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.
9
, No
.
5
,
Octo
ber
201
9
, pp.
3822
~
38
32
IS
S
N:
20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v
9
i
5
.
pp3822
-
38
32
3822
Journ
al h
om
e
page
:
http:
//
ia
es
core
.c
om/
journa
ls
/i
ndex.
ph
p/IJECE
Fra
m
ework fo
r c
ost
-
effe
ctive an
alytic
al m
odelling
for senso
ry
data ov
er
cloud
en
vironment
Manujak
shi B
.
C
.
1
,
K
B Ram
esh
2
1
Depa
rtment of
Com
pute
r
Scie
n
ce
and
Engi
ne
ering,
Presiden
c
y
Univer
sit
y
,
Indi
a
2
Depa
rtment of
El
e
ct
roni
cs
&
In
strum
ent
at
ion
E
ngine
er
ing, RV
Coll
ege of
Eng
i
nee
ring
,
Ind
ia
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Ja
n
7
, 201
9
Re
vised
A
pr
9
,
201
9
Accepte
d
Apr
21
, 201
9
In
orde
r
to
offe
r
sensory
da
ta
as
a
servic
e
over
t
he
cl
oud,
i
t
is
nec
essar
y
to
exe
cu
te
a
cost
-
eff
ective
and
yet
pre
ci
se
d
at
a
ana
l
y
t
ical
log
ic
withi
n
th
e
sensing
unit
s.
How
eve
r,
it
is
qui
te
questi
on
abl
e
as
such
form
s
o
f
ana
l
y
t
ic
a
l
oper
ation
a
re
q
uit
e
resourc
e
d
epe
nden
t
whi
ch
ca
nno
t
b
e
off
ere
d
b
y
th
e
resourc
e
constr
a
int
sensor
y
unit
s
.
The
re
fore
,
th
e
proposed
pape
r
i
ntroduc
es
a
novel
appr
o
ac
h
of
per
form
ing
c
ost
-
eff
ective
data
anal
y
tica
l
m
ethod
in
orde
r
to
ext
r
ac
t
knowl
edge
from
big
da
ta
over
the
c
loud
.
The
proposed
s
tud
y
uses
a
novel
con
ce
p
t
o
f
the
fr
eque
n
t
p
at
t
ern
a
long
wit
h
a
tr
ee
-
b
ase
d
a
pproa
ch
in
orde
r
to
dev
el
op
an
anal
y
tical
m
odel
for
c
arr
y
ing
out
the
m
ini
ng
oper
ation
i
n
the
l
arg
e
-
sc
ale
sensor
depl
o
y
m
ent
ove
r
th
e
c
l
oud
envi
ronm
en
t.
Us
ing
a
sim
ula
ti
on
-
base
d
appr
oa
ch
o
ver
the
m
at
hema
ti
c
a
l
m
odel
,
the
pro
posed
m
odel
exhi
bit
r
educed
m
ini
ng
dura
ti
on
,
cont
rol
le
d
en
e
rg
y
dissipation,
and
high
l
y
opti
m
iz
ed
m
emor
y
demands for a
ll
th
e
r
esourc
e
c
onstrai
nt
nodes.
Ke
yw
or
d
s
:
Cl
oud
c
om
pu
ti
ng
Data analy
ti
cs
In
te
r
net
of thin
gs
Kno
wled
ge
e
xtracti
on
Mi
nin
g
Sensor
n
et
wor
k
Tree
Copyright
©
201
9
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
:
Ma
nuj
a
ks
hi B
.
C
.
,
Assistant
Profe
sso
r
,
Dep
a
rtm
e
nt of C
om
pu
te
r
Science a
nd
E
ng
i
neer
i
ng,
Pr
esi
de
ncy
Un
i
ver
sit
y, Be
ng
al
uru, Ka
rn
at
a
ka
, In
dia
.
Em
a
il
:
m
anu
j
a
ks
hi
bc
2014@
gm
ai
l.co
m
1.
INTROD
U
CTION
The
util
iz
at
ion
of
t
he
se
ns
in
g
un
it
s
has
bee
n
witnesse
d
am
ong
t
he
com
m
ercial
us
er
s
f
rom
m
or
e
than
a
deca
de
in
the
form
of
W
i
rel
ess
Se
ns
or
Network
(
WSN)
[
1].
T
he
c
onve
ntion
al
a
ppr
oa
ch
of
WSN
cal
ls
for
perform
ing
two
basic
operati
on
s
i.e.
data
fu
sio
n
an
d
data
agg
re
gatio
n
[
2].
It
is
al
read
y
kn
ow
n
that
sens
or
app
li
cat
io
ns
ar
e
norm
ally
dep
loye
d
in
the
sc
enar
i
o
w
hich
is
hazar
dous
or
ph
ysi
cal
ly
chall
eng
in
g
f
or
hu
m
ans
to r
eac
h.
H
e
nc
e, the
r
obus
t
ne
ss in
the se
ns
in
g
pe
rfor
m
ance
reall
y
m
atters
in
WSN. Th
e
re
are v
a
rio
us
ess
entia
l
con
ce
pts
that
sta
te
a
senso
r
network
e
ncoun
te
rs
var
i
ous
fo
rm
s
of
cha
ll
eng
e
s
e.g.
se
cur
it
y
prob
le
m
s
[
3
]
,
routin
g
prob
le
m
s
[
4
]
,
traff
ic
m
anag
e
m
ent
pro
blem
[
5
]
,
and
ene
r
gy
pro
blem
s
[
6
]
.
O
ut
of
al
l
t
hese
set
of
pro
blem
s,
ener
gy
is
one
of
t
he
m
os
t
chall
eng
in
g
prob
le
m
s
in
WSN
t
o
be
encou
ntere
d
w
it
h.
It
is
belie
ve
d
that
energy
facto
r
i
s
directl
y
pr
op
or
ti
onal
ly
to
the
data
forw
a
rd
i
ng
perform
ance
in
W
SN
[
7
]
.
Ther
e
f
or
e,
t
here
has
been
va
rio
us
s
ign
ific
a
nt
rese
arch
w
orks
to
wards
a
ddressi
ng
t
he
prob
le
m
of
ene
r
gy
con
st
raint
in
W
SN
[
8
]
.
Howe
ver,
su
c
h
s
olu
ti
ons
ar
e
no
t
resea
rc
hed
en
ough
wh
e
n
WSN
i
s
integ
rated
with
va
rio
us
up
c
om
ing
app
li
cat
io
ns
over
a
reconfi
gura
ble
net
wor
k.
At
pr
e
sent,
sens
or
s
a
re
c
on
si
der
e
d
a
n
integ
ral
com
ponen
t
of
In
te
r
net
-
of
-
T
hi
ng
s
(IoT
)
w
here
sensors
are
c
onnected
with
m
ul
ti
ple
oth
er
form
s
of
netw
orkin
g
de
vices
wh
e
re
the
va
ri
ous
m
e
diu
m
of
com
m
un
ic
at
io
n
can
be
util
iz
ed
[
9
]
.
Hen
ce
,
the
c
ha
ll
eng
in
g
pr
oblem
in
this
aspect
is
that ene
rg
y
de
m
and
s
for
t
he
s
ens
or
node
s are
v
e
ry m
uch
d
i
ff
e
ren
t i
n WS
N
a
nd in I
oT [
1
0
].
At
pr
ese
nt,
the
re
are
few
res
earch
w
ork
to
wards
the
dire
ct
ion
of
s
ol
ving
energy
pro
ble
m
s
in
IoT
even,
but
the
r
e
is
again
a
pote
ntial
resear
ch
ga
p
a
nd
t
ha
t
gap
is:
exis
ti
ng
syst
em
of
Io
T
do
not
ty
pical
ly
consi
der
t
he
sta
nd
a
r
d
sens
or
node
c
onfig
urat
ion
f
or
wh
ic
h
reas
on
it
is
near
ly
va
gue
to
un
der
sta
nd
how
t
o
so
lve
the
e
ne
r
gy
pro
blem
s.
Anothe
r
as
pect
to
obser
ve
in
this
sit
uation
i
s
that
the
data
forw
a
r
ding
pro
cess
is
the
pri
m
e
reaso
n
f
or
e
nergy dep
le
ti
on. I
t
is al
so
kn
own
tha
t
Io
T
is
m
eant
for
not on
ly
ag
gr
e
gatin
g
the dat
a
bu
t
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
Framew
or
k
for
co
st
-
ef
fe
ct
iv
e
analyti
cal
mod
el
li
ng
for
sens
or
y
data
over c
loud
…
(
M
anuj
aks
hi B
.
C
.
)
3823
al
so
pe
rfo
rm
s
a
pr
e
dicti
ve
op
erati
on
on
th
e
sens
or
y
data.
Ther
e
are
m
ulti
ple
ad
van
ta
ge
s
in
this
a
sp
ect
viz.
i)
the p
r
eci
sion
a
s w
el
l as accur
acy
o
f
pre
dicti
on
i
ncr
ease
s if t
he
analy
ti
cal
o
per
at
io
n
is pe
rfor
m
ed
rig
ht af
t
er th
e
data
is
fused
within
a
se
nso
r
a
nd
be
f
or
e
th
e
f
us
e
d
data
is
f
orwarde
d
t
o
the
sin
k
,
ii
)
t
he
siz
e
of
the
a
naly
zed
data
is
al
w
ay
s
le
sser
tha
n
ag
gregate
d
data
w
hich
will
co
nsum
e
le
ss
channel
-
based
re
sources
as
well
a
s
le
sser
com
pu
ta
ti
on
al
dem
and
s
to
pr
ocess
it
.
H
ow
e
ver,
the
re
is
st
il
l
an
uns
olv
e
d
quest
io
n
e
vo
l
ved
f
or
this
process
wh
ic
h
is
how
to
desig
n
suc
h
analy
ti
cs
wh
ic
h
can
be
execu
te
d
ove
r
re
s
ource
co
ns
trai
nt
se
nsor
nodes
.
The
m
e
m
or
ie
s
of
the
sens
or
nodes
are
qu
it
e
lim
it
ed
and
so
is
the
proces
sing
ca
pab
il
it
y.
In
or
der
to
e
xecu
te
any
analy
ti
cal
protoc
ol
ove
r
a
sens
or
no
de,
the
re
is
a
nee
d
of
s
uff
ic
ie
nt
m
e
m
or
y
as
the
m
ajo
rity
of
he
predict
ive
s
chem
es
in
Io
T
are
hi
gh
ly
it
erati
ve
and
it
de
pe
nds
upon
the
scal
e
of
the
dat
a
that
are
re
qu
i
red
t
o
be
a
naly
ti
cal
ly
proces
sed
to
t
he
datace
nter
or
wa
reho
us
es.
W
it
h
the
e
voluti
on
of
big
da
ta
,
the
c
om
pl
exity
of
the
struct
ur
e
dness
of
the
da
ta
is
con
sist
e
ntly
kep
t
on
i
ncr
easi
ng.
Alt
hough
t
her
e
a
re
existi
ng
software
fr
am
ewo
r
ks
f
or
distrib
uted
st
or
a
ge,
they
ar
e
le
ast
capab
le
of
pe
rfor
m
ing
extracti
on
of
sp
eci
fic
knowl
edg
e
wh
ic
h
le
ad
s to
the em
erg
ence
of a
nov
el
m
eth
od
of a
naly
ti
c
s.
Ther
e
f
or
e,
this
pa
per
intr
oduc
es
a
novel
a
nd
cost
-
e
ff
ect
ive
arch
it
ect
ural
a
ppr
oach
bas
ed
on
f
reque
nt
patte
rn
s
in
or
de
r
to
pe
rfo
rm
an
a
naly
ti
cal
op
erati
on
over
sens
or
y
data.
