[3]
A
.
D
u
t
t
,
M
.
A
.
I
s
m
a
i
l
,
a
n
d
T
.
H
e
r
a
w
a
n
,
“
A
S
y
s
t
e
m
a
t
i
c
R
e
v
i
e
w
o
n
Educ
ation
a
l
D
a
ta
M
ining,
”
I
E
EE Acc
e
ss
,
vol.
5,
pp. 15991-1600
5, 2017
, doi:
10
.1109/ACCESS.2017.2654247.
[4]
A.
T
riay
ud
i
and
I.
F
itri,
“
Comp
arison
of
p
aramet
er-free
agglome
rative
hierar
chical
c
lustering
methods,”
ICIC
Ex
pre
ss Le
tte
r
s
, vol. 12
, no
. 10
, p
p. 973-980
, 201
8, doi: 10
.24507
/icicel.12
.
1
0.973
.
[5]
S.
T
.
Ahmed,
R
.
Al-Hamdani,
an
d
M.
S
.
Croock,
“
Developed
thir
d
itera
tive
di
chot
om
izer
b
as
ed
on
feature
de
cis
i
v
e
values
f
or
e
ducational
data
m
ining,”
Indon
esian Journal of Electrical
Engineerin
g and Computer Science
, vol.
18
,
no. 1
, pp
. 209-2
17, 2020
, doi:
10
.11591/ijeecs.v1
8
.i1.pp209-217
.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISS
N
:
2252-
8
938
I
n
t
J
A
r
tif
In
t
ell,
V
o
l
.
10
,
No
.
3
,
S
ep
tem
b
er
2
021
:
7
64
-
770
77
0
[6]
R
.
A
s
i
f
,
A
.
M
e
r
c
e
r
o
n
,
S
.
A
.
A
l
i
,
a
n
d
N
.
G
.
H
a
i
d
e
r
,
“
A
n
a
l
y
z
i
n
g
u
ndergr
a
d
u
ate
s
t
ud
ents
'
p
e
rform
ance
us
in
g
educa
tiona
l
d
a
ta
m
ining,”
Computers and Edu
c
ation
,
vol.
113,
p
p
.
177-194,
2017,
doi: 10
.1016/j.co
mpedu.2017.05.007.
[7]
A.
D
owah,
H.
A
l-Samarraie,
and
W.
M
.
Fauzy
,
“
Educational
d
a
ta
m
ining
and
learning
analy
t
ics
for
21st
centu
r
y
higher
edu
catio
n:
A
r
eview
and
s
y
nth
e
sis,”
Telematics a
nd Informatics
,
vol.
37,
pp.
13-49,
2019,
doi: 10
.1016/j.tele.2019
.01.007
.
[8]
M.
W
.
Rodrigues,
S
.
Isotani,
a
nd
L
.
E.
Z
ar
ate.
“
Educational
Dat
a
Mining:
A
r
eview
of
e
valuation
process
in
t
he
e
-
learn
i
ng,”
Tel
e
matics and
Informatics
, vol. 35
, no
. 6
, pp
. 1701-17
17
, 2018
, doi: 10
.1016/j.tele.2018
.04.015
.
[9]
C.
R
omero
and
S.
V
entura,
“Educa
tional
data
m
ining
and
learning
a
naly
tics:
A
n
updated
survey
,”
Wi
l
ey
Interdisciplinary Reviews: Data
Mining and Knowledge Disco
very
,
vol.
10,
no.
3
,
Art.
N
o.
e
1355,
2019,
doi: 10
.1002/widm.1355.
[10]
B.
B
akhshinateg
h
,
O.
R
.
Zaian
e
,
S.
E
latia,
a
nd
D.
I
ppe
rc
ie
l,
“
E
ducat
ional
d
a
ta
m
ining
a
pplications
a
nd
tasks:
A
survey
o
f
the
la
st
10
y
e
a
r
s,”
Education
and
Information Technologies
,
vol.
23,
no.
1
,
pp.
537-553,
2018
,
doi: 10
.1007/s10
639-017-9616-z.
[11]
M.
I
njad
at,
A.
M
oubay
e
d
,
A
.
B.
N
a
ssif,
a
nd
A
.
Shami,
“
S
y
stemati
c
ensemble
m
odel
selection
approach
f
or
educa
tiona
l
d
a
ta
m
ining
,
”
Knowledge-
B
ased System
s
,
vol.
200
,
Art.
N
o.
105992,
2
020,
doi: 10
.1016/j.k
nos
y
s
.2020.1059
92.
[12]
T.
D
ev
as
ia,
Vin
u
s
h
ree
T
P
,
a
nd
V.
H
egde,
“Prediction
of
s
tudent
s
performance
u
s
ing
Educational
Data
M
ining
,
”
in
2016 Internatio
nal Conferen
ce on Data Mi
ning and Advan
c
ed Computing
(
S
APIENCE)
,
2016,
pp.
91-9
5
,
doi: 10
.1109/SAPIENCE.2016.7
684167.
[13]
E.
F
ernandes
,
M
.
Holanda,
M.
V
ictorino,
V
.
Borges,
R.
C
arv
a
lho,
a
nd
G.
V
.
Erven,
“
Educational
data
m
inin
g:
Predictiv
e
an
al
ysis
o
f
academ
ic
p
erform
ance
of
public
s
chool
s
t
u
d
ents
i
n
the
capi
t
al
o
f
Bra
z
il
,”
Jo
urnal of Busines
s
Research
, vol. 9
4, pp
. 335-343
, 2019, doi:
10
.10
16/j.jbusr
e
s.201
8.02.01
2.
[14]
R
.
A
h
u
j
a
,
A
.
J
h
a
,
R
.
