Int
ern
at
i
onal
Journ
al of
El
e
ctrical
an
d
Co
mput
er
En
gin
eeri
ng
(IJ
E
C
E)
Vo
l.
10
,
No.
4
,
A
ugus
t
2020
,
pp.
4217
~
42
25
IS
S
N:
20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v
10
i
4
.
pp
4217
-
42
25
4217
Journ
al h
om
e
page
:
http:
//
ij
ece.i
aesc
or
e.c
om/i
nd
ex
.ph
p/IJ
ECE
A compr
ehensi
ve study o
n di
sea
s
e risk p
redictions
in m
ac
hi
ne lea
rning
G.
S
ar
anya
,
A
. Pr
av
in
Dep
art
m
ent
o
f
C
om
pute
r
Scie
n
ce a
nd
Engi
n
ee
rin
g,
Sath
y
a
bama
I
nstit
ute of
Sc
ie
n
ce
and Technolo
g
y
,
Indi
a
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Oct
4,
2019
Re
vised Feb
22, 2
020
Accepte
d
Fe
b 29,
2020
Over
rec
ent
y
e
ars,
m
ult
ipl
e
d
isea
se
risk
pre
dic
ti
on
m
odel
s
have
bee
n
deve
lop
ed.
Thes
e
m
odel
s
use
var
ious
patien
t
cha
ra
cteri
sti
cs
t
o
esti
m
at
e
the
proba
bi
li
t
y
of
outc
om
es
over
a
c
ert
a
in
per
iod
of
ti
m
e
and
hold
the
pot
ent
i
al
to
i
m
prove
dec
ision
m
aki
ng
and
i
ndi
vidua
lize
ca
r
e.
Discove
ring
hidde
n
p
at
t
ern
s
and
int
e
racti
o
ns
from
m
edi
c
al
d
at
ab
ase
s
wi
th
growing
eva
lu
at
ion
of
th
e
disea
se
pre
di
cti
on
m
odel
has
b
e
come
cru
c
ial.
I
t
nee
ds
m
an
y
tri
al
s
in
tr
adition
al
c
li
ni
cal
find
in
gs
tha
t
coul
d
co
m
pli
ca
t
e
dise
ase
pre
di
ct
io
n
.
A
Com
pre
hensi
ve
stud
y
on
diffe
ren
t
stra
te
gi
es
used
to
pre
dic
t
disea
se
is
conf
err
ed
in
th
i
s
pape
r.
Appl
y
i
ng
the
se
t
ec
hn
i
ques
to
he
al
th
care
data,
has
improvem
ent
of
risk
pre
diction m
odel
s to
find
out
the
p
at
i
ent
s who would
ge
t
bene
fi
t
from
disea
se
m
ana
ge
m
e
nt
progra
m
s
to
red
uce
hospit
al
rea
dm
ission
and
he
al
th
ca
r
e c
ost,
but
th
e
r
esult
s of
th
ese
endea
vors ha
ve
bee
n
s
hift
ed
.
Ke
yw
or
d
s
:
Breast
cance
r
Disease
pr
e
dicti
on
Hear
t
disease
Ma
chine
le
a
rn
i
ng
Pr
e
dicti
on
m
odel
s
Copyright
©
202
0
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
:
G.
Sara
nya,
Re
search
Sc
hola
r
,
Dep
a
rtm
ent o
f
Com
pu
te
r Sci
ence a
nd E
ng
i
neer
i
ng,
Sathya
bam
a Insti
tute of
Scie
nc
e an
d
Tec
hnol
og
y,
Jep
piaar
Nag
a
r
, Raj
i
v Gand
hi
Sala
i, Che
nn
ai
-
60
0 119, Tam
il
nad
u,
India
.
Em
a
il
:
saran
ya
g3@s
rm
is
t.edu.in
1.
INTROD
U
CTION
Ma
chine
le
ar
ni
ng
is
one
of
t
he
m
os
t
pr
evale
nt
m
et
ho
ds
us
e
d
in
m
ulti
ple
com
pu
te
r
en
gine
erin
g
areas
and
has
bee
n
c
omm
on
ly
us
ed
in
the
pr
ocess
ing
of
natu
ral
la
ngua
ge,
pictu
re
proce
ssin
g,
patte
rn
rec
ogni
ti
on
,
cy
ber
sec
ur
it
y,
and
m
ulti
ple
areas.
On
e
of
the
m
os
t
dynam
ic
research
es
of
m
achine
le
arn
i
ng
is
healt
hcar
e
industry
[1
]
.
A
s
healt
hca
re
fir
m
s
are
try
ing
t
o
c
ollec
t
rec
ords
of
patie
nts,
est
i
m
at
es
sh
ow
that
the
re
a
re
abou
t
1
tril
li
on
byte
s
of
in
f
or
m
at
ion
,
w
hich
is
in
creasin
g
e
ver
y
day.
Th
ese
da
ta
nee
d
to
be
pr
operly
e
xtra
ct
ed
to
ob
ta
in
preci
ous
infor
m
at
ion
[2
]
.
So
m
et
i
m
es
patie
nts
fail
to
de
fine
thei
r
m
e
dical
pro
blem
s
correct
ly
and
the
res
ults
of
la
borato
ry
researc
h
ca
n
le
ad
t
o
s
om
e
degree
of
m
ist
ake.
Sp
eci
al
ist
s
fin
d
it
diff
ic
ult
to
m
ake
decisi
on
s
ab
out
the
il
l
nesses
because
they
m
ay
no
t
hav
e
sk
il
ls
in
al
l
areas.
To
a
ddres
s
this
issue,
it
is
necessary
to
de
velo
p
a
di
sease
pr
e
dicti
on
syst
e
m
that
com
bin
es
m
edical
know
le
dg
e
with
an
inte
gr
at
e
d
s
yst
e
m
to
produ
ce
the
bigges
t
resu
lt
s
and can
h
el
p
s
ociet
y
[3
]
.
Seve
ral
e
arli
er
inv
est
igati
ons
trie
d
to
us
e
patie
nt
la
borat
or
y
te
sts
[
4
-
6]
and
dru
gs
[7]
to
pr
e
dict
the
occ
urre
nce
of
disease
.
S
uch
prot
otypes
wer
e
al
s
o
use
d
to
de
fine
unkn
own
risk
f
act
or
s,
oft
en
wh
il
e
enh
a
ncin
g
se
nsi
ti
vity
and
spe
ci
fici
ty
of
de
te
ct
ion
si
m
ultaneo
u
sly
.
Re
ce
nt
stud
ie
s
have
been
e
ff
ect
i
ve
in
pr
e
dicti
ng
dis
ease
thr
ough
m
ul
ti
ple
m
et
h
od
s
,
incl
ud
i
ng
supporti
ng
ve
ct
or
m
achine
s
[8
-
10
]
,
l
og
i
sti
cal
regressio
n [
11
]
, r
a
ndom
f
orest
s
[
3,
12]
, neu
ra
l netw
orks
[
4,
13
]
,
and tim
e s
eries m
od
el
ing t
echn
i
qu
e
s
[14].
In
s
umm
ary,
th
e
pap
e
r
f
ocuse
s
on
var
i
ou
s
di
sease
pr
e
dicti
on
m
od
el
s
of
m
achine
le
ar
ning.
T
he
pa
per
is
struct
ur
e
d
i
n
the
f
ollo
wing
way.
Sect
io
n
2
e
xp
la
in
s
t
he
basic
c
on
c
ept
of
m
achin
e
le
arn
i
ng,
pr
e
dicti
on
m
od
el
s
and
ty
pes
of
pre
dicti
on
m
od
el
s.
Sec
ti
on
3
e
xp
l
ai
ns
the
su
r
vey
of
disease
predict
ion
m
od
el
s.
Se
ct
ion
4
exp
la
in
s
the
c
om
par
at
ive
study
of
hea
rt
dise
ases.
Sect
i
on
5
su
m
m
arizes
a
com
par
at
ive
st
ud
y
of
br
east
c
ancer
.
Sect
ion
6
give
s
the c
on
cl
us
io
n an
d fu
t
ur
e e
nh
ancem
ent.
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.
10
, No
.
4
,
A
ugus
t
2020
:
4217
-
422
5
4218
2.
INTROD
U
CTION
T
O
MACHI
NE LEA
RN
I
NG
Ma
c
hin
e
le
a
rn
i
ng
(ML
)
is
an
arti
fici
al
intel
lig
ence
(
A
I)
bra
nch
that
help
s
analy
ze
the
dat
a
struc
ture
and
fit
the
in
form
ation
into
m
od
el
s
c
orrectl
y.
It
is
one
of
th
e
com
pu
te
r
sci
ence
fiel
ds,
a
nd
di
ff
e
rs
f
ro
m
oth
e
r
com
pu
ti
ng
te
c
hnology
by
th
e
way
of
trai
ni
ng
t
he
c
om
pute
rs
base
d
on
data
inputs
an
d
it
us
es
sta
ti
sti
cal
analy
sis
to
get
the
pro
per
ou
tpu
t.
For
t
his
reason,
ML
is
us
e
d
i
n
a
utom
at
ed
decisi
on
-
m
aking
m
od
el
s
li
ke
Faci
al
r
eco
gnit
ion
,
Recom
m
e
nd
at
io
n
e
ngine
s,
OCR a
nd S
e
lf drivin
g
ca
r
a
pp
li
cat
io
ns
.
Ma
chine
le
a
r
ni
ng
m
et
ho
ds
are
cat
eg
or
iz
e
d
into
t
hr
ee
cl
assifi
cat
ion
s
accor
ding
to
the
trai
ning
processes
us
e
d.
T
he
cat
egories
are
super
vised
m
achine
le
arn
in
g,
uns
up
e
r
vised
m
a
chine
le
ar
ning
an
d
reinfo
rcem
ent
le
arn
in
g.
In
s
uper
vised
le
ar
ni
ng,
the
data
sa
m
ples
with
categ
ory
la
bels
are
us
e
d
in
trai
ning.
Cl
assifi
cat
ion
and
re
gr
essi
on
m
od
el
s
are
e
xam
ples
of
s
uper
vised
le
a
rni
ng
.
The
al
gor
it
h
m
s
us
ed
in
this
appr
oach
a
re
de
ci
sion
tree,
na
ïve
bayes
et
c.
In
un
s
uper
vis
ed
le
arn
i
ng,
th
e
data
sam
ples
are
directl
y
use
d
in
trai
ning
wi
th
out
cat
egory
la
bel.
Cl
us
te
ring
t
echn
i
qu
e
s
an
d
encode
rs
are
the
basic
exam
ples
of
unsupe
rv
ise
d
appr
oach.
Re
inf
or
cem
ent
le
arn
i
ng
is
a
m
ixtur
e
of
pri
or
t
w
o
ap
proac
hes.
It
us
es
a
ge
nt
that
fin
ds
th
e
c
orrect
act
ion
to
ac
hieve
the
ove
rall
g
oal
of the
app
li
cat
io
n
[
15
]
.
3.
