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.
15
,
No.
1
,
Febr
uary
20
25
, pp.
894
~
907
IS
S
N:
20
88
-
8708
, DO
I: 10
.11
591/ij
ece.v
15
i
1
.
pp
894
-
907
894
Journ
al h
om
e
page
:
http:
//
ij
ece.i
aesc
or
e.c
om
Compari
son of
m
achine le
arning a
lgorit
hm
s to ide
ntify a
nd
prevent
low back i
njury
Ch
ri
sti
an
Ova
ll
e
Pau
li
no
1
,
Jorg
e
Hu
am
an
i
Correa
2
1
Dep
artm
en
t of
E
n
g
in
eering
,
Facu
lty
of
Eng
in
eering
,
U
n
iv
ersid
ad
T
ecno
ló
g
ica
d
el Per
ú
,
Li
m
a,
Per
ú
2
Academic
C
en
tre
for M
aterials
and
Nano
tech
n
o
lo
g
y
,
AGH Univ
ersity
o
f
Kr
ak
o
w,
K
rakó
w,
Po
lan
d
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
A
pr
4,
2024
Re
vised
A
ug 22, 2
024
Accepte
d
Se
p 3, 2
024
Wi
th
the
adv
ancem
en
t
of
techno
logy,
r
em
ot
e
wo
rk
and
v
irt
ua
l
class
es
have
bec
om
e
inc
r
ea
si
ngly
com
mon
,
le
ad
ing
to
prol
onged
per
iods
i
n
front
of
com
put
ers
and,
conse
quently,
to
disco
mfort
and
eve
n
lower
b
ac
k
pa
in.
Thi
s
study
com
p
are
s
machine
l
ea
rn
i
ng
al
gor
it
hms
t
o
ide
n
ti
fy
and
pre
ve
nt
low
bac
k
p
ai
n
,
a
common
he
al
th
prob
le
m
.
A
pre
di
ct
iv
e
mod
el
for
ea
r
l
y
dia
gnosis
and
pre
v
ent
ion
of
the
se
inj
uri
es
was
deve
lop
ed
using
dataset
s
from
open
dat
a
r
eposit
ori
es.
Six
machine
learni
ng
mod
el
s
were
used
to
tr
a
in
the
d
at
a
.
Result
s
show
ed
tha
t
logi
st
i
c
r
egr
ession
was
the
most
eff
ec
t
ive
mode
l
,
with
per
forma
n
ce
cur
ves
of
70%
,
90
%,
and
99%.
Pe
rform
ance
me
tr
i
cs
indicated
86%
a
cc
ur
ac
y
,
85%
re
call,
and
86%
F1
-
score
.
Acc
ura
cy
of
70
%,
recal
l
of
71%,
and
F1
-
s
c
ore
of
63%
r
efle
ct
the
robust
abilit
y
of
th
e
model
to
addr
ess
the
p
roblem.
In
addition
,
an
i
ntui
ti
v
e
interfa
c
e
was
im
p
le
m
e
nte
d
using
Gradi
o
Softwar
e
to improve data
visual
izat
ion.
Ke
yw
or
d
s
:
Algorith
m c
omparis
on
Com
pu
ta
ti
onal
med
ic
in
e
Lumbar
inj
ur
ie
s
M
ac
hin
e lea
rn
i
ng
Pr
e
dicti
ve
m
odel
s
This
is an
open
acc
ess arti
cl
e
un
der
the
CC
BY
-
SA
l
ic
ense
.
Corres
pond
in
g
Aut
h
or
:
Christi
an O
valle
Pau
li
no
Dep
a
rtme
nt of
En
gin
eeri
ng, Fac
ulty
of Engi
neer
i
ng, Uni
ve
rsida
d
Tec
noló
gica d
el
Per
ú
Lima
, P
e
rú
Emai
l:
d
oval
le
@u
t
p.
e
du.p
e
1.
INTROD
U
CTION
Lo
w
back
disorde
rs
c
onside
ra
bly
strai
n
pu
blic
an
d
occ
upat
ion
al
healt
h,
re
pr
ese
ntin
g
a
round
40%
of
al
l
mu
sc
uloske
le
ta
l
disorde
rs
relat
ed
t
o
w
ork
gl
ob
al
l
y.
It
is
of
c
on
ce
rn
t
hat
75%
of
these
injur
ie
s
or
igi
na
te
in
every
day
act
iv
it
ie
s,
su
ch
as
l
ifti
ng
,
le
a
ding
to
one
in
t
hr
e
e
wor
ker
s
w
orl
dw
ide
facin
g
low
back
pro
bl
ems,
making
the
m
on
e
of
the
le
adin
g
c
auses
of
work
a
bs
e
nteei
sm
[
1]
–
[
3]
.