M
a
u
r
y
a
,
a
n
d
R
.
S
r
i
v
a
s
t
a
v
a
,
“
A
n
a
l
y
s
i
s
o
f
e
d
ucat
ional
da
ta
m
ining,”
in
Harmony Search and
Nature
Inspired Optimization Al
gorithms
, pp. 89
7
-907, 2019
, doi: 10.1007
/978-9
81-13-0761-4_85.
[15]
E.
B
.
Cos
t
a
,
B
.
F
ons
eca,
M
.
A.
S
antana
,
F
.
F
.
D
e
A
raujo
,
a
nd
J
.
R
ego,
“
Ev
aluating
the
effe
ct
iven
es
s
of
e
duc
ation
a
l
data
m
ining
techniques
for
early
pr
ediction
of
s
tudents'
a
cadem
ic
f
ailur
e
i
n
in
troductor
y
p
rogr
amming
courses,”
Computers in H
u
man Behavior
, vol. 73
, pp
. 247-
256, 2017
, doi
: 1
0
.1016/j.chb
.
201
7.01.047
.
[16]
C.
S
.
S
ilva
and
J
.
M
.
F
ons
eca,
“
Educa
tiona
l
Dat
a
M
ining:
a
l
i
t
e
r
a
ture
r
ev
iew,
”
in
Europe and MENA Cooperation
Advances in
Info
rmation
and Co
mmunication Technologies
, 201
7, pp
. 87-94
, doi: 10.1007
/978-3-
319-46568-5_9.
[17]
R
y
an
B
aker
,
“
C
hall
enges
for
th
e
future
o
f
edu
c
ation
a
l
da
ta
m
in
i
ng:
T
h
e
B
ak
er
l
earning
ana
l
yti
c
s
prizes
,
”
JEDM
|
Journal of Edu
c
ational Da
ta Mining
, vo
l
. 11, no. 1, pp. 1-17, 201
9.
[18]
R.
M
.
Awangga,
S.
F
.
Pane,
K.
T
unnisa,
and
I
.
S
upriana,
“K
m
ean
s
clustering
an
d
meanshift
analy
s
is
f
or
g
roupin
g
the
dat
a
o
f
coal
t
erm
in
P
uslitbang
Tekm
ira
,
”
TELKOMNIKA Telecommunica
ti
on Computing Electronics and
Control
, vo
l
. 16, no. 3, pp. 1351-
1357
, 2018
, doi: 10.12928/telkomnika
.v16
i3.8910
.
[19]
S.
S
later
,
S
.
Joksimović,
V
.
K
ovanovic,
R
.
S.
B
aker,
and
D.
G
as
evic,
“
T
ool
s
for
educat
iona
l
data
m
ining
:
A
review,”
Journal of
Educa
t
ional and
Behavioral Sta
tis
tics
, vol.
42,
no.
1
,
pp.
85-106.
2017,
doi: 10
.3102/10
76998616666808.
[20]
A
. Tria
y
ud
i, O
.
W
. W
id
y
a
rto
, an
d
V
. Ros
alina
, "CLG
Clus
tering
fo
r Mapping Pattern Analy
s
is of Student Acad
emic
Achievem
ent
.
"
ICIC Ex
pre
ss Le
tte
rs,
vol. 14
, no
.
12, pp
. 1225-12
34, 2020
.
[21]
Cobo
Rodríguez,
“
Parameter-fr
ee
a
gglomerativ
e
hier
arch
ical
c
lus
tering
to
m
odel
l
earn
e
rs'
a
c
tivit
y
in
o
n
line
discussion foru
ms,” Doctoral pr
ogramme, Universitat Oberta de C
atalun
ya
,
2014.
[22]
K
.
X
i
n
g
,
C
.
H
u
,
J
.
Y
u
,
X
.
C
h
e
n
g
,
a
nd
F.
Z
hang
,
“Mutual
Privacy
Preserv
i
ng
$k$
-Means
Clustering
in
S
ocial
Particip
ator
y
Sensing,”
IEEE Tr
ansactions on I
ndustrial Informatics
,
vo
l.
1
3
,
n
o.
4
,
pp
.
2066-2
076,
A
ug.
2017,
doi: 10
.1109/TII
.
2017.2695487
.
[23]
I.
W
ah
y
u
d
i
n,
T
.
Djatna,
and
W.
A
.
Kusuma,
“Cl
u
ster
a
naly
s
i
s
for
S
M
E
r
is
k
anal
ys
is
docum
ents
b
as
ed
on
P
ill
ar
K
-
M
eans
,
”
TELKO
M
NIKA Telecom
m
unication Com
puting Electronics and Control
, vol. 14
, no
. 2, pp
. 674-683, 2016,
doi: 10
.12928/telkomnika.v14i1
.
2385.
[24]
G.
G
an
a
nd
M
.
Kwok
Po
N
g,
“
K-mean
s
cluster
i
ng
with
outlier
r
emo
val,”
Pa
ttern Recognition
Letters
,
vo
l.
9
0
,
pp. 8-14
, 2017
, doi: 10
.
1016/j.patrec.2017.03
.008
.
[25]
D.
P
.
Sari,
D.
R
osadi,
A
.
R.
E
ff
endie,
a
nd
D
anardono,
“
K-means
and
Ba
yes
i
an
n
etworks
to
d
e
t
erm
i
ne
build
ing
dam
a
ge
l
ev
els
,
”
TELKOMNIK
A
Telecommunication Co
mputing Electronics
and Control
,
vol.
17,
no.
2
,
pp. 719-727
. 20
19, doi: 10
.1292
8/telkomnik
a
.v1
7
i2.11756.
Evaluation Warning : The document was created with Spire.PDF for Python.