PREDI
C
TIO
N
MO
DELS
A
predict
io
n
m
od
el
is
c
har
act
erized
as
a
m
od
el
that
pro
vide
s
a
way
to
ev
al
uate
the
ind
i
vidual
da
nge
r
of
a
patie
nt fo
r
the outcom
e o
f
a d
ise
ase
. T
he
q
ue
sti
on
of
wh
e
n,
w
hat and ho
w
to
us
e th
ese m
od
el
s ar
ise
s w
it
h
the
gro
wth
of
su
c
h
predict
io
n
m
od
el
s.
T
he
se
m
od
el
s
can
be
ta
ug
ht
ove
r
tim
e,
prov
i
din
g
t
he
dem
and
s
of
the co
m
pan
y,
to react
to ne
w i
nfor
m
at
ion
or
views
.
Tw
o
ty
pes
of
pr
e
dicti
on
m
od
el
s
exist.
T
he
y
are
m
od
el
s
of
cl
assifi
cat
ion
pr
e
dicti
ng
cl
ass
outc
om
es,
and
m
od
el
s
of
re
gress
io
n
predict
in
g
the
relat
ion
s
hip
be
tween
a
res
ponse
va
riable
Y
a
nd
a
predict
or
var
ia
ble
X.
Va
rio
us
basic
a
nd
ad
va
nced
al
gorithm
,
li
ste
d
in
Fig
ure
1
,
Figure
2
c
onduct
s
data
anal
ysi
s
and
sta
ti
sti
cal
anal
ysi
s
and
dete
r
m
ine
inf
or
m
at
i
on
tre
nd
s
an
d
patte
rn
s
.
Whi
l
e
m
achine
le
arn
i
ng
an
d
pre
dicti
on
analy
sis
can
prov
i
de
an
op
por
tun
it
y
fo
r
a
ny
i
m
ple
m
entat
io
n,
the
ha
phaza
rd
im
ple
m
enta
ti
on
of
the
se
opti
ons
will
dr
ast
ic
al
ly
i
m
ped
e
their
abili
ty
to
pro
vid
e
i
ns
ig
ht
into
the
de
m
and
s
of
t
he
orga
nizat
ion
without
consi
der
i
ng
h
ow
t
hey
fit
into
e
ver
y
day
op
e
rati
ons.
Orga
nizat
ion
s
need
t
o
gua
ran
te
e
that
th
ey
have
the
arc
hitec
tur
e
in
place
to
s
upport
these
al
te
rn
at
ives,
a
s
well
as
high
-
qual
it
y
info
rm
ation
to
feed
t
he
m
and
assist
them
lea
rn, to
m
ake th
e m
os
t of
pr
e
dic
ti
on
a
naly
ti
cs
and m
achine learn
i
ng [16
]
.
Figure
1.
Ba
sic
predict
io
n
t
ec
hn
i
qu
e
s
Figure
2. A
dva
nced p
re
dicti
on
t
ech
niques
4.
E
X
ISTI
NG
T
ECHNIQ
UES
FO
R DISEA
SE PRE
DI
CTION
Mi
ng
yu
Pak,
Mi
young
S
hi
n
[17]
,
e
xplo
re
d
fe
w
e
nvir
onm
ental
var
ia
bl
es
li
nk
e
d
t
o
ty
pe
-
2
diabetes
d
i
s
e
a
s
e
a
n
d
s
e
l
e
c
t
e
d
s
om
e
v
a
r
i
a
b
l
e
s
t
o
d
e
v
e
l
o
p
a
n
a
n
a
l
y
t
i
c
a
l
p
r
o
t
o
t
y
p
e
o
f
p
r
e
d
i
c
t
i
o
n
o
f
d
i
s
e
a
s
e
.
F
o
r
t
h
e
cho
ic
e
of
im
po
rtant
va
riables,
they
first
pr
e
-
proce
ssed
al
l
external
env
ir
on
m
ental
factor
s
into
nu
m
erical
dat
a
an
d
then
est
im
at
e
d
the
m
axim
um
/
m
ini
m
u
m
p
r
obabili
ty
rati
os
of
al
l
the
s
or
te
d
exoge
nous
va
riables.
The
top
-
n
pos
it
ion
ed
va
riabl
es
wer
e
the
n
chosen
as
input
var
ia
bles
f
or
the
forecast
m
od
el
dep
en
din
g
on
ansa
n/ans
ung
coho
rt
2
data
c
ollec
te
d
from
t
he
K
or
ea
n
Nati
on
al
I
ns
ti
tute
of
Healt
h
(
KNIH),
the
d
ise
as
e
ris
k
factor
pr
e
dicti
on
m
od
el
was
create
d
us
i
ng
SV
M.
T
heir
predict
ion
m
od
el
sh
owed
the
outp
ut
of
65.
97
per
ce
nt
pr
eci
sio
n
a
nd
sh
owe
d
ver
y
i
den
ti
cal
pe
rfo
r
m
ance
only
w
it
h
ge
netic
fac
tors
with
par
ti
cular
e
nv
i
ron
m
enta
l
var
ia
bles
to t
he
m
od
el
[
17]
.
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
A com
pr
ehe
ns
i
ve stu
dy on dis
ease r
isk
p
re
di
ct
ion
s i
n ma
c
hi
ne
le
arni
ng
(
G. S
ar
any
a
)
4219
The
re
sea
rc
h
s
how
n
in
[
18
]
f
ocuses
on
dise
ase
forecast
in
g
fr
om
m
edical
inform
at
ion
su
ppli
ed
by
New
Y
ork'
s
Pre
sb
yt
erian
H
ospit
al
.
Since
t
he
se
are
m
edical
inf
or
m
at
ion
,
pr
edict
ing
c
om
pu
te
rized
predic
ti
on
s
is
gen
erall
y
uniqu
e
an
d
sim
pl
er
than
f
oreca
s
ti
ng
us
e
r
te
xt
inputs.
Nico
la
e
Dragu
et
al
disc
us
se
d
-
on
forec
ast
of
serio
us
c
onta
gi
ou
s
disea
ses
from
web
m
at
erial
so
urces,
w
hich
is
al
so
a
sp
eci
fic
or
i
gin
w
her
e
ope
n
c
li
nical
te
rm
s
are
us
ed
[19]
.
Ma
ny
at
tem
pts
hav
e
been
m
ade
to
f
or
ecast
sel
ect
ive
diseases
[
20,
21]
.
For
i
nst
ance,
the
auth
ors
in
[20]
deal
wit
h
the
pr
e
dicti
on
by
m
ining
c
on
te
nt
of
co
ronar
y
he
art
dis
ease.
The
re
ar
e
al
so
an
im
po
rtant
a
m
ou
nt
of
st
udy
wor
ks
on
m
edici
nal
se
rv
ic
es
de
bates
t
hat
ha
ve
bee
n
ca
r
ried
out.
La
v
petr
ov
us
e
d
NLP
to
e
valuate
an
d
break
dow
n
us
e
r
re
m
ark
s
to
predict
disease
and
c
on
ce
ntrate
un
c
omm
on
resp
onses
to dr
ug
s
[
22]
.
R
y
an
McDon
al
d
et
al
,
u
se
a
te
rm
ino
log
y
-
la
de
n
inter
face
(i.
e.
cl
ie
nts
need
to
ex
plo
re
a
l
ong
li
st
of
hin
ts)
.
It'
s
an
a
wkward
jo
b
from
the
us
er'
s
view,
a
nd
the
op
e
rati
on
is
als
o
te
dious.
Als
o,
if
the
custo
m
ers
do
no
t
fin
d
a
ce
rt
ai
n
in
dicat
ion
,
they
are
forced
to
pre
ven
t
t
ha
t
sy
m
pto
m
that
is
not
in
a
ny
m
ann
er
desi
re
d
[
23]
.
Cl
ie
nt'
s
directe
d
te
xt
in
pu
t
[
24,
25
]
.
T
hey
re
ly
on
ne
gligibl
e
sy
m
pto
m
-
dis
ease
relat
ion
s
hi
p
syst
e
m
[2
6,
27
]
in
any
sit
uation
a
nd
us
e
com
plete
con
te
nt
data
base.
These
f
ra
m
ewo
r
ks
be
gin
to
searc
h
f
or
accurate
te
rm
m
at
ch
at
the
input
of
the
cl
i
ent
from
each
inp
ut
li
ne
in
the
da
ta
base.
F
or
e
xam
ple,
if
a
cl
ie
nt
with
sy
m
pto
m
s
equ
i
valent
is
not
sp
eci
fied
i
n
the
data
base
,
then
we
can
not
m
a
tc
h
the
in
f
or
m
at
ion
co
rr
e
ct
ly
.
If
the
us
e
r
in
put
sh
oul
d
co
ntain
a
gr
eat
e
r
num
ber
of
sem
i
-
tech
nical
te
rm
s
than
a
ntici
pated,
it
will
de
grade
it
s
pe
rform
ance.
The
syst
em
u
sed
is es
pecial
ly
stron
g
a
nd r
est
rict
ed
to
p
a
rtic
ular dat
a ty
pes
.
Xiao
y
an
W
ang
et
al
,
pro
po
se
d
an
a
uto
m
at
ed
syst
e
m
fo
r
dis
ease
predict
io
n
that
base
d
on
the
cl
ie
nt'
s
dr
i
ven
fee
db
ac
k.
T
heir
struct
ur
e
re
qu
ire
s
in
pu
t
from
the
cl
ie
nt
su
ch
as
t
he
nam
es
of
s
ym
pto
m
s
and
certai
n
sign
ific
a
nt
para
m
et
ers
and
prov
i
des
a
li
st
of
li
kel
y
il
lnesses
(m
axi
m
u
m
i
nf
ect
io
ns
are
m
or
e
l
ikely
to
occur)
.
The
acc
ur
acy
of
t
he
a
uto
m
a
te
d
disease
pr
edict
ion
sc
he
m
e
(A
D
PS)
e
valuated
the
s
olu
ti
on
of
t
he
curre
nt
schem
e
with
a
sta
nd
ar
d
14.
35
perce
nt
great
er
accuracy
in
exam
inati
on
[28].
S
Ma
nim
ek
al
a
et
a
l,
s
ugge
ste
d
that
the
Au
tom
at
ic
Disease
Pr
edict
ion
te
ch
ni
qu
e
determ
ine
the
m
os
t
pr
oba
ble
disease
bas
ed
on
the
fee
dback
of
t
he
patie
nt
t
hat
facil
it
at
es
early
detect
ion.
T
he
m
od
el
use
s
data
m
ining
al
go
rithm
apr
iori
-
f
reque
nt
patte
r
n
(A
-
FP
)
t
o
id
en
ti
fy
il
lnesses
t
hro
ugh
healt
h
da
ta
m
ining
f
oc
us
e
d
on
the
sig
ns
of
in
put.
T
hi
s
m
e
tho
d
is
use
d
for
fin
ding m
edica
l datase
ts fr
om
w
hich
to c
reate associa
ti
on
rules. Th
e
goal
is to id
entify
f
r
om
the h
eal
th d
at
aset
releva
nt and
frequ
e
nt il
lnesse
s
[
29
]
.
Cristi
nel
Ab
abe
i
et
a
l,
dis
cussed
the
ove
rall
co
ntext
of
com
pu
te
r
f
ram
ewo
r
ks
on
m
ulti
cor
e
process
or
syst
e
m
s,
they
pro
vid
e
d
a
discuss
i
on
on
the
m
os
t
com
m
on
pr
e
dict
ion
a
nd
cl
a
ssific
at
ion
m
et
hods
.