The
pro
posed
syst
e
m
m
ake
s
us
e
of
gr
a
ph
-
base
d
tr
ee
structu
r
e
in
order
t
o
co
ns
tr
uct
the
propos
ed
data
a
naly
ti
cal
con
ce
pt
al
ong
with
novel
us
a
ge
of freq
ue
nt p
at
te
rn
a
nd a
dif
fe
ren
t
var
ia
nt
of
est
i
m
at
es.
This
sect
io
n
i
s
in
c
onti
nu
at
ion
of
the
re
view
of
e
xisti
ng
li
te
ratur
e
f
ro
m
our
pr
i
or
w
ork
[11].
Althou
gh,
the
re
ar
e
var
i
ous
stud
ie
s
al
rea
dy
carried
out
towards
ap
pl
yi
ng
so
phist
ic
at
ed
data
an
al
yt
ics
appr
oach
es
ov
er
the
sen
sory
data
colle
ct
ed
in
the
IoT
ec
os
yst
em
.
The
corre
la
ti
on
-
ba
s
ed
analy
sis
is
on
e
of
the
sim
plest
form
s
of
s
hortli
sti
ng
or
gr
oupi
ng
the
sim
il
ar
form
s
o
f
data
and
he
nce
co
nsi
der
e
d
as
one
of
the
cost
-
e
ff
ect
ive m
echan
ism
of
data
analy
ti
cs.
The
work
ca
rr
i
ed
out
by
Be
rt
r
and
a
nd
Mo
onen
[
12]
has
pre
sente
d
a
m
echan
is
m
to
com
pu
te
th
e
co
rr
el
at
ion
factor
on
t
he
distrib
uted
s
cal
e
us
in
g
a
tree
-
ba
sed
ap
proac
h.
The
c
om
plete
analy
sis
of
the
data
is
car
rie
d
ou
t
on
the
ba
sis
of
a
co
rr
e
la
ti
on
al
facto
r
al
so
co
ns
i
der
i
ng
the
tem
po
ral
at
trib
ute
al
ong
with
it
.
Howev
e
r,
su
c
h
f
or
m
s
of
data
al
so
have
a
higher
c
om
plexit
y.
The
wor
k
carried o
ut b
y
Parwez et
al. [13
]
whe
re th
e
com
plex
data f
ro
m
m
ob
il
e d
at
a h
as b
e
en
c
on
sidere
d
f
or
e
va
luati
ng
the anom
al
y patt
ern
ass
ociat
ed wit
h
the
d
at
a
. T
he
a
uthors h
ave
us
e
d un
s
up
erv
ise
d
le
ar
ning
cl
ust
erin
g
sc
hem
e.
A
sim
il
ar
directi
on
of
the
w
ork
has
al
s
o
be
en
car
ried
ou
t
by
Ra
h
m
an
et
al
.
[1
4]
us
in
g
tim
e
-
series
analy
sis.
Re
hm
an
et
al
.
[1
5]
hav
e
prese
nted
a
disc
us
s
ion
of
bi
g
dat
a
analy
ti
cs
wh
ere
the
a
utho
rs
ha
ve
pr
ese
nted
t
he
s
ign
ific
a
nce
of
con
ce
ntric
c
om
pu
ti
ng
ap
pro
ach
in
the
pres
ented
syst
em
.
A
sim
i
la
r
direc
ti
on
of
the
resea
rc
h
w
ork
has
bee
n
carried
out
by
Sun
et
al
.
[
16]
with
res
pec
ti
ve
to
the
dis
cussion
of
ana
ly
ti
cal
op
e
rati
ons
ove
r
the
inte
rn
et
of
t
hings
an
d
big
data
analy
ti
cs.
The
rece
nt
w
orks
of
Wang
et
al
.
[
17]
ha
ve
pr
ese
nted
a
dis
cussion
of
anal
yt
ic
al
op
erati
on
over
ene
r
gy
-
base
d
data.
Bi
g
data
ana
ly
ti
cal
app
r
oac
h
w
as
al
so
discusse
d
by
Y
ue
et
al
.
[
18]
towa
r
ds
c
on
st
ruct
ing
the
case
of
s
pecific
e
ve
nt
detect
io
n
over
s
ocial
netw
orki
ng
app
li
cat
io
n
s
pe
ci
fic to
geog
ra
ph
ic
al
lo
cat
ion.
Most
rece
ntly
,
Ram
os
et
al
.
[
19
]
ha
ve
pr
e
se
nted
a
wor
k
w
her
e
se
ns
ory
da
ta
has
bee
n
e
xtracted
f
ro
m
the
inter
face
of
t
he
m
ixed
sign
al
us
in
g
non
-
li
near
qua
nt
iz
at
ion
m
et
ho
d.
Sh
a
rm
a
and
Wang
[20]
hav
e
dev
el
op
e
d
a
fra
m
ewo
r
k
t
hat
can
perf
or
m
m
ining
op
e
rati
on
on
e
dge
a
nd
c
loud
over
it
s
re
sp
ect
ive
data.
WSN,
wh
e
n
inte
gr
at
e
d
with
I
oT
,
of
fer
s
m
assive
pr
oces
sin
g
of
di
ff
e
ren
t
f
or
m
s
of
data.
H
ow
e
ver,
if
the
data
bear
s
higher
siz
e
a
nd
c
om
plexity
(e.g
.
m
ultim
edia
data)
tha
n
processin
g
it
al
ong
with
str
uctur
iz
at
io
n
becom
es
the
m
os
t
co
m
plex
ta
sk
.
T
his
pro
bl
e
m
was
addre
ssed
by
Ca
o
et
al
.
[
21
]
w
her
e
a
un
i
qu
e
a
nal
yt
ic
al
op
erati
on
ha
s
been
ca
rr
ie
d
out
us
i
ng
a
sel
f
-
opti
m
iz
ing
ap
proac
h
over
m
ultim
edia
con
t
ents.
T
he
opti
m
iz
at
ion
was
carried
ou
t
on
the
ba
sis
of
c
on
te
xt
in
order
to
e
ns
ure
t
hat
it
offe
rs
resou
rce
fr
ie
ndly
ope
rati
on
a
ppli
cable
f
or
a
pr
act
ic
a
l en
vir
on
m
ental
sit
uation. Ba
sed
on t
he
c
on
te
xt,
the
m
od
el
p
r
oto
ty
pi
ng
is ca
rr
ie
d o
ut that is ca
pabl
e o
f
fine
-
t
un
i
ng
th
e
ene
rg
y
dem
ands
of
the
s
ens
or
s
in
t
he
Io
T
en
vir
onm
ent.
Howe
ver,
the
sel
f
-
op
ti
m
iz
at
ion
pr
i
nciple
do
es
n'
t
of
fe
r
c
os
t
-
e
ff
ect
ive
c
om
plex
a
nd
unstr
uc
ture
d
data
ha
ndli
ng
m
echan
ism
apar
t
from
vide
o
data.
S
uch
a pr
ob
le
m
was
al
so
in
vestigat
ed
by
oth
e
r
re
sear
cher
s where
a syst
e
m
-
on
-
c
hi
p
is
basical
ly
de
sign
e
d
in
or
der
t
o
s
up
port
the
com
plex
processi
ng
of
data
a
nd
othe
r
heav
y
al
gori
thm
s.
The
a
uthors
h
a
ve
al
s
o
use
d
a
m
achine
le
arn
i
ng
a
ppr
oach
f
or
carryin
g
ou
t
f
eat
ur
e
e
xtracti
on
a
nd
trai
ning
.
a
pa
rt
from
thi
s
fog
com
pu
ti
ng
wa
s
al
so
fou
nd
to
be
in
vo
l
ved
in
perform
ing
an
analy
ti
cal
operati
on
in
the
e
xisti
ng
s
yst
e
m
.
A
un
i
que
fog
com
pu
ti
ng
bas
ed
m
od
el
ing
w
as
c
arr
ie
d
out
us
in
g
Ra
spbe
rry
Pi
in
order
t
o
assess
the
perf
or
m
ance
of
dif
fer
e
nt
data
analy
ti
cal
m
od
el
(
He
et
al
.
[
22
]
)
.
T
he
syst
e
m
assist
s
in
perform
ing
la
rg
e
scal
e
of
analy
sis
w
ork
over
m
ul
ti
ple
wo
r
kst
at
ion
s
in
orde
r
to
offe
r
a
gr
eat
er
deal
of
scal
abili
ty
.
The
stud
y
outc
om
e
was
f
ound
to
offe
r
reducin
g
jo
b
com
pu
ta
ti
on
ti
m
e
with
increasing
c
om
plexity
.
It
is
al
so
cl
aim
ed
to
of
f
er
bette
r
ser
vice
an
d
resou
rce m
anag
em
ent.
Applyi
ng
a
nal
yt
ic
-
based
a
ppro
ac
h
ha
s
al
so
been
ca
rr
ie
d
out
over
se
ns
it
ive
data
e.
g.
he
al
thca
re
data
.
Unfortu
natel
y,
su
c
h
data
gro
ws
so
m
ass
ively
that
ret
ention
of
t
hei
r
pr
iv
acy
is
an
utm
os
t
co
ncern.
Applyi
ng
nor
m
al
secur
it
y
al
gorithm
s
cannot
be
ca
rr
ie
d
ou
t
owin
g
t
o
t
he
c
o
m
plica
ti
o
ns
as
so
ci
at
ed
with
it
.
A
li
ght
-
weig
ht
analy
ti
cal
ap
proac
h
has
be
en
pr
es
ente
d
by
G
ong
et
al
.
[
23
]
us
i
ng
a
predict
ive
ap
proac
h.
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.
9
, N
o.
5
,
Oct
ober
201
9
:
3
8
2
2
-
3
8
3
2
3824
The
stu
dy
ha
s
encode
d
pri
va
cy
to
the
trai
ni
ng
data
w
hile
us
e
d
re
gr
es
sio
n
m
od
el
for
pe
rfor
m
ing
pre
di
ct
ion
ov
e
r
par
ti
ti
on
e
d
data
.
T
he
stu
dy
al
so
s
howe
d
that
a
naly
ti
cs
can
be
highly
helpful
f
or
le
ver
a
ging
the
secur
it
y
sta
nd
a
rds
to
o
wh
ic
h
is
on
e
of
the
tr
oubles
hootin
g
pro
blem
s
ov
er
the
di
stribu
te
d
a
nd
dy
nam
ic
ecosyst
e
m
of
the
cl
oud
e
nv
i
ronm
ent.
Fu
rt
he
r
stu
dies
towa
rd
s
healt
hca
re
data
a
re
car
ried
out
by
Li
e
t
al
.
[2
4]
w
hic
h
has
discusse
d
t
he
e
xtracti
on
of
m
edical
inf
or
m
ation
within
t
he
a
m
bu
la
tory
ve
hicle
s.
T
he
m
od
el
de
velo
pe
d
offer
s
consi
ste
nt
m
onit
or
in
g
of
the
he
al
th
factor
of
the
dr
i
ver
in
a
veh
ic
le
.
Re
al
-
ti
m
e
senso
r
-
ba
s
ed
data
extract
i
on
of
healt
h
sta
ti
sti
cs
has
be
en
ca
rri
ed
out
in
orde
r
to
assess
t
he
even
t.
Stu
dy
towa
r
ds
predict
ive
schem
e
has
bee
n
al
so
prom
oted
by
Yildirim
et
al
.
[25]
co
ns
id
erin
g
the
case st
ud
y
of
wi
nd
tur
bin
es
.
Th
e
auth
or
s h
a
ve
pr
esented
an
integ
rated
f
ram
ewo
rk
t
hat
is
respo
ns
ible
for
pe
rfo
rm
in
g
m
ai
ntenan
ce
ov
e
r
the
data
colle
ct
ed
f
ro
m
wind
far
m
s.
The
pa
per
ha
s
al
so
pr
e
sente
d
a
scheduli
ng
fr
am
ewo
r
k
f
or
assessi
ng
the
per
f
or
m
ance
of
dynam
ic
m
od
el
s.