More
over,
it
is
project
e
d
t
hat
appr
ox
imat
el
y
200
bill
ion
do
l
la
rs
a
re
s
pe
nt
each
yea
r
on
t
r
eat
ing
l
ower
ba
ck
pain
[4]
.
Con
se
quently
,
du
e
to
the
hi
gh
c
os
t
and
ti
me
re
quired
for
dia
gn
os
is,
th
e
nee
d
for
s
pecial
iz
ed
kn
ow
le
dge
in
this
a
rea
be
comes
evide
nt
[
5]
.
I
n
t
h
i
s
c
o
n
t
e
x
t
,
t
h
e
i
n
c
r
e
a
s
i
n
g
i
n
c
i
d
e
n
c
e
o
f
m
a
l
i
g
n
a
n
t
s
p
i
n
a
l
a
b
n
o
r
m
a
l
i
t
i
e
s
h
i
g
h
l
i
g
h
t
s
t
h
e
u
r
g
e
n
t
n
e
e
d
f
o
r
e
a
r
l
y
d
e
t
e
c
t
i
o
n
t
o
p
r
e
s
e
r
ve
t
h
e
q
u
a
l
i
t
y
o
f
l
i
f
e
[6]
.
B
e
y
o
n
d
t
h
e
s
p
i
n
a
l
d
e
g
e
n
e
r
a
t
i
o
n
a
s
s
o
c
i
a
t
e
d
w
i
t
h
a
g
i
n
g
,
w
h
i
c
h
c
a
n
c
a
u
s
e
a
c
u
t
e
o
r
c
h
r
o
n
i
c
l
o
w
b
a
c
k
pa
i
n
a
n
d
f
u
n
c
t
i
on
a
l
d
i
s
a
b
i
l
i
t
y
a
t
a
l
l
a
g
e
s
,
t
h
e
r
e
a
r
e
s
e
v
e
r
a
l
a
d
di
t
i
o
n
a
l
c
o
n
d
i
t
i
o
n
s
,
s
u
c
h
a
s
s
c
o
l
i
o
s
i
s
a
n
d
i
n
j
u
r
i
e
s
r
e
s
ul
t
i
n
g
f
r
o
m
i
m
p
r
o
p
e
r
p
o
s
t
u
r
e
,
t
h
a
t
e
x
a
c
e
r
b
a
t
e
t
h
e
p
r
o
b
l
e
m
.
S
t
u
d
i
e
s
s
h
o
w
t
h
a
t
m
a
i
n
t
a
i
n
i
n
g
p
o
o
r
p
o
s
t
u
r
e
w
h
i
l
e
s
i
t
t
i
n
g
f
o
r
l
o
n
g
p
e
r
i
o
d
s
c
a
n
c
a
u
s
e
v
a
r
i
o
u
s
h
e
a
l
t
h
p
r
o
b
l
e
m
s
,
i
n
c
l
u
d
i
n
g
u
p
p
e
r
a
n
d
l
o
w
e
r
b
a
c
k
a
n
d
n
e
c
k
d
i
s
c
o
m
f
o
r
t
.
T
h
i
s
r
e
s
u
l
t
s
f
r
o
m
u
n
e
v
e
n
p
r
e
s
s
u
r
e
d
i
s
t
r
i
b
u
t
i
o
n
o
n
t
h
e
s
p
i
n
e
,
p
o
t
e
n
t
i
a
l
l
y
l
e
a
di
n
g
t
o
b
o
n
e
i
nj
u
r
i
e
s
,
s
a
r
c
o
p
e
n
i
a
,
a
n
d
i
m
p
a
i
r
e
d
c
i
r
c
u
l
a
t
i
o
n
.
T
h
e
r
e
f
o
r
e
,
i
t
i
s
e
s
s
e
nt
i
a
l
t
o
m
a
i
n
t
a
i
n
p
r
o
p
e
r
s
i
t
t
i
n
g
p
o
s
t
u
r
e
,
e
s
p
e
c
i
a
l
l
y
f
o
r
t
h
o
s
e
e
n
g
a
g
e
d
i
n
l
o
n
g
w
o
r
k
o
r
s
t
u
d
y
s
e
s
s
i
o
n
s
[7]
–
[
1
3
]
.
To
ta
c
kle
the
s
e
pro
blem
s
,
re
cent
a
dv
a
ncem
ents
in
arti
fici
al
intel
li
gen
ce
(AI)
ha
ve
ope
ned
up
ne
w
po
s
sibil
it
ie
s
f
or
dia
gnos
i
ng
an
d
mana
ging
lo
we
r
back
inju
ries.