They
intr
oduce
d
s
om
e
pr
edict
ion
syst
em
s
from
si
m
ple
to
m
or
e
com
plicated,
wh
il
e
hi
ghli
g
hting
one
frequ
e
nt
basic
them
e:
e
ach
of
thes
e
syst
e
m
s
m
isuses
the
pri
or
histo
r
y
of
the
va
riab
le
of
inte
rest
[
30
]
.
Anki
ta
De
wan
et
al
,
i
n
c
onjun
ct
ion
with
t
he
back
-
prop
a
gati
on
m
et
ho
d,
t
he
y
su
ggest
e
d
a
su
ccess
f
ul
ge
netic
al
gorith
m
fo
r
pr
e
dicti
ng
hea
rt
diseas
e
an
d
create
d
a
m
od
el
that
co
uld
identify
an
d
extract
un
known
data
(
patte
rn
s
a
nd
relat
ion
s
hip
s
)
li
nk
e
d
to
c
oro
na
ry
hear
t
disea
se
from
arch
iv
ed
data
base
re
cords
of
hear
t
disease
[
31]
.
Table
1
sh
ows the
diff
e
ren
t m
et
ho
ds
an
d
thei
r
ad
van
t
ages and
disa
dvanta
ge
s.
T
he
m
erit
s an
d
de
m
erit
s ar
e sp
eci
fied
in
te
rm
s o
f
ef
fici
ency,
pr
e
dicti
on accu
racy an
d how ea
sil
y a m
od
el
c
ould
be
im
ple
m
ented.
Table
1.
Me
rits an
d dem
erit
s o
f pr
e
dicti
on
tec
hn
i
qu
e
s
Metho
d
s
Pros
Co
n
s
KNN
Si
m
p
le,
h
ig
h
acc
u
r
acy
Local in
f
o
r
m
atio
n
stru
ctu
re
sen
sitiv
ity
.
Cal
cu
lated
slo
w
ly
.
Linear Regr
ess
io
n
Accurac
y
of
good
p
redictio
n
Tr
ain
in
g
n
eeds
.
W
h
en
f
resh
info
r
m
ati
o
n
ar
rive,
co
ef
f
icien
ts
m
a
y
n
eed to
be u
p
d
ated
con
stan
tly
.
SVM
Practicall
y
ef
f
icien
t,
v
er
y
po
p
u
lar
The size of
the tra
i
n
in
g
in
f
o
r
m
at
io
n
can i
m
p
rov
e the
n
u
m
b
er
of
bias
f
u
n
ctio
n
s. Selection
of
para
m
ete
rs
d
ep
en
d
s o
n
in
f
o
r
m
atio
n
.
Bayes cl
ass
if
ier
h
ig
h
acc
u
racy
,
ef
f
icien
t
Tr
ain
in
g
n
eeds
LDA
Fast, I
m
p
le
m
en
tati
o
n
is co
m
p
a
rativel
y
si
m
p
l
e
Tr
ain
in
g
n
eeds
4.1.
Descripti
on
of
existin
g
tech
n
iques f
or hear
t
dise
as
e
predi
ction
P.
K.
A
nooj
[32],
s
uggeste
d
a
decisi
on
-
m
a
king
sc
hem
e
based
on
fu
zzy
ru
le
s
t
o
as
sess
the
le
vel
of
risk
of
hea
rt
disease.
He
fi
rst
pre
-
process
ed
the
i
nfor
m
at
ion
f
or
m
iss
ing
values
in
his
m
et
ho
d.
H
e
the
n
perfor
m
ed
fu
z
zy
set
s
ge
nerat
ion
a
nd
cre
at
ed
a
f
uzzy
decisi
on
-
m
aking
syst
em
.
T
he
te
ch
nique
sel
ect
s
the
re
qu
i
red
a
tt
ribu
te
ba
sed
on
t
he
nu
m
ber
of
e
ven
ts
i
n
the
databas
e
to
create
w
ei
gh
te
d
f
uzzy
la
ws
.
These
wei
gh
te
d
f
uzzy
ru
le
s
are
then
us
e
d
to
create
a
de
ci
sion
-
b
ase
d
s
chem
e
of
assist
ance.
He
te
st
ed
his
su
ggest
e
d
sc
he
m
e
on
th
ree
di
sti
nct
dataset
kinds
that
a
re
colle
ct
ed
f
ro
m
V.A.
Cl
evela
nd
dataset
.
M
edical
centre
with
303
cases
of
trai
ning
data
f
or
202
docum
ents
an
d
te
st
inf
orm
ation
f
or
10
1
doc
um
ents.
P.
K
.
Ano
oj c
ontrast
ed his m
od
el
wi
th m
od
el
s b
as
ed on ne
ural
n
e
tworks
and
obta
ined
t
he great
est
p
recisi
on [3
2]
.
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.
10
, No
.
4
,
A
ugus
t
2020
:
4217
-
422
5
4220
Tsipou
ras
M.
G.
,
et
al
.
[
33]
su
ggest
ed
a
n
aut
om
at
ed
syst
e
m
based
on
f
uzzy
m
od
el
in
g
a
nd
inf
or
m
at
ion
m
ining
to
pr
e
di
ct
hear
t
diseas
e.
His
m
od
el
inclu
des
m
ea
su
r
es
su
c
h
as
inferrin
g
in
f
orm
at
ion
decisi
on
tree,
e
xtracti
ng
guide
li
nes,
f
or
m
ulatin
g
c
risp
m
od
el
,
trans
f
or
m
ing
crisp
m
od
el
int
o
f
uzzy
m
od
el
,
an
d
op
ti
m
iz
ing
it
.
The
inf
orm
a
ti
on
gat
her
e
d
fr
om
Invasive
Ca
r
dio
lo
gy
,
Un
i
ver
sit
y
Dep
a
rtm
ent,
Io
an
nina
Ho
s
pital
. T
he
t
echn
i
qu
e
pr
ov
i
des 8
0% pr
eci
s
ion
i
n
a
war
e
ne
ss and
65%
pr
e
ci
sion
i
n
s
pecif
ic
it
y [3
3]
.
Chait
rali
S.
D
ang
a
re
et
al
.
[34]
,
us
e
d
he
art
disease
pr
edict
ion
data
m
ining
m
et
ho
d.
He
first
pre
-
pr
ocesse
d
the
m
issi
ng
inf
or
m
at
ion
us
in
g
the
m
ean
m
o
de
te
ch
nique
a
nd
us
e
d
t
he
pe
rcep
tr
on
m
ulti
-
la
ye
r
m
od
el
to
m
ap
the
in
form
ation
.
T
o
a
naly
se
the
data
base
of
hea
rt
disease
,
Nai
ve
bays,
neural
netw
orks
a
nd
decisi
on
tree
w
ere
us
e
d.
He
ga
there
d
303
do
cum
ents
from
t
he
reposit
ory
of
Cl
evela
nd
he
art
disease
a
nd
use
d
it
as
a
trai
ning
se
t
and
gat
he
red
270
docu
m
ents
as
te
st
inf
or
m
at
ion
f
rom
the
reposit
ory
of
Stat
log
Hear
t
Disease.
T
he
inf
or
m
at
ion
set
con
sist
s
of
at
tribu
te
,
i
nput
and
key
predi
ct
able.
His
m
od
el
prov
i
des
100%
accuracy
for n
eur
al
netw
orks
, D
eci
si
on tree
with
99.62 a
nd
N
aï
ve
b
ay
es
wi
th 90.74%
pr
e
ci
sion
[34].
Ma
i
Sh
ou
m
an
et
al
.
[35]
,
by
inco
rpor
at
i
ng
decisi
on
tree
and
k
-
m
ea
ns
cl
us
te
ri
ng,
su
ggest
ed
a
te
chn
iq
ue
f
or
diag
nosin
g
pa
ti
ents
with
he
art
disease.
F
or
k
-
m
eans,
the
m
et
ho
d
util
iz
es
centro
i
d
sel
ect
ion
te
chn
iq
ue
an
d
decisi
on
tree
is
us
e
d
t
o
deter
m
ine
the
cl
us
t
ers.
Thirtee
n
di
sti
nct
char
act
erist
ic
s
are
gat
her
e
d
from
the
Cl
eveland
Cl
inic
F
oundat
ion
Hea
rt
Disease
.
Co
m
par
ed
to
t
he
current
decisi
on
tree
,
the
inte
gr
at
e
d
m
od
el
o
f k
-
m
eans a
nd
decisi
on tree
obtai
ne
d g
reater
ou
tc
om
es o
f
83.
9% [
35
]
.
S.
Pal
et
al
[36
]
,
su
ggest
e
d
usi
ng
in
f
or
m
at
ion
m
ining
to
pr
edict
hear
t
dise
ase.
He
s
urvey
ed
3
disti
nct
cl
assifi
ers
s
uc
h
as
C
ART,
ID3
a
nd
tree
of
Decisi
on.
The
in
f
or
m
at
i
on
set
f
ro
m
the
Cl
evela
nd
Cl
inic
Foundati
on
s
hows
that
83.
49
per
ce
nt
of
the
cl
assifi
cat
ion
and
re
gr
essi
on
tree
(CART
)
pr
eci
sio
n
was
m
uch
bette
r
tha
n
the
decisi
on
tree
and
ID3
(I
te
ra
ti
ve
Dich
oto
m
iz
ed
3)
[
36
]
.
H.
A.
H
uijer
et
al
.
[37]
,
de
sign
e
d
a
decisi
on
sup
port
serv
ic
e
to
find
unre
st
transiti
on
th
rou
gh
decisi
on
tr
u
st
m
easur
e,
tr
us
t
-
base
d
S
V
M
and
trust
-
base
d
m
ulti
le
vel
SV
M
to
disc
over
a
gi
ta
ti
on
transiti
on.
240
S
am
ples
are
gathe
r
ed
via
hum
an
body
sens
or
s
.
T
he
patie
nt
ex
pe
riences
disti
nctive
te
ns
i
on
in
ven
t
or
y
of
sta
te
-
qual
it
y
scale
(T
-
ST
AI),
use
d
t
o
cal
culat
e
un
c
om
fo
rtable
adu
l
ts.
T
he
te
ch
nique
pro
vide
s
91
.4
per
ce
nt
prec
isi
on
w
he
n
co
m
par
ed
to
trad
it
ion
al
vecto
r
s
upporti
ng m
achine w
i
th 90.9
pe
rcen
t
preci
sion [
37]
.
Lat
ha
Pa
rthib
a
n
et
al
.
[38]
,
ou
tl
ine
d
a
pr
e
dicti
on
te
c
hn
i
qu
e
f
or
sm
art
hear
t
disease
pr
e
dicti
on
.
The
m
et
ho
d
is
execu
te
d
u
sin
g
coacti
ve
ne
uro
f
uzzy
in
fer
e
nc
e
syst
e
m
,
neural
net
work,
a
nd
gen
et
ic
al
go
r
it
h
m
.
The
dataset
is
gathe
red
f
ro
m
UCI
a
nd
t
he
prot
otype
is
sim
ulate
d
us
in
g
N
euro
So
l
ution
So
ft
war
e
.