Usage
of
ope
n
source
i
n
fra
m
ing
up
data
analy
ti
cs
was
repor
te
d
to
be
us
e
d
for
de
velo
ping
a
m
on
it
or
ing
m
od
ule
f
or
re
al
-
ti
m
e
da
m
age
with
resp
ect
to
the
struct
ur
al
he
al
th
aspect.
Lia
o
et
al
.
[2
6]
hav
e
dev
el
op
e
d
a
m
od
el
f
or
data
a
naly
ti
cs
that
is
capa
ble
of
ext
racti
ng
intel
li
ge
nce
from
infrast
ru
ct
ure
al
on
g
with
powe
r
ef
fici
en
cy
i
n
it
s
intern
al
op
e
rati
on.
T
he
aut
hors
have
us
e
d
real
-
ti
m
e
sensors
in
order
t
o
pe
rfo
r
m
the
exp
e
rim
ent
wh
ere
the
outc
om
e
is
assessed
by
com
pr
ession
a
nd
delay
m
a
inly
.
M
unoz
et
al
.
[27]
hav
e
pr
ese
nted
a
joint
m
od
el
ing
of
a
naly
ti
cs
us
in
g
inter
net
of
thi
ng,
s
oft
war
e
-
de
fine
d
netw
ork
ov
e
r
cl
oud
env
i
ronm
ent.
An
e
xperim
ental
appro
ac
h
ha
s
bee
n
prese
nt
ed
by
Munoz
e
t
al
.
[27]
w
hic
h
is
m
ai
nly
m
e
ant
f
or
con
t
ro
ll
in
g
bot
tl
eneck
co
ndit
ion
within
t
he
traf
fic
scena
rio.
Dep
l
oym
ent
of
the
a
naly
ti
cal
op
e
rati
on
was
al
so
sh
ow
n
to
a
ssist
s
in
so
lvin
g
cl
assifi
cat
ion
pro
blem
s
(O
te
ro
et
al
.
[28]).
The
stud
y
al
s
o
presents
a
uniq
ue
decisi
on
-
m
aking
syst
em
fo
r
perform
ing
predict
ion
ov
e
r
cl
oud
en
vir
on
m
ent.
An
al
ysi
s
of
t
he
traf
fic
-
relat
ed
data
was
al
so
carried
out
by
Sh
a
o
et
al
.
[29]
offe
rin
g
s
olu
ti
on
to
wards
bo
tt
le
nec
k
tra
f
fic
sit
uation.
S
tud
ie
s
towa
rd
s
ad
va
nc
ed
a
naly
ti
cs
ha
ve
been
car
ried
ou
t
by
I
va
nov
et
al
.
[30]
w
her
e
preci
sio
n
far
m
ing
is
a
dvocated
for
pe
rfo
rm
ing
an
analy
ti
cal
op
e
rati
on
in
W
SN
.
C
hand
rakal
a
and
Ra
o
[
31]
ha
ve
prese
nt
ed
m
igrat
ion
of
VM
to
en
ha
nce
the
secu
rity
of
th
e
cl
oud
c
om
pu
ti
ng.
Rhi
ou
i
and
O
um
nad
[
32
]
hav
e
prese
nted
IoTs
s
urv
ey
for
cal
culat
ing
hum
an
act
ivit
ie
s
from
all
ov
er
.
Sindh
u
et
al
.
[
33
]
ha
ve
pr
e
se
nted
a
new
i
nc
orp
or
at
ed
struc
ture
to
m
ake
su
re
s
uperior
data
qual
it
y
i
n
big
data
analy
ti
cs
on
cl
oud
s
urrou
nd
i
ngs.
T
he
pa
per
has
al
so
pr
ese
nted
a
m
echan
ism
of
so
phist
ic
at
ed
f
arm
m
anag
em
ent
an
d
disc
usse
d
va
rio
us
pr
act
ic
al
chall
eng
es
ass
ociat
ed
with
it
.
The
aut
hors
ha
ve
presente
d
a
prototypin
g
appr
oach
f
or
exp
e
rim
enting
their
con
ce
pt.
Ther
e
fore,
th
ere
is
var
i
ou
s
wor
k
that
has
bee
n
carried
ou
t
to
wards
de
velo
pi
ng
analy
ti
cs
ov
e
r
the
cl
ou
d.
The
nex
t
sect
ion
discusse
s the
problem
that is i
den
ti
fie
d
f
r
om
the r
e
view
.
We
ha
ve
pe
rfor
m
ed
a
ro
ug
h
inv
est
igati
on
of
the
m
ining
te
chn
iq
ues
im
plem
ented
exclusively
for
the
wireless
se
ns
or n
et
w
ork.
It
was
ex
plored
that
the
rece
nt
tren
d
of
kn
ow
l
edg
e
e
xtracti
on
is
m
or
e
or
le
ss
int
o
the
ide
ntific
at
ion
of
a
ny
un
i
qu
e
patte
r
n
from
the
com
plex
data.
How
ever,
with
ov
e
rco
m
ing
of
t
he
data
com
plexity
over
c
lo
ud
usi
ng
SD
aaS
,
the
an
al
ysi
s
cou
ld
be
done
m
or
e
e
ffec
ti
vely
with
le
sser
respo
ns
e
tim
e.
Howe
ver,
the
pro
blem
s
ti
l
l
resides
as
e
xisti
ng
te
ch
niques
cal
l
fo
r
obser
ving
only
the
fr
e
qu
e
nt
patte
r
n
f
ro
m
the
occ
urre
nce
of
a
n
e
ve
nt.
The
m
eaning
of
t
his
occ
urre
nce
is
-
t
hat
al
tho
ug
h
se
ns
or
se
ns
e
al
l
the
data
bu
t
it
on
ly
f
orwards
the
data
w
hich
has
sig
nifican
t
inform
ation
of
occurr
ence
of
a
n
eve
nt.
It
is
done
in
order
t
o
avo
i
d
the
c
omm
un
ic
at
ion
ov
erh
ea
d.
Th
e
si
gn
i
ficant
resea
rch
ga
p
is
-
ti
ll
date
the
f
re
qu
e
nt
patte
r
n
ap
pr
oach
e
s
are
ne
ve
r
acc
urat
e
f
or
e
xplo
r
ing
t
he
po
te
ntial
patte
rn
of
da
ta
as
it
can
only
highli
ght
the
e
po
c
hs
pres
ent
in
the
data
base
t
ha
t
con
ta
in
s
the
fr
e
qu
e
nt
patte
r
ns
.
Su
c
h
pro
ble
m
s
are
ve
ry
da
ng
e
r
ou
s
f
or
he
al
thcare
a
pp
li
cat
ion
as
well
as
a
n
a
pp
li
cat
io
n
t
hat
m
on
it
or
s
the
cl
i
m
atic
conditi
on
,
espe
ci
al
ly
w
her
e
s
ens
ors
are
us
e
d
to
m
on
it
or
the
healt
h
sta
ti
sti
cs.
Finall
y,
s
uch
inc
om
plete
knowle
dge
ge
ts
accum
ulated
in
Ha
doop
t
ha
t
le
ads
to
un
w
anted
cost
a
nd
ex
pe
ndit
ur
e
of
cl
ou
d
ser
vices
to
sto
rag
e
an
d
perf
orm
err
o
r
-
pron
e
analy
sis
of
dat
a.
It
is
e
xplo
re
d
that
su
c
h prob
le
m
s lead to f
ollow
i
ng issues:
−
Li
m
it
a
ti
on
of
Cl
oud
-
base
d
Con
t
ro
l:
It
is
no
t
po
ssible
f
or
a
cl
ou
d
to
trac
k
al
l
the
sensor
data
from
one
po
i
nt
of
the
ba
se
sta
ti
on
.
He
nc
e,
if
the
track
ing
of
data
is
done
f
ro
m
m
ul
ti
ple
po
ints
of
the
base
sta
ti
on
that
fr
eq
ue
nt
pa
tt
ern
s
can
be
extracte
d
bu
t
the
sign
i
ficant
patte
rn
ca
nnot
be
extracte
d
if
the
la
rg
e
netw
ork
with it
s
dynam
ic
it
y i
s n
ot con
sidere
d.
−
In
c
om
patibil
ity
of
e
xisti
ng
m
ining
:
I
f
se
nsors
colla
bo
rate
with
cl
o
ud
a
r
chite
ct
ur
e
t
hat
it
is
essenti
al
that
knowle
dge
dis
cov
e
ry
(
or
m
ining)
of
t
he
data
al
so
t
o
be
done
in
s
am
e
way.
U
nfor
t
un
at
el
y,
the
existi
ng
m
ining
ap
proa
ches
w
ere
never
te
sti
fied
f
or
energy
co
nsu
m
pt
ion
.
It
is
sti
ll
un
so
l
ved.
More
ov
e
r,
t
he
big
qu
e
sti
on
m
ark
is
non
-
ap
plica
bili
ty
of
c
onve
ntion
al
m
ining
te
chn
i
qu
e
if
t
he
se
nsory
dat
a
is
m
assive
an
d
highly
unstr
uctur
e
d.
−
The
s
plit
opin
ion
of
tree
-
ba
sed
a
ppr
oach
:
Stu
dies
to
wa
rd
s
tree
-
base
d
ap
proac
hes
a
re
le
ss
a
nd
m
or
e
crit
ic
iz
ed
irres
pecti
ve
of
it
s
ben
e
fits.
Ti
ll
date
tree
-
bas
ed
ap
proac
h
was
util
iz
ed
only
fo
r
routin
g,
howe
ver,
it
can
be
al
so
util
iz
ed
f
or
co
ns
tr
ucting
the
t
opology
f
or
d
at
a
analy
ti
cs
too
,
wh
ic
h
is
not
the m
uch
-
re
sea
rch
e
d
t
op
ic
.
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
Framew
or
k
for
co
st
-
ef
fe
ct
iv
e
analyti
cal
mod
el
li
ng
for
sens
or
y
data
over c
loud
…
(
M
anuj
aks
hi B
.
C
.
)
3825
−
More
I
ncline
d
towards
sp
e
ci
fic
even
t
proc
essing;
Ma
jorit
y
of
the
exis
ti
ng
r
esearc
h
w
ork
is
essenti
al
ly
fo
c
us
e
d
on
s
pe
ci
fic
f
or
m
s
of
e
ven
t
detect
ion
an
d
does
n’t
co
ns
i
der
th
e
ex
pone
ntial
higher
le
vel
of
chall
enges
ass
ociat
ed
with
th
e
data.
Hen
ce
,
the
existi
ng
str
uctu
re
of
data
cannot
be
di
re
ct
ly
su
bj
ect
e
d
to
pr
ese
nt
data a
na
l
yt
ic
al
ap
pr
oa
ch.
Ther
e
f
or
e,
the
pro
blem
sta
t
e
m
ent
is
""
T
o
dev
el
op
a
s
i
m
ple
fr
am
ewo
r
k
for
ca
rr
yi
ng
out
s
ophist
ic
at
ed
analy
ti
cal
op
er
at
ion
ov
e
r
a
distribu
te
d
cl
ou
d
en
vironm
ent."
In
orde
r
to
ad
dr
ess
this
prob
l
e
m
,
it
is
necessary
t
o
evo
l
ve up wit
h a de
finiti
ve
a
r
chite
ct
ur
e t
hat
is discu
ssed
in t
he
ne
xt secti
on.
2.
SY
STE
M AR
CHI
TE
CT
U
R
E
The
pro
pose
d
stud
y
is
a
con
t
inu
at
io
n
of
our
pr
io
r
im
ple
m
e
ntati
on
of
[
3
4
]
and
[
3
5
]
.
This
fr
am
ewor
k
will
be
m
or
e
f
ocused
to
ward
s
extracti
on
of
know
le
dg
e
disco
ver
y
from
t
he
com
plex
s
ensory
data.
N
ow
that
after
i
m
ple
m
e
ntati
on
of
our
pr
i
or
prot
otyp
e
SD
aaS,
a
f
ra
m
ewo
r
k
is
pr
e
sented
that
ca
n
act
ually
generate
a
stream
of
se
nsor
data
a
nd
use
HBase
to
perform
pr
oper
m
anag
em
ent
of
structu
re
d
data.