AI
has
dem
onstra
te
d
it
s
use
f
ul
ness
i
n
deliveri
ng
acc
ur
at
e
an
d
unde
rstan
dab
le
in
formati
on
to
healt
hcar
e
pro
vid
e
rs,
e
nh
a
n
c
ing
it
s
de
pend
abili
ty
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
omp E
ng
IS
S
N:
20
88
-
8708
Compari
son
of
mach
i
ne
le
arn
ing
algorit
hms
to ide
ntif
y and
pr
eve
nt
…
(
C
hri
sti
an
Ova
ll
e
Pa
ulino
)
895
acro
s
s
dif
fer
e
nt
co
ntexts
[13]
–
[
16]
.
S
pecifica
ll
y,
mac
hin
e
le
ar
ning
(
M
L
)
has
e
me
rg
e
d
as
a
pro
misi
ng
appr
oach to tac
kling t
he dif
fic
ulti
es asso
ci
at
e
d wit
h diag
nos
ing
l
umbar
con
diti
on
s
[
17], [
18]
.
Additi
on
al
l
y,
machine
le
arni
ng
ho
l
ds
the
po
te
ntial
to
trans
form
m
e
dical
pr
act
ic
e
by
offe
rin
g
phys
ic
ia
ns
pre
ci
se
and
ta
il
ored
in
formati
on
,
w
hich
c
ould
help
minimi
z
e
medical
er
r
or
s
an
d
s
urpa
ss
the
eff
ect
ive
ness
of
co
nventi
onal
meth
od
s
[19]
.
M
ac
hin
e
le
arn
i
ng
te
ch
ni
qu
e
s
a
re
tra
nsfo
rming
the
he
al
thcare
sect
or
globall
y,
eq
uippin
g
p
r
of
essi
onal
s
wi
th
inno
vative
too
ls
that
im
pro
ve
the
qu
al
i
ty
an
d
ef
fici
ency
of
medical
care
[
20]
.
Th
is
h
i
gh
l
igh
ts t
heir posi
ti
ve
impact
on
patie
nt care
[
21]
.
D
e
s
p
i
t
e
t
h
e
s
i
gn
i
f
i
c
a
nt
a
d
v
a
n
c
e
s
i
n
A
I
a
n
d
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
,
t
h
e
r
e
a
r
e
s
t
i
l
l
s
u
b
s
t
a
n
t
i
a
l
c
h
a
l
l
e
n
g
e
s
t
o
ov
e
r
c
o
m
e
.
F
o
r
i
n
s
t
a
n
c
e
,
t
h
e
o
c
c
u
r
r
e
n
c
e
o
f
f
a
l
s
e
p
o
s
i
t
i
v
e
s
i
n
m
a
g
n
e
t
i
c
r
e
s
on
a
n
c
e
i
m
a
gi
n
g
(
M
R
I
)
a
n
a
l
y
s
i
s
w
h
e
n
u
s
i
n
g
B
a
y
e
s
'
t
h
e
o
r
e
m
i
s
a
n
o
t
a
b
l
e
i
s
s
u
e
.
T
h
i
s
u
n
d
e
r
s
c
o
r
e
s
t
h
e
o
n
g
o
i
n
g
n
e
e
d
f
o
r
t
h
e
d
e
v
e
l
o
p
m
e
n
t
o
f
m
e
t
h
o
d
s
t
ha
t
c
a
n
e
n
h
a
n
c
e
t
he
a
c
c
u
r
a
c
y
a
n
d
r
e
l
i
a
b
i
l
i
t
y
o
f
l
u
m
b
a
r
d
i
a
g
n
o
s
t
i
c
s
,
t
h
e
r
e
b
y
i
m
pr
o
v
i
n
g
h
e
a
l
t
h
c
a
r
e
[22]
–
[
2
5
]
.
It
is
c
r
ucial
to
delve
int
o
the
pioneeri
ng
resea
rc
h
on
the
a
pp
li
cat
io
n
of
AI
a
nd
M
L
in
th
e
mana
geme
nt
of
l
umbar
inju
ri
es.
F
or
i
ns
t
anc
e,
a
gr
oundbre
akin
g
stu
dy
[
26]
de
vise
d
a
predict
ive
model
us
i
ng
deep
le
ar
ning
and
M
L
te
c
hn
i
qu
e
s
to
f
or
eca
st
recove
ry
ou
tc
om
es
f
ollow
i
ng
lum
ba
r
dis
c
he
rn
ia
ti
on,
there
by
ai
din
g
cl
inica
l
decisi
on
-
ma
ki
ng.