T
he
m
ean
sq
ua
re
er
r
or
of
CAN
FIS
wa
s
0.000
842
[
38]
.
N.
Dee
pika
et
al
.
[3
9],
su
g
geste
d
a
hear
t
at
ta
ck
pa
ti
ent
cl
assifi
cat
ion
m
od
el
.
He
pr
e
-
proce
ssed
his
in
form
ation
set
s
f
or
m
is
sing
val
ues
a
nd
the
n
im
plem
ented
the
sam
e
width
binnin
g
inte
rv
al
ap
proac
h.
Then
nu
m
erical
par
am
et
ers
are
transfo
rm
ed
into
cat
eg
ori
cal
par
am
et
ers
and
fr
e
quent
patt
ern
s
ar
e
m
ined
base
d
on
pr
un
i
ng
-
cl
assifi
c
at
ion
r
ule
al
gorithm
li
nk
ed
to
hea
rt
disease.
His m
od
el
us
e
d
a
n
e
f
fecti
ve
f
oreca
st
of
par
ti
cular
class l
abel
[39].
T.
Tu
rn
e
r
[
40]
pro
po
se
d
the
i
dea
of
dia
gnos
i
ng
h
eart diseas
e
by
com
bin
in
g
naï
ve
bayes and
k
-
m
ean
s
cl
us
te
rin
g
with
disti
nct
ch
oice
of
ce
ntr
oid
s
.
Cl
eveland
cl
ini
c
fou
nd
at
io
n
c
ollec
ts
the
data
set
.
The
preci
sion
of
the
em
bed
ded
k
-
k
-
is
84.5
pe
r
cent
com
par
ed
to
the
tradit
io
nal
al
gorithm
[4
0].
Uzm
a
An
sari
et
al
.
[41]
us
e
d
weig
hted
asso
c
ia
ti
ve
cl
assifi
er
to
de
velo
p
a
m
od
el
for
pr
e
di
ct
ing
hear
t
at
t
acks.
Th
e
data
set
is
gat
her
e
d
fro
m
the
ML
databa
se
of
Irvine
Universit
y
of
Ca
li
forn
ia
(
UCI
).
He
us
e
d
2
cl
ass
la
bels
1
in
his
m
od
el
f
or
"
N
o
hea
r
t
disease
"
a
nd
ot
her
one
for
"
H
eart
disease
"
r
at
her
t
han
getti
ng
5
cl
as
s
la
be
ls
with
1
f
or
no
hear
t
disease
an
d
4
for
fou
r
hear
t
disease
kinds.
He
us
e
d
80%
of
co
nfi
de
nce
value
an
d
25
%
of
sup
port
value.
T
he
pro
toty
pe
pro
po
se
d
achi
eves
preci
sio
n
of
81.
51
pe
r
cent.
He
co
nc
lud
es
that
the
m
easur
ed
ass
ociat
ive
cl
assifi
er
is
the
eas
ie
st
wa
y
to
ac
qu
i
re
e
ff
ic
ie
nt
im
po
rt
ant
patte
rn
s
f
r
om
inform
ation
set
ti
ng
for
c
ard
ia
c
disease
[
41
]
.
The bel
ow Ta
bl
e 2
s
hows
the
com
par
at
ive s
urvey
of
hea
rt d
i
sease t
ech
niqu
es whic
h we
h
a
ve discus
sed
a
bove.
Table
2.
T
he
c
om
par
at
ive s
urvey o
f heart
di
sease
Au
th
o
rs
Year
Data set
Metho
d
o
lo
g
y
Accurac
y
Latha Pa
rthib
an
2007
Un
iv
ersity
of
Calif
o
rnia Ir
v
in
e (
UCI)
Neu
ral
n
etwo
rk an
d
Genetic algo
rith
m
MSE
-
.00
0
8
4
2
Tsip
o
u
ras M
.
G
2008
Inv
asiv
e Cardio
lo
g
y
,
Ho
sp
ital of
L
o
an
n
in
a
Fu
zzy
Mod
ellin
g
and
Data
Minin
g
80%
H.
A
.
Hu
ije
r
2010
Sen
so
r
Data
SVM
9
1
.4%
N.
De
ep
ik
a
2011
Un
iv
ersity
of
Calif
o
rnia Ir
v
in
e (
UCI)
Prun
in
g
-
Clas
sif
ica
tio
n
Asso
ciatio
n
r
u
le
Predicts
ef
f
ectiv
ely
Uz
m
a
Ans
ari
2011
Un
iv
ersity
of
Calif
o
rnia Ir
v
in
e (
UCI)
W
eig
h
ted
Asso
ciat
iv
e Class
if
ier
8
1
.51
%
P.
K.
Ano
o
j
2012
Clev
elan
d
,
Hu
n
g
arian I
n
stitu
te of
Card
io
lo
g
y
,
Un
iv
ersity
Ho
sp
ital,
Zuri
ch
,
Switzerland
W
eig
h
ted
Fuzzy
r
u
les
5
7
.85
%
Ch
aitrali
S.
2012
Clev
elan
d
heart dis
ease datab
ase
Naïv
e baye
s
,
Neu
r
al netwo
rk an
d
Decisio
n
tr
ee
100%
Mai
Sh
o
u
m
an
2012
Clev
elan
d
heart dis
ease datab
ase
Cen
troid
Selection
T
echn
iq
u
e
8
3
.9%
T.
Turn
er
2012
Clev
elan
d
heart dis
ease datab
ase
Integ
rated k
-
m
ean
s an
d
Naïve bayes
8
4
.5%
S.
Pal
2013
Clev
elan
d
heart dis
ease datab
ase
CAR
T,
ID3,
Decis
io
n
T
ree
8
3
.49
%
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
A com
pr
ehe
ns
i
ve stu
dy on dis
ease r
isk
p
re
di
ct
ion
s i
n ma
c
hi
ne
le
arni
ng
(
G. S
ar
any
a
)
4221
4.2.
Descripti
on
of
existin
g
tech
n
iques f
or bre
ast c
an
cer
pred
ic
tion
A
f
uzzy
m
od
el
was
create
d
by
Yassi
et
al
.
[
4
2]
to
diff
e
re
ntiat
e
between
norm
al
and
m
ali
ci
ou
s
br
ea
st
cancer
.
T
he
t
echn
i
qu
e
bro
ught
disor
der
i
nto
t
he
hie
rarc
hical
cl
us
te
r
of
pa
rtia
l
swa
rm
enh
anc
em
ent
of
m
ul
ti
sp
eci
es,
prom
pting
the
i
m
pr
ovem
ent
of
c
hao
ti
c
hierar
c
hical
c
luster
-
base
d
m
ul
ti
sp
eci
es
swar
m
enh
a
ncem
ent
of
par
ti
cl
es
(C
H
CM
SPSE).
C
HCMSPSE
helps
t
o
disti
ngui
sh
t
he
form
of
cance
r
of
the
br
east
and
to
e
nhance
the
f
uzzy
ru
le
s.
T
he
m
od
el
al
so
di
sc
overs
fu
zzy
ru
le
s
ve
r
y
correct
ly
.
Th
e
dataset
is
gather
e
d
from
the
m
achine
le
arn
i
ng
da
ta
base
of
I
r
vin
e
U
niv
e
rsity
of
Ca
li
f
ornia
(
UCI).
T
hus,
f
or
w
or
ld
wide
searc
h
capab
il
it
y,
the
te
chn
iq
ue
util
iz
es
11
c
ha
otic
m
aps.
Sin
usoidal
cha
otic
m
ap
acqu
i
red
99
perce
nt
pr
e
ci
sion
from
tho
se
m
aps
beca
us
e
it
m
a
tc
hed
with
the
posit
ion
of
the
issue.
The
m
od
el
achieves
m
or
e
th
a
n
90%
pr
eci
sio
n [42].
In
de
fining
5,
10
an
d
15
ye
ar
s
of
s
pecific
breast
cance
r
s
ust
ai
nab
il
it
y,
Lu
nd
i
n
M
et
al
.
[
43
]
pro
vid
e
d
a
prototype f
or
acce
ssing
t
he
pr
eci
sio
n
of
ANN.
The
data
s
ource
is
colle
ct
ed
f
ro
m
Ci
ty
H
os
pital
of
T
urk
u
an
d
Turk
u
U
niv
e
rs
it
y
Ce
ntral
Ho
sp
it
al
with
951
instances.
I
n
that
trai
ning
set
of
65
1
instan
ces
and
a
vali
datio
n
set
of
300
ins
ta
nces.
T
his
prot
otype
com
par
es
the
ou
tc
om
es
of
arti
fici
al
neural
net
work
a
nd
lo
gisti
cal
regressio
n.
T
he
preci
sion
of
br
east
ca
nce
r
s
pecific
s
urviva
l
for
5
ye
ars
re
ported
as
0.909,10
ye
ars
repor
te
d
a
s
0.086 a
nd
15 yea
rs rep
or
te
d
a
s 0
.
883 [
43]
.
Delen
D
et
al
.
[
44]
i
m
ple
m
ented
a
data
m
ining
te
c
hn
i
qu
e
com
par
is
on
a
ppr
oach
th
at
involves
log
ist
ic
al
regr
ession,
decisi
on
tree,
a
nd
ar
ti
fici
al
neu
ral
netw
ork
to
predict
breast
cancer
de
velo
pm
ent.
The
m
od
el
util
iz
es
ov
er
200,0
00
cases
of
an
en
or
m
ou
s
i
nfor
m
at
ion
reposit
or
y.
T
hus,
log
ist
ic
al
reg
r
e
ssion
pr
eci
sio
n
is
89.
2%,
decisi
on
tree
pr
eci
si
on
(C5)
93.6%
an
d
arti
fici
al
neural
net
work
91.
2%.
10
-
f
old
cro
ss
-
validat
io
n
f
or
in
form
a
ti
on
te
sti
ng
,
unbiase
d
est
im
at
ion
m
easur
em
ent
and
3
t
echn
i
qu
e
s
pre
dicti
on
.
Re
search
i
nd
ic
at
es
that
th
e
sel
ect
ion
trees
is
the
best
dete
r
m
inant
m
et
ho
d
for
de
fining
breast
cance
r
gr
ow
t
h
relat
ive to t
he a
rtific
ia
l neural
n
et
w
ork
and l
og
ist
ic
r
e
gressi
on [4
4].
Be
ll
aachia
Abdelg
han
i
et
al
.
[45]
us
e
d
in
form
ation
m
in
ing
te
ch
niques
to
pr
ese
nt
a
m
od
el
fo
r
pr
e
dicti
n
g
br
ea
st
cancer
de
vel
op
m
ent.
The
pre
-
cl
assifi
cat
io
n
proce
ss
is
carried
out
in
thr
ee
areas:
recov
ery
of
essenti
al
sta
tus,
recove
ry
of
s
urvival
ti
m
e
and
cause
of
deat
h.
T
hr
ee
te
c
hn
i
qu
e
s
of
m
achi
ne
le
ar
ning,
na
m
el
y:
neural
net
w
ork
prop
a
gate
d
back,
naïve
ba
ye
s
and
C4.
5
f
or
cl
assifi
c
at
ion
perform
ance.