T
he
fr
am
ework
al
s
o
has
a
us
er
-
cen
tric
analy
t
ic
al
m
od
ule
in
SDa
aS.
H
oweve
r,
this
par
t
of
t
he
stud
y
will
be
the
co
ntin
uation
of
SD
aaS
an
d
will
inco
rpor
at
e
s
om
e
of
the
nove
l
and
pote
ntia
l
m
ining
fe
at
ures
f
or
a
bette
r
degree
of
kn
owle
dg
e
extracti
on.
This
pa
rt
of
th
e
stud
y
will
con
si
der
a
dopti
ng
a
n
a
naly
ti
cal
appro
ac
h
c
om
ple
te
ly
.
Fig
ure
1
s
hows
the
ind
ic
at
ive
schem
e
to
be
adopted
.
The
pr
im
e
ob
j
ect
iv
e
will
be
to
extract
a
s
ign
ific
ant
patte
rn
tha
t
el
icits
the
hidden
relat
ion
s
hip
of
pat
te
rn
s
am
on
g
th
e
data
colle
ct
ed
from
senso
rs
.
The
m
at
he
m
a
ti
cal
m
od
el
ing
wi
ll
com
pr
ise
va
riou
s
var
ia
bles
e.g
.
the
num
ber
of
se
ns
ors,
tim
e
slots,
f
orm
ula
t
ion
s
of
f
reque
nt
patte
r
ns
,
et
c
.
We
will
al
so
f
or
m
ulate
a
con
diti
on
of
an
e
poch
for
analy
z
ing
the
f
re
qu
e
nt
patte
rns.
Th
e
ou
tc
om
e
of
SD
aa
S
fr
am
ewo
r
k
will
be
co
ns
i
der
e
d
as
an
i
nput
for
this
par
t
of
th
e
m
ining
te
c
hniqu
e.
He
nce,
it
is
just
an
exte
ns
io
n
of SD
a
aS
fr
am
ewor
k
f
or e
nh
a
ncin
g
m
ining
of
know
le
dge f
r
om
co
m
plex
se
ns
ory
data.
S
D
a
a
S
F
r
a
m
e
w
o
r
k
S
d
1
S
d
2
S
d
n
S
d
3
-
-
-
-
-
T
1
T
2
T
3
T
n
A
p
p
l
y
G
r
a
p
h
T
e
c
h
n
i
q
u
e
t
o
g
e
n
e
r
a
t
e
T
r
e
e
T
r
e
e
m
a
n
a
g
e
m
e
n
t
B
r
a
n
c
h
S
o
r
t
i
n
g
E
x
t
r
a
c
t
c
a
n
d
i
d
a
t
e
F
P
F
i
n
a
l
K
n
o
w
l
e
d
g
e
e
x
t
r
a
c
t
i
o
n
Figure
1 A
dopt
ed
m
et
ho
dolo
gy
A
novel
grap
h
theo
ry
has
been
dev
el
op
e
d
in
orde
r
to
con
str
uct
the
tree
fo
r
eve
r
y
data
being
gen
e
rated
by
the
S
DaaS
f
ra
m
ewo
r
k.
T
he
aim
is
to
ex
plore
the
relat
i
on
s
hip
am
ong
the
tree
br
a
nch
e
s.
The
ge
ner
at
e
d
tree
(T
1
,
T
2
,
…
..
T
n
)
will
be
s
ubj
ect
ed
to
f
urt
her
t
wo
ope
rati
on
s
i.e.
i
)
tree
m
anag
em
ent
and
ii
)
br
a
nc
h
so
rtin
g.
Tree
m
anag
e
m
ent
is
a
ll
abo
ut
arr
a
ng
i
ng
th
e
tree
structur
e
fo
r
facil
it
at
ing
traver
sal
s
ope
rati
ons
in
the
tree
w
hi
le
br
a
nc
h
s
or
ti
ng
will
pe
rtai
n
to
s
or
ti
ng
op
e
rati
on
so
that
t
he
gen
e
rated
t
ree
co
ul
d
be
e
asi
ly
su
bject
e
d
to
t
he
m
ining
op
e
ra
ti
on.
A
no
ve
l
al
go
rit
hm
wi
ll
be
de
velo
pe
d
that
ca
n
a
pply
a
patte
rn
m
inin
g
appr
oach
i
n
or
der
t
o
extr
act
al
l
cand
idate
f
re
qu
e
nt
patte
r
ns
from
the
entire
tree.
Th
e
ad
va
ntage
of
t
his
pr
ocess
will
be
that
it
pr
ese
nts
a
te
ch
nique
of
sop
hi
sti
cat
ed
data
m
ining
te
c
hniq
ue
f
or
a
ver
y
l
arg
e
a
rea
co
nsi
der
in
g
the
com
plexity
fo
r
both
hom
og
e
ne
ous
an
d
heter
og
e
ne
ou
s
ty
pes
of
the
ne
twork
.
W
e
wil
l
al
so
inv
est
iga
te
the
po
s
sibil
it
ie
s
of
inco
rpor
at
in
g
distrib
uted
m
i
ning
ap
proac
he
s
as
well
as
par
al
le
l
m
ining
ap
proac
hes
so
that
sen
s
or
data
co
uld
be
truly
use
d
as
a
cl
oud
s
erv
ic
e
di
rectl
y.
The
disc
us
sio
n
of
ad
opte
d
m
et
hodo
l
og
y
is
f
ur
t
he
r
su
pp
or
te
d by r
at
ion
al
e a
nd it
s r
es
pecti
ve
c
on
tribu
ti
on:
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
9
, N
o.
5
,
Oct
ober
201
9
:
3
8
2
2
-
3
8
3
2
3826
a.
T
he rat
io
nale of the
ad
opte
d
m
et
ho
dolo
gy
It
is
qu
it
e
e
vid
ent
th
ere
sim
il
ar
even
t
-
base
d
in
form
at
ion
can
be
capt
ured
by
dif
fe
re
nt
f
or
m
s
of
sensing
de
vice
wh
ic
h
will
im
po
se
a
big
ge
r
chall
en
ge
in
processi
ng
the
analy
ti
cal
op
e
rati
on
as
the
da
ta
i
s
abs
olu
te
ly
no
t
un
i
qu
e
a
nd
is
highly
redu
nd
a
nt.
S
uch
set
of
equ
i
valent
in
form
ation
am
on
g
the
sensi
ng
de
vices
or
set
of
su
c
h
dev
ic
es
can
be
easi
ly
represe
nted
by
fr
e
que
nt
patte
r
ns
w
hi
ch
a
re
highly
e
ssentia
l
f
or
the
real
-
world
analy
sis
of
bigger
data
stream
.
Techni
cal
ly
,
su
ch
f
orm
s
of
fr
e
que
nt
patte
r
ns
e
xtra
ct
a
hi
gh
e
r
e
xt
ent
of
the
tem
po
ral
r
el
at
ion
sh
i
p
ex
i
sti
ng
am
on
g
di
ff
ere
nt
sensi
ng
unit
s
in
the
Io
T
e
nv
i
ronm
e
nt.
He
nce,
if
there
is
any
sign
ific
a
nt
even
t
than
al
l
the
inf
or
m
at
ion
of
the
c
onne
ct
ed
sensing
unit
s
can
be
ext
racted
an
d
the
y
can
be
util
iz
ed
for
pe
r
form
ing
certai
n
act
uatio
n
f
orm
s
of
ta
sk
f
or
the
s
ens
or
s
.
S
uch
i
nfor
m
at
ion
obta
ine
d
is
hi
gh
ly
i
m
po
rtant
i
n
orde
r
to
pe
rform
m
anag
em
en
t
of
t
he
I
oT
r
eso
ur
ces
.
H
ow
ever,
at
prese
nt
,
there
a
re
no
su
c
h
appr
oach
es
e
ve
r
ev
olv
e
d
in
order
t
o
extrac
t
su
ch
un
i
qu
e
form
s
of
fr
e
quent
patte
r
ns
f
r
om
the
data
st
ream
of
the
sen
sin
g
un
i
ts.
A
par
t
from
this,
it
is
a
hi
ghly
com
pu
ta
ti
on
al
ly
exp
e
ns
i
ve
ta
sk
in
or
der
to
ext
ract
su
c
h
form
s
of
un
i
qu
e
f
re
quent
patte
r
ns
f
ro
m
the
hig
he
r
nu
m
ber
of
se
ns
in
g
unit
s
over
the
cl
oud
en
vir
on
m
ent.
Th
eref
or
e
,
these
m
e
tho
do
log
y
offe
rs
a
highly
distr
ib
ut
ed
and
ye
t
w
el
l
-
synch
roniz
ed
co
nn
e
ct
ivit
y
a
m
on
g
the
s
ensin
g
nodes
in or
de
r t
o
ext
ract esse
ntial
m
ined
in
f
or
m
at
ion
.
b.
T
he
c
on
tri
buti
on of t
he
a
dopte
d
m
et
ho
do
log
y
Fo
ll
owin
g
a
re t
he
c
on
t
rib
ution o
f
the
prop
os
e
d
syst
em
:
−
A
un
i
que
form
of
fr
e
quent
-
pa
tt
ern
base
d
appr
oach
is
intr
oduce
d
to
est
ablish
the
pote
nt
ia
l
relation
sh
i
p
a
m
on
g t
he n
odes as
well
as th
ei
r
res
pecti
ve d
at
a.
−
A
uniq
ue
tree
structu
re
has
been
pr
ese
nt
ed
that
is
cap
able
of
form
ulati
ng
a
un
i
que
com
m
un
ic
at
i
on
process
go
vern
ing
a
uniq
ue fl
ow of a
nal
yt
ic
al
d
at
a w
it
hin
WSN.
−
A
un
i
qu
e
an
d
novel
a
naly
ti
c
al
m
od
el
is
int
rod
uced
t
hat
pe
rfor
m
s
the
co
m
pu
ta
ti
on
of
di
ff
ere
nt
f
orm
s
of
sens
or
y est
im
a
te
s f
or
bette
r gra
nu
la
rity
in
the
m
ining
proces
s.
−
The
pro
pose
d
syst
e
m
has
al
s
o
a
suppo
rtabil
it
y
of
sing
le
-
hop
as
wel
l
as
m
ul
ti
-
hop
r
ou
t
ing
ope
rati
on
a
nd
hen
ce
is sig
nifi
cantl
y uti
li
zed
in the p
resen
t
-
day ap
plica
ti
on of
WSN.
−
A
highly
si
m
plifie
d
com
pu
ta
ti
on
al
m
od
el
in
g
is
carried
out
wh
ic
h
offers
bette
r
te
chn
i
cal
ado
ptio
n
in
pr
ese
nce
of
ne
ar r
eal
-
w
or
l
d
s
ensin
g dem
ands.
Th
ere
f
or
e,
the
pro
po
se
d
syst
e
m
is
m
eant
fo
r
addressi
ng
t
he
pro
blem
associat
ed
with
pe
rfo
rm
ing
an
a
naly
ti
cal
op
e
rati
on
ov
e
r t
he
s
ophisti
cat
ed
a
nd co
m
plex
se
nsory
data
ov
e
r
t
he
cl
ou
d env
i
ronm
ent.
3.
MO
DEL D
IS
CUSSIO
N
The
com
plete
analy
ti
cal
pr
oc
ess
of
t
he
pro
pose
d
syst
e
m
is
desig
ned
on
the
basis
of
a
novel
m
od
el
that
us
es
the
con
ce
pt
of
fr
e
quent
patte
r
ns
a
s
the
backb
one.