T
he
dat
a
set
is
gathe
red
f
ro
m
the
Nati
onal
C
ancer
I
ns
ti
tute
(N
CI
)
Surv
ei
ll
ance,
Epi
dem
i
ology,
a
nd
En
d
Re
s
ults.
T
he
re
are
151.
886 rec
ord
s in
t
he data
set
, w
it
h 1
6
c
ha
ra
ct
erist
ic
s.
The
m
od
el
p
r
ov
i
de
s roug
hly 8
7% pr
eci
sio
n [
45
]
.
H.
Ko
y
un
c
u
et
al
.
[46]
i
m
plem
ented
a
bio
m
edical
patte
rn
base
d
on
arti
fici
al
ne
ur
al
netw
ork
ro
ta
ti
onal
fores
t
al
go
rithm
(RF
-
A
N
N).
Mult
il
ay
er
per
ce
ptron
was
us
e
d
as
the
cl
assifi
er
of
base
an
d
the
m
od
el
us
e
d
RF
al
gori
thm
as
the
cl
a
ssifie
r
of
e
ns
e
m
b
le
.
Usin
g
th
e
m
ai
n
com
po
nen
t
asse
ssm
e
nt,
di
ff
e
ren
t
functi
on
set
s ar
e
gathe
r
ed fr
om
the infor
m
at
ion
set. T
he precisi
on
of
RF
-
ANN
is t
he
refor
e
98.0
5 p
ercent
[46].
S.
M.
Jam
aran
i
H
et
al
.
[
47
]
pro
posed
a
m
eth
od
for
recog
ni
zi
ng
the
disea
se
an
d
helping
rad
i
olog
ist
s
pr
e
dict
breast
cancer
.
T
he
m
od
el
c
om
bin
es
deco
m
po
sit
io
n
of
a
rtific
ia
l
neural
net
wor
k
(AN
N)
a
nd
sub
-
ba
nd
picture
base
d
on
m
ulti
wav
el
et
.
The
te
ch
niq
ue
is
stu
die
d
us
in
g
m
a
m
m
o
gr
a
phic
databa
se
by
m
a
mm
o
gr
a
phic
i
m
age
analy
sis
so
ci
et
y
(MI
AS
)
.
T
he
highest
outp
ut
of
biorth
ogonal
Gero
nim
o
,
H
ard
i
n
a
nd
m
asso
pu
st
m
ul
ti
wav
el
et
2
(BiG
HM2
)
was
am
on
g
th
e
var
io
us
kind
s
of
m
ulti
wavel
et
.
Thu
s
,
Bi
GH
M
2
accom
plishe
d
pr
eci
sio
n
i
n
t
he
operati
ng c
ha
racteri
sti
c cu
rve o
f
the
r
ecei
ve
r
ar
ound
0.9
6 [
47
]
.
T.
N
gu
ye
n
et
al
.
[48]
sug
ge
st
ed
an
aut
om
at
ic
wav
el
et
-
base
d
te
ch
nique
f
or
cl
assify
ing
m
edical
inf
or
m
at
ion
an
d
a
ty
pe
-
2
fu
zz
y
log
ic
.
They
c
arr
ie
d
out
exec
ution
f
r
om
the
UCI
data
base
f
or
m
achine
le
arn
i
ng
on
2
m
edical
dataset
s:
Cl
evel
and
hea
rt
disea
se
and
Wisc
onsin
breast
canc
er.
T
he
outc
om
e
de
m
on
strat
es
that,
com
par
ed
to
oth
e
r
m
achine
le
arn
in
g
m
eth
ods,
the
a
dvantage
of
inte
rv
al
ty
pe
-
2
fuzzy
log
ic
sche
m
e
is
bette
r
[48].
Z.
Ma
h
m
ud
et
al
.
[49]
su
gge
ste
d
a
m
et
ho
d
to
us
e
age,
m
arit
al
s
ta
tus
and
the
ra
py
am
on
g
Ma
la
ysi
an
wom
en
to
find
ou
t
abo
ut
cer
vica
l
cancer.
The
r
ecords
of
patie
nts
with
cer
vical
cancer
are
ga
there
d
from
the
m
edical
center
of
Keb
a
ngsaa
n
Ma
la
ysi
a
Un
iver
sit
y
(U
KM).
T
he
m
od
el
has
f
our
phases
,
wi
th
444
record
s
of
pat
ie
nts
i
m
pacted
by
cerv
ic
al
cancer
,
and
fi
ndin
g
out
the
age
an
d
m
ar
it
al
sta
tus
of
wom
en'
s
m
edical
therapy
.
They
disc
overe
d
that
the
46
-
ye
ar
-
ol
d
fe
m
al
es
are
m
or
e
li
kely
to
devel
op
ce
rv
ic
al
c
ancer
.
So
,
it
is
sug
ge
ste
d
that
Ma
la
ysi
an
w
om
en
unde
rgo
te
sti
ng
pr
i
or
t
o
th
e
age
of
45
a
nd
they
al
so
f
ound
that
Chinese
w
ome
n
unde
r
th
e
age
of
57
ha
ve
m
or
e
li
kel
ihood
of
bein
g
diag
nosed
with
rad
i
oth
e
r
apy
in
the origi
nal ph
ase o
f
ce
rv
ic
al
cancer
[49
]
.
M.
Seera
et
al
.
[50]
sug
geste
d
us
in
g
hy
br
i
d
s
m
art
cl
assificati
on
to
cl
assify
m
edical
info
rm
at
ion
to
pr
e
dict
cance
r.
The
m
od
el
ha
s
a
ra
ndom
fo
r
est
,
cl
assifi
ed
t
rees,
a
nd
a
regressio
n
tree
an
d
a
m
in
-
m
ax
neu
ra
l
netw
ork.
The
te
chn
iq
ue
of
r
andom
fo
rest
is
us
e
d
to
c
re
at
e
a
cl
assifi
cat
ion
a
nd
re
gressi
on
Tree
m
od
e
l
ensem
ble.
Fu
z
zy
m
in
-
m
ax
is
us
ed
for
te
ach
ing
purposes
.
The
tree
of
cl
a
ssific
at
ion
a
nd
regressio
n
is
us
e
d
to
extract t
he
rule
. T
he precisi
on
of this m
od
el
f
or cance
r forec
ast
w
as
98.
84% [50
]
.
W.
Kuo
et
al
.
[51]
sugg
e
ste
d
a
novel
te
ch
ni
qu
e
f
or
breast
tum
ou
r
predic
ti
on
in
cl
inica
l
ultraso
nic
i
m
ages
us
in
g
t
he
decisi
on
tre
e.
The
m
od
el
con
ce
ntra
te
d
on
picture
s
f
ro
m
the
U
nited
Stat
es.
T
he
decisi
on
tree
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.
10
, No
.
4
,
A
ugus
t
2020
:
4217
-
422
5
4222
us
es
t
he
C
5.0
al
gorithm
.
Ma
chine
le
arn
i
ng
with
decisi
on
tree
al
gorith
m
help
predict
breast
tum
our
il
lness
with
93.33%
r
e
sp
onsi
ven
e
ss
preci
sion an
d 9
6.6
7%
s
pecifici
ty
p
recisi
on
[
51]
.
Seon
-
Ha
k
[
52
]
co
nst
ru
ct
e
d
a
prototype
us
i
ng
r
ough
set
structu
res
for
hierar
c
hical
cl
assifi
cat
ion
.
The
m
od
el
is
base
d
on
the
f
ram
ewo
rk
of
hi
erarch
ic
al
gra
nu
la
ti
on
to
fin
d
out
the
la
ws
of
cl
assifi
cat
ion
a
nd
therefo
re
s
ugge
ste
d
a
disc
overy
of
la
ws.
T
he
te
c
hn
i
qu
e
is
validat
e
d
a
ga
inst
the
W
isc
on
si
n
breast
c
ance
r
(
W
BC
)
data
ga
there
d
dataset
.
The
m
od
el
sti
ll
gen
erates
e
xcell
ent
ef
fici
ency
w
hen
l
oaded
with
sim
ple
ru
le
s
and
bri
ef
co
nd
it
ion
al
s.
Th
us,
by
creati
ng
m
i
nim
a
l
cl
assifi
c
at
ion
r
ules,
the
m
od
el
was
e
ffec
ti
ve
in
dec
re
asi
ng
the
num
ber
of
dim
ension
s.
His
m
od
el
m
a
kes
it
si
m
pler
for
us
to
a
na
ly
ze
the
infor
m
at
ion
syst
e
m
[52].
The bel
ow Ta
bl
e 3
s
hows
the
com
par
at
ive st
ud
y
of
br
east
c
ancer
tech
niqu
es whic
h we
h
a
ve discus
sed
a
bove.
Table
3.
T
he
c
om
par
at
ive s
urvey o
f br
ea
st c
ancer
Au
th
o
rs
Year
Data set
Metho
d
o
lo
g
y
Accurac
y
Lun
d
in
M
1999
Turk
u
City
Hos
p
it
al and
Cen
tral
Ho
s
p
ital
o
f
T
u
rku
Univ
ersity
ANN
5
y
rs
-
0
.90
9
,
1
0
y
rs
-
0
.08
6
an
d
15
y
rs0
.88
3
G
W
.
Kuo
2001
Machin
e lea
rnin
g
r
ep
o
sito
ry
C5
.0 alg
o
rith
m
9
6
.67
%
Seo
n
-
Hak
Seo
2001
W
isco
n
sin
b
reast c
an
cer
(
W
BC
)
,
UC
I
d
iab
etes f
o
r
Pi
m
a
I
n
d
ian
s
Hierarch
ical Class
i
f
icatio
n
8
3
.49
%
Delen D
2005
Machin
e lea
rnin
g
r
ep
o
sito
ry
ANN and
log
istic r
eg
ressio
n
9
3
.6%
Ja
m
ar
an
i S.
M.
h
2005
Machin
e lea
rnin
g
r
ep
o
sito
ry
(ANN
),
deco
m
p
o
si
tio
n
of
m
u
lti
wav
elet
-
b
ase
d
p
ictu
re
su
b
-
b
an
d
ROC
-
0
.96
Bellaach
ia
Ab
d
elg
h
an
i
2006
Natio
n
al Can
cer In
stitu
te (
NCI
)
su
rveillan
ce,
ep
id
e
m
io
lo
g
y
and
end
resu
lts
Data
Minin
g
M
eth
o
d
s
87%
Uz
m
a
Ans
ari
2013
Un
iv
ersity
of
Calif
o
rnia Ir
v
in
e (
UCI)
Ro
tatio
n
f
o
rest alg
o
rith
m
(
RF
-
ANN)
9
8
.05
%
M.
S
eera
2014
W
isco
n
sin
breast c
an
cer
(
W
BC
)
,
UC
I
d
iab
etes f
o
r
Pi
m
a
I
n
d
ian
s
Ran
d
o
m
f
o
rest,
cla
ss
if
ied
trees a
n
d
r
eg
r
ess
io
n
tr
ee
(CART)
and
f
u
zzy n
eu
ral
n
etwo
rk
m
in
-
m
ax
9
8
.84
%
A.
Yass
i
2014
Un
iv
ersity
of
Calif
o
rnia Ir
v
in
e (
UCI)
h
ierar
ch
ical
-
clu
ste
ring
90%
T.
Ng
u
y
en
2015
Un
iv
ersity
of
Calif
o
rnia Ir
v
in
e (
UCI)
W
av
elets
an
d
inter
v
al ty
p
e
-
2
f
u
zzy
log
ic sy
ste
m
(
I
T2FLS)
9
7
.40
%
5.
CHALL
ENG
ES A
ND R
E
S
EAR
CH OPP
ORTUNITIE
S
Re
views
of
da
ta
m
ining
m
eth
ods,
cl
assifi
c
at
ion
m
et
ho
ds,
sm
art
m
et
ho
ds
an
d
c
hoic
e
of
feat
ur
es
f
or
dise
ase
pr
e
dict
ion
we
re
discu
ssed
he
re.
As
t
he
sel
ect
ion
of
char
act
e
risti
cs
enab
le
s
us
to
erad
ic
at
e
unnec
essary
data,
la
r
ge
-
dim
ensio
nal
data
m
us
t
be
com
pr
essed
with
ou
t
t
he
lo
ss
of
data
,
w
hich
im
pr
oves
t
he
ef
fecti
ven
e
ss
of
cl
assifi
cat
io
n.
But
the
dif
f
ic
ulty
of
subse
t
at
tribu
te
sel
ect
ion
is
high,
wh
ic
h
is
com
plica
te
d
as
it
req
ui
res
com
plex
interd
epende
nce
on
a
wide
range
of
facto
rs.
W
e
c
ou
l
d
inco
r
pora
te
gu
ideli
nes
a
nd
feat
ur
e
sel
e
ct
ion
for
bette
r
resu
l
ts
in
the
cl
assif
ie
rs
in
the futu
re.
A
dd
it
io
nally
,
ne
w
featu
re
sel
ect
ion
m
et
ho
d
s
uch
as
a
nt co
lo
ny
op
ti
m
iz
ation
,
e
tc
.
is
possible
to
te
st
to
im
pr
ov
e
qual
it
y,
an
d
yo
u
ca
n
at
te
m
pt
exp
erim
enting
with
al
gor
it
h
m
po
te
ntial
for
m
os
t
m
edical
dataset
s
that
incl
ud
e
disti
nct
fe
at
ur
es
su
c
h
as
no
isy
i
nfor
m
ation
,
s
par
sit
y,
m
issi
ng
value, et
c. t
o
e
nh
a
nce m
od
el
accuracy.
6.
CONCL
US
I
O
N
This
pa
per
'
s
pr
im
ary
fo
cus
is
to
discuss
var
io
us
pr
e
di
ct
ion
m
od
el
s
and
te
ch
ni
ques
us
e
d
f
or
pr
e
dicti
ng
hea
rt
disease
a
nd
br
east
ca
ncer.
The
te
ch
niqu
e
al
so
she
ds
li
gh
t
on
t
he
sig
nificance
i
n
m
edical
dataset
of
var
i
ou
s
cl
assifi
cat
ion
te
ch
niques
for
disease
pr
e
dicti
on
.
T
he
da
ta
set
that
we
hav
e
disc
us
se
d
in
s
o
m
any
cur
ren
t
m
et
ho
ds
is
li
nked
to
hear
t
a
nd
breast
ca
nc
er.
As
a
cl
assi
fier,
t
he
diff
e
r
ent
m
achine
le
arn
i
ng
m
et
ho
ds
a
re
use
d
to
c
onstr
uct
a
value
-
ef
fecti
ve
m
od
el
for
predict
in
g
disea
se.
It
is
the
refor
e
well
rec
ognize
d
by
the
e
xhau
sti
ve
stu
dy
that
the
ext
racti
on
of
t
he
necessa
ry
data
from
t
he
cl
inica
l
re
posit
or
y
help
s
us
t
o
prom
ote ex
cel
le
ntly
-
inf
or
m
ed
te
sti
ng
a
nd c
hoic
es.
REFERE
NCE
S
[1]
M.
Jiang,
Y.
Chen,
M.
Li
u
,
S.
T.
Rosenbloom,
S.
Mani,
J.
C.
De
nn
y
,
and
H.
Xu,
“
A
study
of
m
ac
hine
-
le
arn
ing
-
base
d
appr
oa
ches
to
ext
ra
c
t
c
li
n
ic
a
l
entities
and
the
ir
asser
ti
ons
from
discha
rge
summ
ari
es,
”
J
.
Am
Me
d
Inform
Assoc
,
vol
.
18
,
n
o.
5
,
pp
.
601
–
60
6,
2011
.
[2]
R.
Sn
y
der
m
an,
“
Personali
ze
d
hea
l
th
ca
re
:
fro
m
the
or
y
to
pra
ctice,”
B
iot
e
ch
nology
Journal
,
vol.
7,
no.
8
,
pp.
973
–
979
,
Au
g.
2012
.
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
A com
pr
ehe
ns
i
ve stu
dy on dis
ease r
isk
p
re
di
ct
ion
s i
n ma
c
hi
ne
le
arni
ng
(
G. S
ar
any
a
)
4223
[3]
M.
Jiang,
Y.
Chen,
M.
Li
u,
S.
T.
Rosen
bloom,
S.
Mani,
J.
C
.
Den
n
y
,
and
H.
X
u,
“
A
stud
y
o
f
m
ac
hine
-
l
ea
rn
in
g
-
base
d
appr
oa
c
hes
to
ext
ra
ct
c
li
nical
en
ti
t
ie
s
a
nd
the
ir
assert
io
ns
from
discha
rge
sum
m
ari
es,
”
J.
Am
Me
d
Infor
m A
ss
oc
,
vo
l. 18
,
no
.
5
,
pp
.
601
–
606,
2011
.
[4]
N.
Raz
avian
a
nd
D.
Sontag,
“
Te
m
pora
l
convol
uti
ona
l
neu
ral
net
works
for
dia
gnosis
from
la
b
te
sts,”
arXiv
:1511.
0793
8v4
,
2015
.
[5]
R.
Ranga
n
at
h,
J
.
S.
Hirsch,
D.
Bl
ei
,
and
N.
El
h
ad
ad,
“
Risk
pre
dict
ion
for
chr
oni
c
kidney
dise
ase
pr
ogre
ss
ion
using
het
ero
g
en
eous
elec
tron
ic
health
r
ec
ord
data
and
tim
e
serie
s
ana
l
y
s
is,”
Journal
of
t
he
Ame
rican
Me
dic
al
Informatics
Associat
ion:
JAMIA
,
vo
l. 22, no. 4, pp. 872
–
880,
Jul.
2015.
[6]
N.
Ta
ngri
,
L.
A.
Steve
ns,
J.
Griffith,
H.
Ti
g
hioua
rt
,
O.
Dju
rdje
v,
D.
Na
imark,
A.
L
evi
n,
a
nd
A.
S.
Le
v
e
y
,
“
A
pre
dic
ti
v
e
m
odel
for
prog
ression
of
chr
o
nic
kidn
e
y
dise
ase
to
k
idney
f
ai
lur
e,
”
J
AMA
,
vol.
305
,
no.
1
5,
pp.
1553
–
1559
,
2011.
[7]
E.
Choi
,
M.
T.
B
aha
dori
,
A.
Schu
et
z
,
W
.
F.
St
ewa
rt,
and
J.
Sun,
“
Doctor
AI
:
Predi
ct
ing
cl
in
ical
ev
ent
s
via
re
cu
rre
n
t
neur
al
net
works
,
”
Proce
ed
ingsoft
he1
st
Ma
chi
ne
LearningforHealt
hcare
Conf
ere
nc
e,
ser.
Proc
ee
di
ngs
of
Mac
hi
n
e
Learning
Re
sear
ch
,
pp
.
301
–
318
,
Aug 2016.
[8]
N.
Bara
k
at,
A.
P.
Brad
ley
,
and
M.
N.
H.
Bar
akat,
“
Intelli
g
ible
S
upport
Vec
tor
M
ac
hin
es
for
Di
ag
nosis
of
Diabete
s
Mell
it
us,
”
I
EEE
Tr
ansacti
ons on Inform
ati
on
Tec
hnology
in
B
iomedi
c
ine
,
vol
.
14
,
no.
4
,
pp
.
1114
–
1120,
Jul.
2010.
[9]
W
u
ji
ongli
n
M.
S.,
et
al
.
,
“
Pre
dic
ti
on
Model
in
g
Us
ing
EHR
Data
:
Challeng
e
s,
and
a
Com
pa
rison
of
Mac
hine
Le
arn
ing
Appro
ac
hes,
”
Jou
rnal
of
the
medi
cal
c
are
sec
ti
on
,
Ame
rican
publ
ic
h
e
alt
h
ass
oci
at
ion
,
vol.
48
,
no
.
6
,
pp.
S106
-
S113, 2010.
[10]
W
.
Yu,
T.
Li
u
,
R.
Valdez,
M
.
Gw
inn,
and
M.
J.
Khour
y
,
“
Applicati
on
of
sup
port
vector
m
achine
m
odel
ing
f
or
pre
diction
of
co
m
m
on
disea
ses:
the
ca
se
of
di
a
bet
es
and
pre
-
d
i
abe
t
es,
”
BMC
Me
dic
a
l
Inform
ati
cs
and
De
ci
si
on
Mak
ing
,
vol
.
10
,
no.
16,
pp.
1
-
7,
2010.
[11]
N.
Raz
avian,
S.
Ble
cke
r
,
A.
M.
Schm
idt
,
A.
Sm
it
h
-
McLa
llen
,
S.
Nigam,
and
D.
Sontag,
“
Populat
ion
-
L
evel
Predic
ti
on
of
T
y
pe
2
Diab
etes
from
Cla
ims
Data
and
Anal
y
sis
of
Risk
F
ac
tors,
”
Bi
g
Da
ta
,
vol.
3
,
no.
4,
pp.
277
–
287
,
De
c.
2015
.
[12]
A.
V.
Le
bede
v
,
E.
W
estman,
G.
J.
P.
Van
W
est
en,
M.
G.
Kram
ber
ger
,
A.
Lund
erv
old,
D.
Aarsl
and,
H.
Soininen,
I.
Kłosze
ws
ka,
P.
Mec
occ
i
,
M.
Tsola
ki
,
B.
Vellas,
S.
Love
stone
,
and
A.
Simm
o
ns,
“
Random
Fo
rest
ense
m
ble
s
for
det
e
ct
ion
and
pr
edi
c
ti
on
of
Al
zh
ei
m
er’
s
disea
se
with
a
good
b
etw
ee
n
-
cohor
t
rob
ustness,”
NeuroI
mage:
Cli
ni
cal
,
vol.
6
,
pp
.
115
–
1
25,
2014
.
[13]
N.
Ta
ngri
,
L.
A.
Steve
ns,
J.
Griffith,
H.
Ti
g
hioua
rt
,
O.
Dju
rdje
v,
D.
Na
imark,
A.
L
evi
n,
and
A.
S.
Le
v
e
y
,
“
A
pre
dic
ti
v
e
m
odel
for
prog
ression
of
chr
o
nic
kidn
e
y
dise
ase
to
k
idney
f
ai
lur
e,
”
J
AMA
,
vol.
305
,
no.
1
5,
pp.