The
m
od
el
harnesses
the
po
te
ntial
con
c
ept
of
si
m
plici
t
y
in
us
ing
fr
e
qu
e
nt
pa
tt
ern
s
an
d
a
ddresses
the
scal
abili
ty
pr
oble
m
associat
ed
with
i
t
by
introdu
ci
ng
a
un
i
qu
e
c
oncep
t
of
ep
oc
h
m
a
nag
em
ent
and
tree
-
base
d
m
i
ning.
T
he
co
r
e
ideology
of
the
pro
posed
m
od
el
desig
n
is
to
en
su
re
the
i
ntrod
uction
of
the
tree
-
base
d
str
uc
ture
for
the
to
po
l
og
y
c
onstr
uc
ti
on
in
su
c
h
a
way
that
it
do
esn’
t
on
ly
assist
s
i
n
routin
g
data
but
it
of
fer
s
m
or
e
gr
a
nula
riti
es
in
fo
r
wardi
ng
m
or
e
err
or
-
f
re
e
data.
The
propose
d
m
od
el
al
so
off
ers
sig
nificant
ben
e
fits
towa
r
ds
e
xtracti
ng
a
ll
po
ssible
for
m
s
of
relat
ion
s
hip
j
ust
by u
si
ng the
no
vel idea
of fre
quent
patte
r
a
pp
ro
ac
h.
a.
M
od
e
l
par
a
m
et
ers
The
pro
posed
m
od
el
con
side
rs
that
there
ar
e
M
set
of
senso
ry
m
otes
wh
ere
M
={
m
1
,
m
2
,
….,
m
n
},
wh
e
re
n
re
pre
sents
a
total
nu
m
ber
of
se
nsors
.
As
the
se
ns
or
pe
rfor
m
data
colle
ct
ion
on
the
basis
of
it
s
pr
e
def
i
ned
ti
m
e
slots
t
(t
1
,
t
2
,
….t
a
)
s
o
t
he
st
ud
y
c
onside
rs
t
hat
ef
fecti
ve
c
urren
t
ti
m
e
(d
iffe
ren
ce
of
t
s+1
and
t
s
,
wh
e
re
s
ϵ
[1,
a
-
1]
)
is
em
pirical
l
y
represe
nted
a
s
δ, wh
ic
h
a
is t
he
siz
e o
f
the
tim
e
slot.
The
prop
os
ed
m
od
el
al
so
represe
nts
patt
ern
α
as
a
set
of
sp
eci
fic
k
num
ber
of
se
nsors
i.e.
α=
{m
1
,
m
2
,
….
m
k
}.
A
tu
ple
β
(β
t
,
γ
)
is
a
represe
ntati
on
of
an
ep
oc
h
w
hich
is
re
qu
i
re
d
for
c
onstruct
ing
a
se
ns
ory
da
ta
reposit
ory
m
at
rix
db
.
The
refor
e
,
this
m
a
trix
db
is
a
colle
ct
ion
of
t
he
finite
num
ber
of
ep
oc
h
consi
der
i
ng
γ
as
a
sp
eci
fic
pa
tt
ern
co
rr
es
po
nd
i
ng
to
the
e
ve
nt
th
at
has
been
cap
ture
d
by
t
he
se
ns
or
within
t
he
def
i
ned
tim
e
slot.
The
va
riab
le
β
t
is
con
si
de
red
a
s
the
tim
e
slot
of
an
eve
nt.
Th
e
m
od
el
ing
as
pects
c
on
si
der
to
car
ry
out
a
n
analy
ti
cal
op
e
rati
on
us
i
ng
f
r
equ
e
nt
patte
rn
s
co
nce
pt
ap
plica
ble
over
s
op
histi
cat
ed
data
in
distr
ibu
te
d
se
ns
ory
app
li
cat
io
ns
.
T
her
e
fore,
a
c
onditi
on
is set
where
a s
upporta
bili
ty
o
f
a s
pecific
patte
rn
σ
is
on
ly
c
on
si
der
e
d
t
o be
v
al
id
by a
n
e
poch
β
(β
t
, γ) if
,
σ
γ
(1)
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
Framew
or
k
for
co
st
-
ef
fe
ct
iv
e
analyti
cal
mod
el
li
ng
for
sens
or
y
data
over c
loud
…
(
M
anuj
aks
hi B
.
C
.
)
3827
The
occurre
nc
e
of
the
sp
eci
f
ic
patte
rn
σ
over
t
he
re
po
sit
or
y
db
is
a
nu
m
ber
of
the
e
po
c
hs
th
at
are
found
within
db
it
sel
f
. I
t ca
n be m
ath
em
atical
ly
r
e
pr
ese
nted
as,
f
(σ,
db)=|{
β
(β
t
,
γ)
| σ
γ |}
(2)
Fo
r
bette
r
prec
isi
on
,
the
c
ond
it
ion
al
log
ic
is
desig
ne
d
to
hi
ghli
gh
ts
that
a
s
pecific
patte
r
n
σ
can
be
on
ly
cal
le
d
as a fre
qu
e
nt
pa
tt
ern
if
f
(σ,
db
)>m
ini
m
al
cu
t
-
off
s
upport.
b.
Im
ple
m
entat
ion
strat
e
gy
The
c
om
plete
i
m
ple
m
entat
io
n
strat
egy
is
a
ll
abo
ut
c
onstr
ucting
a
nove
l
tree
for
a
give
n
se
ns
ory
reposit
ory
db
a
fter
read
i
ng
al
l
the
data
within
it
in
one
s
hot
.
T
he
im
ple
m
entat
ion
strat
e
gy
al
so
co
ns
i
de
rs
that
there
is
a
no
n
-
redu
nd
a
nt
r
ou
te
betwee
n
two
no
de
s
t
hat
init
ia
ti
ng
from
the
root
node
of
t
he
tr
ee.
The
c
um
ulati
ve
inf
or
m
at
ion
of
t
he
se
nsory
nodes
co
rr
e
spondin
g
t
o
al
l
r
eposi
tory
db
c
an
be
e
xtracte
d
f
r
o
m
this
tree.
T
he
weig
hted
s
en
so
ry
est
im
at
ed
scor
e
is
num
erical
ly
hig
he
r
than
or
eq
ua
l
to
the
c
um
ulati
ve
weig
hted sens
ory
estim
at
es o
f a
ll
the ch
il
dren
nod
e
s.
c.
Syst
em
i
m
p
lem
entat
ion
The
c
om
plete
syst
e
m
is
i
m
pl
e
m
ented
on
t
he
m
echan
is
m
of
grap
h
t
heor
y,
w
hich
is
f
urt
her
div
i
ded
into
tw
o
phase
viz.
gra
ph
c
onstr
uction
pha
se
an
d
m
ining
ph
a
se.
T
he
pr
opose
d
syst
em
us
es
ce
rtai
n
es
sentia
l
at
tribu
te
s
that
are
us
e
d
f
or
as
sessing
t
he
qua
li
ty
of
the
m
ini
ng
o
pe
rati
on
on
the
basis o
f
t
he
est
i
m
at
es
ob
ta
ined
from
each
ep
oc
h for in
div
i
dual
sen
s
or
s
.
−
Sensory
Esti
m
at
es
(S
E):
Ba
sic
al
ly
,
this
a
t
tribu
te
re
pr
ese
nts
an
act
ive
nu
m
b
er
of
ev
entual
rea
ding
captu
red by se
ns
or m
ote for
a
g
ive
n
e
poch
.
−
Ep
och
Se
nsory
Estim
at
es
(eS
E):
This
at
trib
ute
represe
nts
the
cum
ulati
ve
sens
or
y
est
im
a
te
cor
res
pondi
ng
to a
giv
e
n
e
poc
h.
−
Cum
ulati
ve
S
ens
or
y
Estim
a
te
(cSE)
:
This
at
tribu
te
dep
i
ct
s
the
cum
ul
at
ive
value
of
epo
c
h
sens
ory
est
i
m
at
e cor
res
pondin
g
to
all
the e
po
c
h
i
n
se
ns
ory
re
posit
or
y db.
−
Sh
a
re
Estim
ate
(S
hE
):
The
s
har
e
est
im
at
e
at
tribu
te
is
co
m
pu
te
d
as
an
est
i
m
at
ed
score
of
the
gro
up
of
sens
or
s
for a
giv
en
epoc
h divi
ded b
y t
he
c
umulat
ive se
ns
ory
es
tim
at
e fo
r
t
he
sen
s
ory
r
e
posit
or
y.
−
Mi
ni
m
u
m
Sh
are:
The
m
inim
u
m
sh
are
is
basical
ly
a
kin
d
of
c
ut
-
off
value
wh
ic
h
consi
ders
that
the
sh
ari
ng sc
or
e
of
patte
rn
sp
eci
f
ic
to
an
eve
nt
within a
ti
m
es
l
ot is m
or
e tha
n o
r
e
qu
al
t
o
m
i
nim
u
m
sh
are.
The
tree
-
ba
sed
m
ining
a
pproach
is
im
ple
m
ented
t
hat
ta
ke
s
the
in
pu
t
of
n
(t
otal
m
otes),
m
(
m
otes)
wh
ic
h
a
fter
pro
cessi
ng off
e
r
a
n ou
t
pu
t
of
ord
ered t
ree. The
ste
ps
in
volve
d i
n
this al
gorith
m
are:
Algori
th
m
for
Tree
-
b
as
ed
Mining
Inpu
t
: n
, m
Out
p
ut
:
orde
r
ed
tree
St
ar
t
1.
i
nit
2.
For
i=
1:n
3.
co
ns
tr
uct
a tree(
r
node
, c
node
, h
m
at
)
4.
set
e
po
c
h
SE
1
5.
sort(m
)
n
6.
sort(
wS
E
)
& reo
rg
a
nize t
ree
7.
En
d
End
The
desig
n
of
the
pro
posed
i
m
ple
m
entat
ion
consi
sts
of
r
no
de
as
the
r
oo
t
node
,
c
node
is
a
child
no
de,
and
h
m
at
is
a
m
at
rix
to
li
st
head
e
rs.
T
he
com
plete
al
go
r
it
h
m
i
m
ple
m
e
ntati
on
is
c
ar
ried
ou
t
i
n
tw
o
disti
nct
sta
ges
i.e.
m
ap
ping
sta
ge
a
nd
re
-
or
de
rin
g
tr
ee
sta
ge.
T
he
first
sta
ge
of
m
app
in
g
is
ca
r
ried
out
by
org
anizi
ng
the
m
otes
in
a
sp
eci
fic
orde
r
on
the
basis
of
t
heir
i
den
ti
f
ie
r.
F
or
this
pur
pose,
t
he
tre
e
is
co
ns
tr
ucted
by
inco
rpor
at
in
g
al
l
the
e
po
c
h
in
the
entire
s
ens
or
y
reposit
or
y
s
eq
ue
ntial
l
y
i
n
order
to
ob
ta
in
a
final
tree.
The
pro
pose
d
syst
e
m
al
so
const
ru
ct
s
the
m
at
rix
h
m
a
t
fo
r
retai
ni
ng
t
he
li
st
of
hea
der
s
of
the
s
ens
or
.
This
operati
on
is
carried
out
for
m
ai
ntaining
the
or
der
of
the
m
otes
as
well
as
stores
the
weig
hted
s
ens
or
y
est
i
m
at
es
(w
S
E)
ass
ociat
ed
with
t
he
se
nso
r.
In
order
to
m
ai
ntain
the
tr
aver
sal
featu
re
of
the
tree,
th
e
m
od
el
al
so
m
ai
ntains
neig
hbo
rin
g
points.
In
t
he
prel
i
m
inary
sta
ge
s
of
im
ple
m
e
ntati
on
,
the
tre
e
structu
re
is
usual
ly
e
m
pty
and
it
i
niti
at
es
with
t
he
root
node
and
fi
nally
,
the
tree
is
con
struct
e
d
us
in
g
a
ll
the
def
ined
epo
c
h.
The
final
sta
ge
of
im
ple
m
ent
at
ion
is
basical
ly
associat
ed
wi
th
th
e
re
-
ord
erin
g
process.
This
im
ple
m
entat
ion
sta
ges
basical
ly
ta
rg
et
s
m
e
m
or
y
r
ed
uctio
n
and
offe
rs
a
fa
ste
r
proce
ss
in
the
analy
ti
cal
operati
on.
T
he
sta
ge
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.
9
, N
o.
5
,
Oct
ober
201
9
:
3
8
2
2
-
3
8
3
2
3828
beg
i
ns
with
or
der
i
ng
t
he
li
st
of
el
em
ents
in
h
m
at
in
decr
ea
sing
orde
r
wit
h
res
pect
to
w
SE
w
her
e
m
er
ge
a
nd
so
rtin
g
proces
s
can
be
util
iz
ed
fo
ll
owe
d
by
furthe
r
r
e
-
order
i
ng
the
t
ree
struct
ur
e
on
t
he
basis
of
th
e
ne
w
value.
A
bra
nc
h
sorti
ng
m
ech
anism
can
be
us
e
d
for
the
pur
pose
of
sorti
ng
operati
on
of
the
tree.
I
n
t
his
tree
const
ru
ct
io
n
proces
s,
al
l
the
li
n
ks
am
on
g
the
sens
or
node
s
are
basical
ly
so
rted
usi
ng
this
br
anc
h
s
or
ti
ng
al
gorithm
.
The
analy
ti
cal
pro
cess
is
car
ried
ou
t
by
asses
sin
g
the
gro
wth
in
the
patte
r
n
i
n
order
to
e
xtr
act
the
knowle
dge
of
al
l
the
senso
r
y
data
fr
om
t
he
co
ns
tr
ucted
ta
ble.
The
pr
opo
se
d
syst
em
util
iz
es
the
patte
rn
-
grow
t
h
[
3
6
]
co
ncep
t
in
orde
r
to
pe
rfor
m
the
analy
ti
cal
op
e
rati
on
with
a
n
ai
d
of
wei
gh
te
d
sen
sory
est
im
at
es
.
The
com
plete
m
echan
ism
is
al
so
carrie
d
ou
t
us
ing
both
si
ng
le
hop
as
w
el
l
as
the
m
ult
i
-
hop
sc
hem
e
of
data
aggre
gatio
n.
It
will
m
ean
that
pr
op
os
e
d
syst
e
m
of
fer
s
e
xec
ution
of
it
s
ana
ly
ti
cal
op
erati
on
in
both
sin
gle
and
m
ul
ti
ho
p
netw
ork
f
or
offer
i
ng
it
s
su
pp
or
ta
bi
li
t
y
to
al
l
up
com
ing
routin
g
schem
es
that
work
on
the
pri
nciple
of d
ist
ri
bu
te
d m
ining
a
ppr
oa
ch ov
e
r
cl
oud
e
nv
i
r
onm
ent.
The
operati
on
of
the
par
al
l
el
pr
ocess
of
analy
ti
cal
operati
on
is
as
fo
ll
ows:
For
this
pur
po
se
,
the
pro
po
se
d
s
yst
e
m
first
ta
kes
the
input
of
reposit
or
y
db
and
pe
rfo
rm
s
segr
e
gation
of
the
db
base
d
on
n
nu
m
ber
of
se
nsors
.
T
he
outc
om
e
of
this
an
al
yt
ic
a
l
op
erati
on
will
be
fre
quent
pa
tt
er
ns
of
sen
sory
data
as
the
kno
wl
e
dg
e
.
Th
e
process
co
nsi
der
s
the
local
reposit
ory
syst
e
m
fo
r
each
lo
cat
ion
an
d
pe
r
form
s
the
so
rting
of
al
l
the
ep
oc
h
β
base
d
on
thei
r
sp
eci
fic
ide
ntifie
r.
It
the
n
i
nc
lud
es
the
value
of
e
po
c
h
β
int
o
the
de
sig
nated
tre
e
structu
re
f
ollo
wed
by
the
up
dating
op
e
rati
on
of
t
he
hea
der
ta
ble.
The
proces
s
the
n
forw
a
r
ds
th
e
pri
m
ar
y
sens
or
y
est
i
m
a
te
s
dep
e
nd
i
ng
upon
it
s
epo
c
h
and
the
weig
ht
value
.
Thes
e
values
are
f
orwarde
d
to
th
e
root
node.
T
he
se
nsory
es
ti
m
at
ed
base
d
on
sec
onda
ry
e
po
c
h
a
nd
wei
gh
t
fact
or
is
co
ns
tr
uct
ed
by
the
root
node.
The
seco
ndary
sensory
est
i
m
at
es
in
this
routing
ta
ble
are
then
re
-
orde
re
d
in
dec
reasin
g
orde
r
an
d
th
en
it
is
forw
a
r
ded
to
t
he
pri
m
ary
locat
ion
of
the
re
po
sit
or
y.
T
he
f
inal
tree
is
reco
ns
t
ru
ct
e
d
on
the
basis
of
the
new
ly
acqu
i
red
orde
r
fo
ll
owe
d
by
the
ide
ntific
at
io
n
of
t
he
sim
i
lar
val
ue
of
t
he
sens
or
y
est
im
a
te
s
with
e
po
c
h
an
d
weig
ht
ov
e
r
a
ll
the
ro
utes
and
the
n
they
are
integrate
d
to
the
sin
gl
e
sensor
m
ote
.
The
ne
xt
ste
p
of
the
analy
ti
cal
op
e
rati
on
is
c
arr
ie
d
out
onl
y
if
the
pa
rtit
ion
e
d
re
posit
ory
is
processe
d.
T
he
us
e
r
obta
ins
the
m
ini
m
u
m
s
har
e
a
nd
the
n
a
ll
the
senso
r
m
ote
chec
ks
if
epo
c
h
ba
sed
se
nsory
est
im
a
te
s
are
fou
nd
m
or
e
than
the
m
ini
m
u
m
cut
-
off
value
of
t
he
se
nsory
es
tim
at
es.
In
the
posit
ive
ca
se,
the
sen
sor
m
ote
is
include
d
i
n
the
upc
om
ing
patte
rn
li
st
f
ollow
e
d
by
an
it
e
rati
ve
analy
ti
cal
op
e
rati
on
th
at
pr
e
fixes
t
he
m
ote
by
con
str
uctin
g
a
rev
ise
d
tree
structu
re.
All
t
he
upcom
ing
pa
tt
ern
s
are
ad
de
d
with
the
se
ns
or
m
ote
in
t
he
upcom
ing
li
st
and
then
the
upco
m
ing
patte
r
ns
are
f
orwarde
d
to
the
r
oo
t
node
f
ro
m
the
entire
pri
m
ary
node.
Finall
y,
al
l
the
fr
e
qu
e
nt
patte
r
ns
a
re
ob
ta
in
e
d
from
the
roo
t
node
w
hich
no
t
only
re
du
c
es
the
ti
m
e
of
analy
sis
but
90%
c
ut
sh
ort
the
pr
oce
ss of
m
ining
operati
on
w
hen
perform
ed
fro
m
the roo
t
node
.
4.
RESU
LT
DI
S
CUSSIO
N
The
analy
sis
of
the
propose
d
syst
e
m
is
carried
out
in
MATLAB
us
in
g
a
sim
ula
ti
on
-
base
d
ap
proach.
Norm
al
sys
tem
us
ing
4G
B
RAM
an
d
windows
OS
is
use
d
f
or
the
sim
ulati
on
stu
dy.
On
e
of
t
he
ess
entia
l
factors
to
be
c
on
si
der
e
d
duri
ng
the
analy
si
s
of
t
he
resu
lt
is
the
data
as
the
c
om
plete
pr
oces
s
of
the
m
inin
g
op
e
rati
on
is
carried
out
on
t
hi
s.
The
pr
opose
d
syst
em
consi
der
s
the
m
ote
config
ur
at
io
n
base
d
on
M
EMSIC
nodes
.
The
pro
po
s
ed
syst
em
i
s
si
m
ulate
d
in
order
t
o
ge
ne
r
at
e
m
axi
m
iz
ed
tup
le
s
of
se
nsory
da
ta
again
st
any
sp
eci
fic
en
vir
onm
ental
data.
The
analy
sis
was
al
so
car
rie
d
out
that
data
aggreg
at
io
n
ha
s
been
ca
rr
ie
d
out
in
the prese
nce
of inf
e
rio
r qu
al
it
y of t
he wirel
e
ss m
ediu
m
f
rom
the
transm
it
t
ing
source
.
The
analy
sis
of
the
pro
posed
ou
tc
om
e
a
lso
assum
ed
that
i
f
there
a
re
any
fo
rm
s
of
m
iss
ed/ski
ppe
d
sens
or
y
data
t
han
it
is
consi
der
e
d
as
un
detect
ed
an
ev
ent
that
assist
s
in
gen
e
rati
ng
sk
i
pp
e
d
se
nsory
r
eadin
g.
A
syntheti
c
da
ta
is
gen
erat
ed
that
act
ual
ly
do
es
n'
t
of
f
er
any
f
orm
of
i
nfor
m
at
ion
of
th
e
sh
a
re
data
corres
pondin
g
to
the
al
l
the
it
e
m
s
equ
ivale
nt
to
al
l
the
un
it
transacti
ons.
In
order
t
o
co
nne
ct
/
m
ap
al
l
the
it
e
m
s
with
the
exa
ct
even
t
ual
data,
the
pro
po
se
d
syst
e
m
add
s
an
ar
bitrary
nu
m
ber
associat
ed
with
al
l
the
it
e
m
s.
A
sim
ulatio
n
e
nv
i
ronm
ent
of
100x10
0
m
2
is
const
ru
ct
e
d
in
MATLAB
with
50
se
ns
or
no
des
bei
ng
distr
ibu
te
d
rand
om
l
y
ov
er
it
.
As
the
pr
opose
d
syst
em
i
s
desig
ne
d
ove
r
the
c
on
ce
pt
of
I
oT;
there
f
ore,
a
gateway
node
is
po
sit
io
ne
d
in
the
center
of
t
he
sim
ulatio
n
area
that
is
m
eant
for
assist
ing
in
the
tra
nsl
at
ion
al
serv
i
ces
of
diff
e
re
nt
ty
pes
of
r
ou
ti
ng
op
e
rati
on
i
n
WSN.
The
a
naly
sis
is
carried
out
with
res
pect
to
two
sce
na
rios
wh
e
re
the
data
a
ggre
gation
is
car
ried
ou
t
us
in
g
a
sing
l
e
hop
com
m
un
ic
at
ion
syst
em
as
well
as
distri
bute
d
(or
m
ul
ti
ho
p) c
omm
un
ic
at
ion
syst
e
m
. F
ig
ure
2
a
nd Fig
ure
3 hig
hligh
t t
he
sim
ulati
on
pr
ocess of
d
at
a a
ggre
ga
ti
on
.
The
a
naly
sis o
f
the
ou
tc
om
e o
btained
fro
m
the sim
ulati
on
s
tud
y:
a.
A
naly
sis o
f
tim
e req
ui
red
for
a
naly
ti
cal
o
per
at
io
n
Pr
oc
essin
g
ti
m
e
play
s
an
es
s
entia
l
ro
le
in
offer
i
ng
cl
a
rity
towards
a
f
ast
er
res
pons
e
rate
of
the
m
ining
proces
s.
As
the
pro
pose
d
syst
e
m
is
i
m
ple
m
ented
ov
e
r
the
distri
bu
te
d
tree
struc
ture,
it
is
essenti
al
to
unde
rstan
d
tha
t
ho
w
f
a
st
the
mini
ng
oper
at
ion
is
possibl
e
?
Othe
rw
ise
,
the
a
pp
li
cabil
it
y
of
the
pro
po
s
ed
m
ining
oper
at
ion
is
diff
ic
ult
to
unde
rstan
d.
Figure
4
high
li
gh
ts
the
com
par
at
ive
a
naly
sis
of
the
pro
po
s
ed
syst
e
m
with
e
xisti
ng
f
re
qu
e
nt
patte
rn
al
gorithm
ov
er
inc
re
asi
ng
hy
po
t
het
ic
al
data.