1553
–
1559
,
2011.
[14]
A.
Perotte,
R.
Ranga
na
th,
J
.
S
.
Hirsch
,
D.
Bl
ei
,
and
N.
El
ha
dad,
“
Risk
pr
ed
ic
ti
on
for
chr
on
ic
k
idney
d
isea
s
e
progre
ss
ion
using
het
ero
g
ene
ous
el
e
ct
ron
ic
h
ea
l
t
h
rec
ord
d
ata
an
d
ti
m
e
seri
es
an
aly
s
is,”
Journal
of
the
Ame
rica
n
Me
dic
a
l
Informa
ti
cs
Associat
ion:
JA
M
IA
,
vol
.
22
,
no.
4,
pp.
872
–
8
80,
Jul.
2015.
[15]
J.
Qiu,
Q.
W
u,
G.
Ding,
Y.
Xu,
and
S.
Feng,
“A
survey
of
m
a
chi
ne
l
ea
rning
f
or
big
dat
a
proc
essing,
”
EURA
S
IP
J.
Adv.
S
ignal
Proce
ss
.
,
vol. 2016
,
De
c. 2016.
[16]
B.
Shicke
l
,
P.
Ti
ghe
,
A.
B
ihor
ac
,
and
P
.
Rash
idi
,
"D
e
ep
EHR
:
A
surve
y
of
r
ec
en
t
adv
ances
in
de
ep
l
ea
rning
te
chn
ique
s
for
e
le
c
troni
c
h
ea
l
th
rec
ord
(EHR)
ana
l
y
sis,"
IE
EE
J.
B
iomed.
He
alt
h
Informatic
s
,
vol.
22
,
no.
5
,
pp.
1589
-
1604
,
2018.
[17]
Ming
y
u
Pak
,
Mi
y
oung
Shin
,
“
Deve
lopi
ng
Dis
ea
se
Risk
Predi
ct
ion
Mode
l
B
a
sed
on
Envi
ron
m
ent
al
Fac
tors
,
”
The
18th
I
EEE
I
nte
rnational
Sy
mpos
ium
on
Consum
er
El
ectroni
cs
(
ISCE
2014)
,
2014.
[18]
Xiao
y
an
W
ang
,
Am
y
Chused
,
N
om
ie
El
had
ad,
C
aro
l
Friedman
,
a
nd
Maria
n
th
i
M
ark
at
ou
:
“
Autom
at
ed
Know
l
edge
Acquisit
ion
fro
m
Cli
nical
N
arr
a
ti
ve
Report
s,
”
A
MIA
2008
S
ymp
osium P
roce
ed
in
gs
,
pp.
783
-
787,
2008.
[19]
Nicol
a
e
Dragu,
Fouad
El
khour
y
,
Ta
kunar
i
Ral
ph
and
A.
Morelli
Nicol
as
di
T
ada,
“
Ontolog
y
-
Bas
ed
Te
x
t
Mining
for
Predicting
D
i
sea
se
Outbr
ea
k
s,”
Proc
ee
dings
of
th
e
Tw
enty
-
Thir
d
Inte
rnatio
nal
Fl
orida
Artific
ia
l
Int
el
l
ige
n
c
e
Re
search
So
ciet
y
Conf
ere
nc
e
(
FLAIR
S
2010)
,
20
10
.
[20]
Kum
ar
Sen,
Sha
m
sher
Baha
dur
Pate
l
and
D
.
P.
Shukla,
“
A
Dat
a
Mining
Techn
ique
for
Predi
ct
i
on
of
Coron
a
r
y
Hea
rt
D
ise
ase
Us
ing
Neuro
-
Fuzz
y
,
”
Int
ernati
o
nal
Journal
o
f
Engi
ne
ering
an
d
Computer
Scienc
e
,
vol
.
2
,
no.
9,
pp.
2663
-
2671
,
2013.
[21]
Saba
Bashir,
Us
m
an
Qam
ar,
Far
han
Hass
an
Kha
n,
“
BagMOO
V:
A
novel
ense
mbl
e
for
hea
rt
dis
ea
se
pre
di
ct
ion
bootstra
p
aggr
e
gat
ion
with
m
ul
ti
-
object
ive
optim
iz
ed
voti
ng
R
ec
e
ive
d
,
”
Austr
alasian
Phy
sica
l
&
Engi
nee
ring
Sci
en
ce
s in
Me
d
i
ci
ne
,
vo
l. 38, pp. 305
-
323,
2015.
[22]
la
v
Petrov,
Dip
a
nja
n
Das
and
Ry
an
McDona
ld,
“
A
Univer
sal
P
art
of
-
Speec
h
T
agse
t,”
Proceedi
ngs
of
the
Ei
ght
Inte
rnational
Co
nfe
renc
e
on
Language
R
esource
s and
E
val
uat
ion (
LREC
2012
),
2
012.
[23]
Ma
y
o
clinic,
[O
nli
ne]
.
Ava
il
ab
le:
ww
w.m
a
y
o
cl
in
ic
.
org
[Acc
essed
17/10/
2015]
[24]
Isabe
l, [Onli
n
e]
.
Avail
ab
le
:
ww
w.i
sabe
lhe
a
lt
h
ca
re
.
com [Access
ed 12/10/
2015]
[25]
Pati
ent,
[Onl
ine
]
.
Avai
la
b
le
:
ww
w.pa
tient
.
co
.
uk
[
Acc
essed
11/10
/
2015]
[26]
[Online
]
.
Avai
lable: www
.
s
y
m
ptomchec
ker
.
isabelhealt
hc
are.c
om
[Acc
essed
30
/10
/2015]
[27]
Md.
Ta
hm
id
Ra
hm
an
La
skar,
Md.
Ta
hm
id
Hos
sain,
Abu
Raihan
Mos
tofa
Kam
al
,
Nafi
u
l
Ra
shid
,
“
Autom
at
ed
Disea
se
Predi
ction
S
y
st
em
(AD
PS
):
A
Us
er
Input
-
base
d
Rel
ia
b
le
Ar
chi
t
ec
tur
e
for
Dis
ea
se
Pred
ic
t
ion
,
”
Inte
rnational
Jo
urnal
of
Comput
er
Applications,
vol.
133
,
no
.
15
,
pp.
24
-
29
,
Jan
.
2
016.
[28]
Xiao
y
an
W
ang,
Am
y
Chused,
N
om
ie
El
had
ad,
C
aro
l
Friedman,
a
nd
Maria
nthi
M
ark
at
ou
,
“
Automa
te
d
Know
le
dg
e
Acquisit
ion
f
ro
m
Cli
nical
N
arr
a
ti
ve
Report
s,
”
A
MIA
2008
S
ymp
osium P
roce
ed
in
gs
,
pp.
783
-
787,
2008.
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.
10
, No
.
4
,
A
ugus
t
2020
:
4217
-
422
5
4224
[29]
S
Manimekal
a
i,
R
Suguna,
S
Ar
ulsel
var
ani
,
“
An
Intelli
g
ent
Auto
m
at
ic
Mul
ti
-
Dis
ea
se
Pred
ic
t
ion
Te
chn
ique
using
Data
Min
ing
Algorit
hm
s
and
Bi
g
Data
,
”
Int
ernati
onal
Journal
of
Computer
Sci
ences
and
Engi
nee
ring
,
vol
.
6
,
no.
10
,
Oct
.
201
8
[30]
Cristi
nel
Abab
e
i,
Mila
d
Ghorb
ani
Moghadda
m
,
“
A
Surve
y
of
Predic
ti
on
a
nd
Cla
ss
ifi
catio
n
Te
chni
qu
es
i
n
Multi
cor
e
Proc
e
ss
or
Sy
st
ems
,
”
I
EE
E
Tr
ansacti
o
ns on
Parallel
a
nd
Distribu
te
d
S
yste
ms
,
vol
.
1
,
n
o.
1
,
Sep
.
2017
.
[31]
Ankita
Dewan
,
and
Meghn
a
Sharm
a,
“
Predic
ti
on
of
he
art
disea
se
using
a
h
y
br
id
t
ec
hn
iq
ue
in
d
ata
m
ining
cl
assifi
ca
t
ion,”
2015
2nd
Int
e
rnational
Conf
ere
nce
on
Co
mputing
for
S
ustainabl
e
Glo
bal
Dev
el
opme
nt
(
INDIACom
)
,
2015.
[32]
P.
K.
Anooj,
“
Cl
ini
c
al
Dec
ision
Support
sy
st
em:
Risk
le
vel
pre
d
iction
of
hea
rt
dise
ase
using
weighted
fuz
z
y
rul
es,
”
Journal
of
King
Saud
Unive
rs
ity
-
Computer
and
I
nformation
Sc
ience
s
,
vo
l. 24, pp. 27
-
40,
2012.
[33]
Tsipoura
s
M.G.
,
Exa
r
chos
T.
P
.
,
Fotia
dis
D.I
.
,
Kotsia
A.P.
,
V
ak
a
li
s
K.V.,
Naka
K.K.,
Mi
cha
l
is
L.
K.
,
“
Autom
at
e
d
dia
gnosis
of
cor
onar
y
arter
y
dis
ea
se
bas
ed
on
d
at
a
m
ini
ng
and
f
uzzy
m
odel
ing
,
”
IEE
E
T
.
Inf
.
Te
chno
.
B
.
,
vol.
12
,
no.
4
,
pp
.
447
–
4
58,
2008
.
[34]
Chai
tr
al
i
S.
Dan
gar
e
and
Sul
abh
a
S.
Apt
e,
“
Im
prove
d
Stud
y
of
H
ea
rt
Disea
se
Pre
dic
ti
on
S
y
st
em
u
sing
Data
Minin
g
Cla
ss
ifi
c
at
ion
Techni
ques,
”
In
te
r
nati
onal Journal
of
Comput
er
Ap
pli
cations,
vo
l. 4
7,
no
.
10
,
pp
.
44
-
48,
Jun.
2012.
[35]
Mai
Shoum
an,
Ti
m
Turne
r
and
Rob
Stocke
r,
“
Inte
gra
ti
ng
Dec
ision
Tre
e
and
K
-
Mea
ns
Cluste
rin
g
with
Diffe
re
n
t
Init
ial
Cent
ro
id
Sele
c
ti
on
Metho
ds
in
the
Diagn
osis
of
Hea
rt
Disea
se
Patients,
”
Proce
ed
ings
of
t
he
Inte
rnationa
l
Confe
renc
e
on
Data
Mini
ng
,
20
12.
[36]
V.
Chaura
sia
an
d
S.
Pal,
“
Ea
rl
y
pre
diction
of
he
art
disea
s
es
using
dat
a
m
ini
ng
t
e
chni
ques,
”
Cari
bb
ean
Journal
o
f
Sci
en
ce and
Tec
hnology
,
vol
.
1
,
pp.
208
-
217
,
20
13.
[37]
G.
E.
Sakr
,
I.
H
.
El
ha
jj
and
H
.
A.
Huije
r
,
“
Support
vector
m
ac
hin
es
to
def
in
e
and
det
e
ct
ag
itati
on
t
ran
siti
on
,
"
IEEE
Tr
ansacti
ons on Aff
e
ctive Compu
ti
ng,
vol. 1
,
no
.