The
ou
tc
om
e
sh
ows
that
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
Framew
or
k
for
co
st
-
ef
fe
ct
iv
e
analyti
cal
mod
el
li
ng
for
sens
or
y
data
over c
loud
…
(
M
anuj
aks
hi B
.
C
.
)
3829
al
tho
ug
h
proce
ssing
tim
e
incr
eases
f
or
both
appr
oach
es
,
it
is
com
par
at
ive
ly
bette
r
f
or
th
e
pro
posed
sys
tem
i
n
con
t
rast
to
th
e
existi
ng
a
ppr
oac
h.
T
he
pr
im
e
reaso
n
beh
i
nd
t
his
outc
om
e
is
that
existi
ng
fr
e
qu
e
ntly
patte
rn
just
re
cords
the
occ
urre
nces
of
it
em
s
wh
ic
h
al
s
o
expo
nen
ti
al
ly
gro
w
with
th
e
increase
in
t
raffic
.
Howe
ver,
the
pro
po
se
d
syst
em
ob
ta
ins
hi
ghly
tim
e
-
based
uniq
ue
data,
wh
ic
h
no
t
only
reduces
t
he
siz
e
of
patte
rn
s
but als
o offers
f
ast
e
r m
ining
op
e
rati
on.
Figure
2 Si
ngle
hop data a
ggr
egati
on
Figure
3 Dist
ri
bu
te
d data a
ggreg
at
io
n
Figure
4
C
om
par
at
ive a
naly
sis o
f
m
ining
ti
m
e
b.
A
naly
sis o
f
energy
dep
le
ti
on for analy
ti
cal
o
pe
rati
on
Energy
is
one
of
the
m
os
t
pract
ic
al
par
am
e
te
rs
in
orde
r
to
ju
dg
e
t
he
effe
ct
iveness
of
a
ny
form
of
processi
ng
bei
ng
us
e
d
withi
n
the
sen
sor
m
ote.
Fo
r
pr
ac
ti
cal
op
erati
on,
it
is
al
ways
e
xp
ect
e
d
that
a
sens
or
m
ote
sh
ou
l
d
not
de
plete
e
nergy
at
a
ver
y
fa
ste
r
rate
i
n
ord
er
to
su
sta
i
n
a
bette
r
f
orm
of
netw
ork
sta
bili
ty
and
li
fe
tim
e.
Figu
r
e
5
highli
gh
ts
t
hat
the
pro
po
se
d
syst
e
m
of
fer
s
hig
hly
re
du
ce
d
ene
rg
y
co
nsu
m
pt
ion
com
pared
to
existi
ng
f
re
qu
e
nt
patte
rn
base
d
m
ining
a
ppr
oach.
The
pr
i
m
e
reaso
n
be
hi
nd
this
is
exis
ti
ng
fr
e
quent
pa
tt
ern
s
retai
ns
a
m
axim
u
m
nu
m
ber
of
patte
r
ns
w
hi
ch
i
ncr
ease
s
th
e
siz
e
of
the
tree
if
the
incomi
ng
data
is
consi
der
e
d
as
a
bit
strea
m
.
This
cause
s
excessi
ve
de
pleti
on
of
tran
sm
issi
on
ene
r
gy
towards
a
gg
reg
at
io
n
fo
ll
owed
by
m
ining
in
the
cl
oud
en
vir
on
m
ent.
H
ow
e
ve
r,
pro
po
se
d
ad
dr
ess
this
prob
le
m
by
intro
duci
ng
a
tree
co
ns
t
ru
ct
io
n
and
re
-
order
i
ng
w
her
e
the
m
i
ning
operati
on
is
carried
out
ov
e
r
r
oo
t
node
with
al
l
up
dat
es
causin
g
m
a
xim
u
m
energy sa
ving.
c.
An
al
ysi
s
of
m
e
m
or
y con
s
um
pt
ion
for
a
na
ly
ti
cal
o
per
at
io
n
The
pro
pose
d
syst
e
m
of
fer
s
r
edu
ce
d
m
e
m
or
y
con
s
um
ption
w
it
h
each
no
de
program
m
ed
with
10
00
KB
of
inter
nal
m
e
m
or
y.
Figure
6
highli
gh
ts
that
existi
ng
f
reque
nt
patte
rns
-
bas
ed
m
ining
appro
ac
h
c
ons
um
es
higher
m
e
m
or
y
ow
in
g
to
the
increase
of
da
ta
siz
e
ov
er
th
e
tree,
w
her
eas
the
pro
po
se
d
syst
e
m
m
ai
ntain
s
th
e
m
ined
dat
a
onl
y i
n
the
r
oo
t
no
de.
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.
9
, N
o.
5
,
Oct
ober
201
9
:
3
8
2
2
-
3
8
3
2
3830
Figure
5
.
Com
par
at
ive
an
al
ys
is of ene
r
gy d
e
pleti
on
Figure
6
.
Com
par
at
ive
an
al
ys
is of m
e
m
or
y
5.
CONCL
US
I
O
N
The
c
or
e
ide
a
of
t
he
pro
pose
d
pa
pe
r
is
to
s
howca
se
that
th
ere
is
a
highe
r
deal
of
com
plexity
in
order
to
plan
a
desi
gn
of
an
arc
hitec
ture
pe
rfor
m
ing
a
n
analy
ti
cal
op
erati
on
over
la
r
ge
sen
s
or
y
data
on
th
e
cl
ou
d
env
i
ronm
ent.
Stud
ie
s
an
d
ap
proac
hes
us
in
g
the
e
xisti
ng
s
yst
e
m
are
m
or
e
f
ocu
s
ed
on
case
-
s
pecific
m
inin
g
wh
il
e
the
num
ber
of
stu
dies
of
knowle
dge
extracti
on
exis
ts
fo
r
th
e
sens
or
netw
ork.
H
ow
e
ve
r,
sig
nif
ic
ant
researc
h
to
wards
kn
ow
le
dg
e
extracti
on
of
s
ens
or
y
data
over
I
oT
en
vir
on
m
ent
is
j
us
t
in
the
nasce
nt
sta
ge
an
d
need m
or
e ex
pl
or
at
ive
proces
s.
T
he
pr
opos
e
d
syst
em
arg
ue
s that f
reque
nt
patte
rn
can
be
t
reated as
kn
owle
dg
e
bu
t
they
ca
nnot
be
directl
y
us
ed
i
n
s
uch
com
plex
and
distrib
uted
e
nv
i
ronm
ent.
Ther
e
fore,
t
he
pape
r
introd
uces
a
sign
i
ficant
nove
lt
y
in
the
exist
ing
syst
em
of
fr
e
qu
e
nt
patte
r
ns
an
d
intr
oduc
es
var
i
ou
s
se
ns
or
y
est
i
m
at
es
to
it
in
orde
r
t
o
off
er
m
or
e
gran
ul
arit
ie
s
in
the
a
naly
ti
cal
ou
tc
om
e.
The
st
ud
y
al
so
us
es
t
ree
-
base
d
topolo
gy
in
or
der
to
m
ake
a
structu
re
d
of
da
ta
m
anag
em
e
nt
associat
ed
with
m
ining
outc
om
e.
The
pro
pose
d
m
od
el
is
si
m
ulate
d
in
M
AT
LAB
ov
e
r
nor
m
al
syst
e
m
con
fi
gurati
on
to
fin
d
that
it
off
ers
reduce
d
e
ne
rg
y
consum
ption
,
re
du
ce
d delay
, a
nd b
et
te
r
m
e
m
or
y uti
li
zation
.
REFERE
NCE
S
[
1
]
Udoh,
Emm
anue
l,
"Evol
ving
Deve
lopments
i
n
Grid
and
C
l
o
ud
Com
puti
ng:
Advanc
ing
R
e
sea
rch
:
Advanc
i
ng
Resea
rch
"
,
IGI
Global
,
pp
.
383
,
2012
[
2
]
Ahm
ed
Abdelga
wad,
Magd
y
Ba
y
oum
i,
"Resourc
e
-
Aw
are
Dat
a
Fus
ion
Algorit
hm
s
for
W
ire
le
ss
S
ensor
Networks",
Springer
Scienc
e
&
Bu
siness Me
dia
,
pp
.
108
,
2012
[
3
]
Javie
r Lópe
z, Jian
y
ing
Zhou
,
"W
i
rel
ess Sensor Ne
twork
Secur
i
t
y
"
IOS
Press
,
Comput
ers,
pp
.
313
,
2018
[
4
]
Za
m
an,
Noor
,
"
W
ire
le
ss
Sensor
Networks
and
Ene
rg
y
E
fficie
n
c
y
:
Protoco
ls,
R
outi
ng
and
Man
age
m
ent
:
Protoc
ols,
Routi
ng
and
Ma
nage
m
ent
",
IGI
Global
,
pp
.
655
,
2012
[
5
]
olge
r
Karl
,
And
rea
s
W
il
li
g
,
"P
r
otoc
ols
and
Arc
hit
e
ct
ure
s
for
W
ire
le
ss
Sensor
Networks",
Jo
hn
Wil
e
y
&
So
ns
,
pp.
497
,
2007
[
6
]
Holger
Karl
,
Andrea
s
W
il
li
g
,
"
Protocol
s
and
Archi
tectur
es
fo
r
W
ire
le
ss
Sensor
Networks",
J
ohn
Wil
e
y
&
So
ns,
Technol
ogy
&
E
ngine
ering
,
pp
.
497,
2007
[
7
]
Moham
m
ad
S.
Obaida
t
,
Sudip
Misra,
"P
rinc
ip
l
es
of
W
ire
l
ess
Sensor
Network
s"
,
Cambridge
Univer
sit
y
Press
,
pp.
415
,
2014
[
8
]
Frode
Ei
k
a
Sa
ndnes,
Yan
Zh
ang,
Chunm
ing
Rong,
La
ur
en
ce
Ti
anr
uo
Ya
ng,
"U
biqui
tous
Intelli
g
enc
e
a
nd
Com
puti
ng”
5th
Inte
rnational
Confe
renc
e
,
UIC
2008
,
Pro
ce
ed
ings
",
Springer
Os
lo,
Norw
a
y
,
June
23
-
25,
pp.
763
,
2008
[
9
]
David
Hane
s,
G
onza
lo
Salgueiro
,
Patr
ic
k
Gros
set
et
e
,
Robe
rt
B
arton,
Jerom
e
Henr
y
,
"IoT
Fund
amentals:
Networki
ng
Te
chno
logi
es,
Pr
otoc
ols,
and
Us
e
Cases
for
the In
te
rne
t
of
Thi
ngs"
,
Cis
co
Press
,
pp
.
576
,
2017
[
1
0
]
Christophe
r
Siu,
"IoT
and
Low
-
Pow
er
W
ire
le
ss
:
Circ
uit
s,
Archi
t
ec
tur
es,
and
Tec
hnique
s"
,
CRC
Press,
Technol
o
gy
&
E
ngine
ering
,
pp.
394
,
2018
[
1
1
]
Conti
,
Franc
esc
o,
e
t
al.
"A
n
Io
T
endpoi
nt
s
y
st
em
-
on
-
chi
p
for
sec
ure
and
ene
r
g
y
-
e
fficie
nt
ne
a
r
-
sensor
ana
l
y
tic
s
"
,
IEE
E
Tr
ansacti
o
ns on
Circuits a
nd
Syste
ms
I:
R
e
gular P
apers
,
vo
l
64
,
no.
9,
pp.
2
481
-
2494
,
2017
.
[
1
2
]
Bert
ran
d
,
Alex
a
nder
,
and
Marc
Moonen.
"D
istributed
c
anoni
c
al
cor
relati
on
anal
y
sis
in
wir
el
ess
sensor
net
works
with
applic
at
ion
to
distr
ibut
ed
b
li
nd
sourc
e
sepa
rat
ion
.
"
IE
EE
Tr
ansacti
ons
on
Si
gnal
Proc
essing
,
vol.
63,
No.
18
,
pp.
4800
-
4813
,
2015
.
[
1
3
]
Parwez
,
Md
Sali
k,
Danda
B.
Rawa
t,
and
Mos
es
Garuba
.
"Big
dat
a
an
aly
t
ic
s
for
user
-
ac
ti
v
ity
ana
l
y
sis
and
user
-
anomal
y
de
tection
in
the
m
obil
e
wire
le
ss
net
work."
IEE
E
Tr
ansacti
ons
on
In
dustrial
Informa
ti
cs
,
vol
.
13
,
no.
4,
pp.
2058
-
2065
,
2017
.
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
Framew
or
k
for
co
st
-
ef
fe
ct
iv
e
analyti
cal
mod
el
li
ng
for
sens
or
y
data
over c
loud
…
(
M
anuj
aks
hi B
.
C
.
)
3831
[
1
4
]
Rahman,
As
hfa
qur,
John
McCull
och
,
and
Qua
zi
Mam
un.
"P
re
dic
ti
on
wi
th
un
ce
rt
ai
nt
y
:
a
nov
el
fra
m
ework
f
or
ana
l
y
z
ing
sensor
data
str
e
ams
.
"
I
EE
E
Sensors
Jo
urnal
,
vol
15,
no
.
1,
pp
.
382
-
386
,
2015
.
[
1
5
]
Rehman,
Ur
Muham
m
ad
Habib,
et
a
l.
"Big
Data
Anal
y
t
ic
s
in
Industria
l
Io
T
Us
ing
a
Conce
ntr
ic
Com
puting
Model.
"
I
EE
E
C
omm
u
nic
ati
ons
Magazine
,
vo
l
56
,
no
2,
pp
37
-
43
,
2018
.
[
1
6
]
Sun,
Yunchua
n,
et
a
l.
"In
te
rn
et
o
f
thi
ngs
and
b
ig
dat
a
a
na
l
y
tics
fo
r
sm
art
and
conn
ec
t
ed
comm
unit
i
es.
"
IE
EE
a
ccess
,
vol
4,
pp.
766
-
7
73
,
2016
.
[
1
7
]
W
ang,
Ying,
As
hish
Pandhar
ip
a
nde,
and
Pe
te
r
Fuhrm
ann.
"Ene
r
g
y
D
at
a
Anal
y
t
i
cs
for
Nonintrus
ive
Li
ght
ing
As
set
Monitori
ng
and Ene
rg
y
Disaggr
e
gat
ion
.
"
IE
EE Sens
ors
Journal
,
v
ol
18
,
no
7,
pp
.
2
934
-
2943
,
2018
.
[
1
8
]
Yue,
Peng,
et
al
.
"A
n
SD
I
appr
oac
h
for
big
dat
a
an
aly
t
ic
s:
The
ca
se
on
sensor
web
eve
nt
det
e
ct
ion
an
d
geopr
oce
ss
ing
workflow."
IEEE
Journal
of
Se
le
c
te
d
Topics
in
Appl
ie
d
Earth
Obs
erv
ati
on
s
and
Re
mote
Sensi
ng
,
vol
8
,
no.
1
0
,
pp.
4720
-
4728
,
201
5
.
[
1
9
]
Pena
-
Ramos
,
Juan
-
Carl
os,
e
t
al.
"A
Fully
Co
nfigura
bl
e
Non
-
Li
ne
ar
Mixed
-
Signal
Interface
for
Multi
-
Sensor
Anal
y
tics."
IE
E
E
Journal
of
S
olid
-
Stat
e
Circuits
,
vol
53
,
no
11
,
pp
.
3140
-
3149
,
20
18
.
[
2
0
]
Sharm
a,
Shree
Krishna,
and
Xi
anbi
n
W
ang.
"Li
ve
Dat
a
Anal
y
t
i
c
s
W
it
h
Coll
abo
r
at
iv
e
Edg
e
and
Cloud
Proce
ss
in
g
in
W
ire
le
ss
IoT
Ne
tworks."
IEEE A
cc
ess
,
vol.
5
,
no.
99
,
pp.
4621
-
46
35
,
2017
.
[
2
1
]
Cao,
Ning
y
uan
,
et
al.
"S
el
f
-
optim
iz
ing
IoT
wire
le
ss
vide
o
sensor
node
with
in
-
si
tu
dat
a
ana
l
y
t
ic
s
and
cont
ex
t
-
driv
en
ene
rg
y
-
awa
re
re
al
-
ti
m
e
ada
pt
at
i
on.
"
IE
EE
Tr
ansacti
ons
on
Circ
uit
s
and
Syste
ms
I:
Re
gular
Pap
ers
,
vol
.
64
,
no.
9
,
pp.
2470
-
2480
,
2017
.
[
2
2
]
He,
Jianhua,
e
t
al
.
"M
ultiti
er
fo
g
computing
with
la
rg
e
-
sca
l
e
io
t
data
anal
y
tics
for
sm
art
ci
ties"
IEEE
Int
er
ne
t
of
Things
Journal
,
vol
5
,
no.
2
,
pp.
677
-
686
,
2018
.
[
2
3
]
Gong,
Yanm
in,
Yuguang
Fang
,
and
Yuanxion
g
Guo.
"P
riva
te
dat
a
anal
y
tics
on
biomedic
al
sensing
dat
a
vi
a
distri
bute
d
computat
ion
"
IEEE/
ACM
transacti
o
ns
on
computat
ional
biol
og
y
a
nd
bioi
nformati
cs
,
vol
.
13
,
no
.
3
,
pp.
431
-
444
,
20
16
.
[
2
4
]
Li
,
Xian
,
Hui
Huang,
and
Ye
Sun.
"D
riT
ri:
An
in
-
vehi
cle
wire
le
ss
sensor
net
work
pla
tfo
rm
for
dai
l
y
he
al
t
h
m
onit
oring
"
SE
NSORS,
IE
EE
,
2
016.
[
2
5
]
Yildi
rim,
Murat
,
Na
gi
Z.
Gebr
ae
e
l,
and
Xu
And
y
Sun,
"Inte
g
rat
ed
Predi
ct
iv
e
Anal
y
t
ic
s
and
Optimiza
ti
o
n
for
Opportunisti
c
Maintena
n
c
e
and
Opera
ti
ons
in
W
ind
Farm
s
"
IEE
E
Tr
ansacti
ons
on
Pow
er
Syst
e
ms
,
vol,
32
,
no.
6
,
pp.
4319
-
4328
,
2017
.
[
2
6
]
Li
ao
,
Yizhe
ng
,
et
al
,
"S
nowfort:
An
open
sourc
e
wire
le
ss
sensor
net
work
for
d
at
a
an
aly
t
ic
s
in
infra
struc
ture
an
d
envi
ronm
ent
a
l
m
onit
oring"
,
I
E
EE Se
nsor
s J
ournal
,
vol
.
14
,
no
.
12
,
pp.
4253
-
4263
,
2014
.
[
2
7
]
Muñoz,
Raü
l,
et
al
.
,
"Int
egr
a
ti
on
of
IoT,
Tr
ansport
SD
N,
and
Ed
ge/
Cloud
Com
p
uti
ng
for
D
y
na
m
ic
Distribut
ion
of
IoT
Anal
y
t
ic
s
and
Eff
i
cient
Us
e
of
Network
Resourc
es,
"
J
ournal
of
Light
wave
Techno
lo
gy
,
vo
l.
36
,
no
.
7
,
pp.
1420
-
1428
,
2018
.
[
2
8
]
Oter
o,
Ca
rlos
E
.
,
et
a
l.
"A
wire
le
ss
sensor
netw
orks'
anal
y
tics
s
y
stem
for
pre
dic
ti
ng
per
fo
rm
anc
e
in
on
-
dem
and
depl
o
y
m
ent
s,
"
I
EE
E
Syste
ms
Jo
urnal
,
vo
l
.
9
,
no
.
4
:
1344
-
1353
,
2
015
.
[
2
9
]
Shao,
Yun,
e
t
al
.
"H
eur
isti
c
opti
m
iz
a
ti
on
f
or
rel
i
able
data
conge
stion
an
aly
tics
in
cro
wds
ourc
ed
eHe
alth
net
works
,
"
IE
EE A
c
ce
ss
,
vo
l.
4
:
p
p.
9174
-
9183
,
2
016
.
[
3
0
]
Iva
nov,
Step
an,
Kriti
Bharg
ava,
and
W
il
l
ia
m
Donnelly
,
"P
re
c
ision
far
m
ing:
Sensor
ana
l
y
t
ic
s
,
"
IE
EE
Int
el
l
ig
ent
systems
,
vol
.
30
,
no.
4
,
pp
76
-
80
,
2015
.
[
3
1
]
Chandra
ka
la,
N
.
,
and
B
.
Thi
ru
m
al
a
Rao
.
"M
i
gra
ti
on
of
Vir
t
ual
Ma
chi
ne
to
improve
th
e
S
ec
uri
t
y
in
Clou
d
Com
puti
ng,
"
Int
ernati
onal
Journ
al
of
Elec
tric
al
&
C
o
mput
er
En
gine
ering
,
vol
.
8
,
no
.
1
,
pp
.
210
-
219,
2018
.
[
3
2
]
Rghioui
,
Am
ine
,
and
Abdelmajid
Oum
nad.
"Int
ern
et
of
Thi
ngs:
Surve
y
s
for
M
ea
suring
H
um
an
Acti
vit
i
es
from
Eve
r
y
where
"
Int
ernati
onal
Journ
al
of
Elec
tric
al
&
C
o
mputer
En
gine
ering
,
vol
7,
no.
5
,
pp.
2474
-
2482
,
2017.
[
3
3
]
Sindhu,
C.
S.,
a
nd
Naga
rat
na
P.
Hegde
.
"A
Novel
Inte
gra
te
d
Fra
m
ework
to
Ensu
re
Bet
t
er
Data
Q
ual
ity
in
Big
Da
ta
A
naly
tics
over
Cloud
Envi
ron
m
ent
"
Inte
rnatio
nal
Journal
of
El
e
ct
rica
l
&
Co
mputer
Engi
ne
e
ring
,
vol.
7
,
no
.
5
,
pp.
2798
-
2805
2017
.
[
3
4
]
Manuja
ksh
i,
B
.
C.
,
and
K.
B
.
Ramesh.
"S
Daa
S:
fra
m
ework
of
s
ensor
dat
a
as
a
servic
e
for
l
eve
r
agi
ng
services
i
n
Inte
rne
t
of
T
hings."
In
In
te
r
nati
onal
Con
fer
enc
e
on
Em
erging
R
ese
arc
h
in
Computi
ng,
Informatio
n,
Comm
unic
ati
on,
and
App
li
ca
ti
on
s
,
pp.
351
-
363.
Springer,
Singapo
re,
2016
.
[
3
5
]
Manuja
kshi,
B
.
C.
,
and
K
.
B.
R
amesh.
"A
Novel
Expe
r
imenta
l
Protot
y
p
e
for
As
sess
ing
IoT
Perform
anc
e
on
Real
-
Ti
m
e
Anal
y
t
ic
s
.
" In
Computer
S
ci
en
ce Onli
ne
C
onfe
renc
e
,
pp.
4
6
-
55.
Springe
r,
Cham,
2018.
[
3
6
]
Rashid,
Md
Mam
unur,
Iqba
l
Gondal,
and
Joar
der
Kam
ruz
za
m
a
n.
"S
har
e
-
fre
qu
ent
sensor
patte
rns
m
ini
ng
from
wire
le
ss
sensor
net
work
data."
IEE
E
Tr
ansacti
ons
on
Par
all
el
and
Distr
ibut
ed
Syst
ems
,
vol
26,
no
.
1
2
,
pp.
3471
-
3484
,
2
015
.
Evaluation Warning : The document was created with Spire.PDF for Python.