2,
pp
.
98
-
108
,
D
ec
.
2010
.
[38]
La
th
a
Parthi
ban
and
R.
Subrama
nia
n
,
“
Inte
ll
ig
e
nt
Hea
rt
Disea
se
Predic
ti
on
S
y
s
te
m
using
CA
N
FIS
and
Gene
tic
Algorit
hm
,
”
Inter
nati
onal Journal
of
Bi
o
logi
ca
l
a
nd
Life Sc
i
ence
,
vol
.
15
,
pp
.
157
–
160,
2007
.
[39]
K.
Chandra
she
kar
and
N.
De
e
pika
,
“
As
socia
tion
rule
fo
r
cl
as
sific
a
ti
on
of
He
art
At
ta
ck
Patie
nts,”
In
te
rnation
al
Journal
of
Ad
va
nce
d
Engi
ne
erin
g
Scienc
e
and
Te
chnol
ogi
es
,
vol
.
11,
no
.
2
,
pp
.
25
3
-
257,
2011
.
[40]
M.
Shoum
an,
T.
Turne
r
and
R.
Stocke
r,
“
Inte
gr
at
ing
na
ïve
ba
y
e
s
and
K
-
Mea
ns
cl
usteri
ng
with
diffe
ren
t
ini
tia
l
ce
ntro
id
sel
ectio
n
m
et
hods i
n
the di
agnosis
of
he
a
rt
dise
ase
p
at
i
ents
,
”
ICAITA
,
pp
.
125
-
137,
2012
.
[41]
Uzm
a
Ans
ari
,
Dipesh
Sharm
a,
Jy
o
ti
Soni
and
Sunita
Soni,
“
Inte
lligent
and
E
ffe
ctive
Hea
r
t
Disea
se
Predicti
on
S
y
stem
using
W
ei
ghte
d
As
socia
ti
v
e
Cla
ss
ifi
er
s,”
Inte
rnat
ional
Journal
on
Computer
Sci
en
ce
and
Engi
ne
ering
(
IJCSE
),
vol
.
3
,
no.
6
,
pp
.
2385
-
2392,
Jun.
2011.
[42]
A.
Yass
i,
M.
Ya
ghoobi,
M.
Yass
i
,
“
Distingui
shin
g
and
Clust
eri
ng
bre
ast
ca
n
ce
r
accordi
ng
to
hierar
chi
c
al
stru
ct
ure
s
base
d
on
cha
ot
i
c
m
ult
ispe
ci
es
p
art
i
ca
l
sw
arm
opti
m
iz
ation,
”
Ir
anian
Confe
ren
c
e
on
Int
ellige
n
t
Syste
ms
,
pp
.
1
-
6
,
Feb
.
2014
.
[43]
Lundi
n
M.,
Lun
din
J.,
Burke
B.
H.,
Toi
kkan
en
S.
,
P
y
lkkä
nen
L
.
and
Joensuu
H.,
“
Artifi
c
ia
l
Neura
l
Network
s
Applie
d
to
Sur
viva
l
Pred
ic
t
ion
in
Brea
st
C
an
ce
r,”
Oncolog
y
Inte
rnational
Jo
urnal
for
Canc
er
Re
saerch
an
d
Tr
eat
ment
,
vo
l.
57,
no
.
4
,
pp
.
28
1
-
286,
1999
.
[44]
Dele
n
D.
,
W
al
k
er
G.,
Kada
m
A.,
“
Predicting
bre
ast
ca
nc
er
su
rviva
bilit
y
:
a
co
m
par
ision
of
th
ree
d
at
a
m
ini
ng
m
et
hods,”
Arti
f
I
nte
ll Me
d
,
vo
l. 3
4,
no
.
2
,
pp
.
113
-
27,
2005
.
[45]
Bel
la
a
chia
Abd
el
ghan
i
and
Erhan
Guven,
"P
red
i
ct
ing
Br
ea
st
C
an
ce
r
Surviv
abi
l
ity
using
Data
Mini
ng
Techni
ques,
"
N
int
h
Workshop
on
Min
ing
Sc
ientif
i
c
and
Engi
n
ee
ring
Datase
ts
in
con
junc
t
ion
wit
h
th
e
S
ix
th
SI
AM
Int
ernati
ona
l
Confe
renc
e
on
Data
Mini
ng
,
20
06
.
[46]
H.
Ko
y
uncu
a
nd
R.
Ce
y
la
n
,
“
Artifi
ci
a
l
ne
ura
l
net
work
b
ase
d
on
rota
t
io
n
fore
st
for
biomedic
al
p
at
t
er
n
cl
assifi
ca
t
i
on,
”
i
n
36th
Inte
rnati
onal
Confe
renc
e
on
Tele
com
-
municat
ions
and
Signal
Proce
ss
i
ng
(TSP)
,
IEE
E
,
pp.
581
–
585
,
20
13.
[47]
Jam
ara
ni
S.
M.
h.
,
Behn
am
H.
and
Rez
a
ira
d
G.
A.,
“
Multi
wave
l
et
Based
N
eur
al
Ne
twork
for
Brea
st
Cancer
Diagnosis,
”
Inter
nati
onal Enf
orm
ati
ka
C
on
fe
ren
ce
,
IEC'05
,
pp
.
1
9
-
21,
2005
.
[48]
T.
Ngu
y
en,
A.
Khos
rav
i,
D.
Creighton,
and
S.
Naha
vandi,
“
Medi
-
ca
l
data
class
ifi
c
at
ion
using
int
er
val
t
y
pe
-
2
fu
z
z
y
logi
c
s
y
st
em a
nd
wave
l
et
s,
”
Applied
So
ft
Comput
i
ng
,
vol
.
30
,
pp
.
8
12
–
822,
2015
.
[49]
Z.
Mahm
ud
and
S.
Sulong,
“
C
onfounding
eff
e
ct
s
of
age
,
m
arit
al
stat
us
and
tr
e
at
m
ent
on
ce
rv
i
ca
l
c
ancer
stages
among m
al
a
y
s
ian wom
en,
”
Con
f
ere
nce
:
Sta
ti
sti
c
s in
Sc
ie
n
ce
,
Bus
ine
ss
,
and
Engi
n
ee
ring (
ICSSBE
)
,
2012
.
[50]
M.
Seer
a
and
C.
P.
Li
m
,
“
A
h
y
brid
intellige
nt
s
y
stem
for
m
edi
ca
l
d
ata
class
ifi
cation,”
E
x
pert
Syste
ms
wi
th
Appl
ic
a
ti
ons
,
vo
l.
41
,
no
.
5
,
pp
.
2
239
–
2249,
2014
.
[51]
W
.
Kuo,
R.
Cha
ng,
D.
Chen
and
C.
C.
Le
e
,
“
Dat
a
Mining
with
Dec
ision
Tr
ee
s
f
or
Diagnosis
of
Brea
st
Tumor
i
n
Medic
a
l
Ultr
aso
nic
Im
age
s
,
”
Bre
ast Canc
er
Re
se
archs and
Tr
e
at
ment
,
vo
l. 66, no
.
51
-
57
,
2001
.
[52]
Chul
-
Heui
L
ee,
Seon
-
Hak
Seo,
Sang
-
Chul
Choi
,
“
Rule
Discov
e
r
y
using
hierarchical
c
la
sifi
catio
n
struct
ur
e
wit
h
rough
sets,
”
IFS
A
World
Congre
ss
and
20th
NA
F
IPS
Int
ernati
ona
l
Conf
ere
nce
,
20
01.
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
A com
pr
ehe
ns
i
ve stu
dy on dis
ease r
isk
p
re
di
ct
ion
s i
n ma
c
hi
ne
le
arni
ng
(
G. S
ar
any
a
)
4225
BIOGR
AP
HI
ES OF
A
UTH
ORS
Ms.
G.
Saran
ya
rec
ei
v
ed
B.
T
ec
h
degr
ee
in
I
nform
at
ion
Tec
hnolog
y
from
Sri
Venka
t
eswar
a
Coll
ege
of
Eng
i
nee
ring
,
Sriperu
m
budur,
Chenna
i,
India
in
2009.
M.T
ec
h
degr
ee
in
Inform
at
ion
Te
chno
log
y
fro
m
Vel
Te
ch
Multi
Tech
Dr.Ra
ng
ara
j
an
Dr.Sakunthala
Eng
ine
e
rin
g
Coll
ege,
Avad
i,
Chenna
i
,
Indi
a
i
n
2011.
She
is
cu
rre
ntly
pursuing
her
Ph.D
in
Co
m
pute
r
Scie
nc
e
and
Engi
n
ee
ring
at
Sath
y
ab
ama
Instit
ute
of
science
and
Techn
olog
y
,
Chennai,
India
and
works
as
A
ss
ist
ant
Profess
or
in
Inform
at
ion
Techn
olog
y
Dep
art
m
e
nt
at
SR
M
Insti
tut
e
of
Sc
ie
nc
e
and
Technol
og
y,
India
.
She
h
as
7
+
y
ea
rs
of
te
a
ch
ing
expe
r
ie
n
ce
a
nd
2
y
ea
rs
of
In
dustr
y
Exp
erien
ce
.
H
er
rese
arc
h
int
er
ests
include
Big
Dat
a
Ana
l
y
t
ic
s,
Ma
chi
n
e
l
earning
and
Dee
p
l
ea
rning
.
Dr.
A.
Pr
av
in
rec
e
ive
d
th
e
B.
E
degr
ee
in
Com
p
ute
r
Scie
n
ce
&
Engi
ne
eri
ng
fro
m
Bhara
th
Niketan
Engi
ne
eri
ng
Co
ll
eg
e,
Madu
rai
Kam
ara
j
Univ
er
sit
y
,
Madur
ai,
I
ndia
in
2003
,
M.E
d
egr
e
e
in
Com
pute
r
Scie
n
ce
&
Eng
ine
er
in
g
from
Sath
y
abam
a
Univer
sit
y
,
Chenna
i
,
Indi
a
i
n
2005
and
Ph.
D
degr
ee
in
Com
pute
r
Sci
ence
&
Engi
ne
eri
ng
at
Sath
y
aba
m
a
Uni
v
ersity
,
Ch
ennai
,
India
in
2014
.
He
works
cur
ren
tly
as
an
As
sociate
Profess
or
for
the
Depa
rt
m
ent
of
Com
pu
te
r
Scie
n
ce
and
Eng
ineeri
ng
at
Sath
y
ab
ama
Institute
of
Scie
n
ce
a
nd
Technol
og
y
,
Chenna
i
and
h
e
has
14
Yea
rs
of
te
a
chi
ng
expe
r
ience
.
He
h
as
par
ticipated
and
pr
ese
nte
d
m
an
y
Res
ea
rch
Paper
s
in
I
nte
rna
ti
ona
l
and
Nati
ona
l
Confer
enc
es
and
al
so
publi
shed
m
an
y
pape
rs
in
Int
er
nat
ion
al
and
Na
ti
onal
Journals.
His a
rea of
inter
ests
includes
Sof
twar
e Engi
n
ee
ri
ng,
Dat
a
m
ini
ng
,
Int
ern
e
t
of
Things
and
Big
da
ta
.
Evaluation Warning : The document was created with Spire.PDF for Python.