This
resear
ch
retr
ospect
iv
el
y
exami
ned
cl
inica
l
data
from
470
patie
nt
s
and
app
li
ed
a
ra
ng
e
of
al
gorithm
s,
su
c
h
as
ra
ndom
f
orest
(R
F)
,
e
xtre
me
gr
adient
boos
ti
ng
(
X
GBoost),
su
pp
or
t
vecto
r
machi
ne
(
SVM
),
dec
isi
on
tree
(
D
T),
K
-
near
est
neig
hbor
(KN
N)
,
l
og
ist
ic
re
gr
essi
on
(LR)
,
li
gh
t
gr
a
dient
boost
ing
mac
hin
e
(L
GBM),
an
d
mul
ti
la
yer
pe
rcep
t
ron
(
M
LP
).
Th
e
res
ults
re
veal
ed
a
lo
w
c
orrelat
io
n
betwee
n
t
he
fe
at
ur
es,
as
de
pi
ct
ed
in
the
c
orrelat
ion
matri
x
heat
ma
p.
A
nothe
r
st
udy
[
27]
cra
fted
a
m
achine
-
le
arn
in
g
al
gori
thm
to
eval
uat
e
the
c
onnecti
on
bet
ween
lu
m
bar
disc
heig
ht
on
r
adi
ograph
s
a
nd
t
he
prese
nce
of
disc
bulges
o
r
herniat
ion
s
.
By
anal
yzin
g
data
from
458
patie
nts,
the
y
identifie
d
c
ruci
al
factor
s
li
nked
t
o
lumb
a
r
disc
he
rn
ia
ti
on
(
LDB
H)
,
inclu
di
ng
L4
-
5
-
disc
heig
ht,
a
ge,
a
nd
L
1
-
2
-
disc
heig
ht
.
A
DT
-
ba
sed
model
was
de
velo
ped
for
cl
inica
l
de
ci
sion
-
maki
ng,
achievi
ng
F1
-
sc
ore
of
0.7
06,
0.7
78,
0.569,
0.729,
an
d
0.7
06
f
or
the
le
ast
absol
ute
sh
ri
nk
a
ge
and
sel
ect
io
n
op
e
rato
r
(
L
A
SSO
)
,
m
ulti
va
riat
e
adap
ti
ve
regressio
n
s
pl
ines
(
MARS
)
,
DT
,
RF,
a
nd
X
GBoost
m
odel
s,
res
pecti
vely
,
with
the
MARS
m
od
el
at
ta
ini
ng
the
highest
F1
-
scor
e
.
In
a
se
par
at
e
study
[28]
,
th
e
eff
ect
ive
nes
s
of
a
tra
nsfo
raminal
e
pidur
al
ste
ro
id
inje
ct
ion
(TFE
SI)
wa
s
evaluate
d
in
pa
ti
ents
with
lu
mbosacral
ra
di
cular
pai
n
du
e
to
lu
mb
a
r
sp
i
na
l
ste
nosis
(LSS),
a
le
ss
c
ommo
nly
stud
ie
d
co
ndit
ion.
A
c
onvolut
ion
al
ne
ural
ne
twork
(CN
N)
was
trai
ne
d
with
data
f
rom
193
patie
nts,
ac
hievin
g
an
a
rea
un
der
the
c
urve
(
AUC
)
of
0.9
20
a
nd
a
n
accu
rac
y
of
87.
2%
,
dem
on
st
rati
ng
the
model’s
outst
a
nd
i
ng
pr
e
dicti
ve
ca
pa
bili
ty.
Simi
la
rl
y,
an
oth
e
r
stu
dy
by
Haide
r
et
al.
[29
]
i
ntrod
uc
ed
a
n
ad
va
nc
ed
machi
ne
le
a
rn
i
ng
te
chn
iq
ue
us
in
g
bo
otstrap
ping
an
d
data
ba
la
ncing
meth
ods
to
ide
ntif
y
low
back
pain
.
The
y
em
ployed
a
sta
nd
a
rd
dataset
co
ntainin
g
310
rec
ords
a
nd
pr
opos
e
d
t
he
rand
om
f
orest
gr
a
di
ent
boos
ti
ng
X
GBo
os
t
ensem
ble
(RG
XE)
meth
od,
wh
ic
h
c
ombin
ed
RF
,
gr
a
dien
t
boos
ti
ng
(G
B
),
a
nd
XG
B
oos
t,
surpa
ssin
g
previ
ou
s
methods
with
a
rema
r
kab
le
accurac
y
of
0.99.
A
st
udy
[
30]
al
s
o
c
reate
d
a
me
dical
te
st
to
help
heal
thcare
prof
e
ssio
nals
choose
a
nd
as
sign
phys
ic
al
treat
ments
f
or
no
nspeci
fic
l
ow
bac
k
pain
pa
ti
ents.
This
stu
d
y
assessed
se
veral
M
L
al
gorith
ms,
i
nclu
ding
LR,
D
T,
SVM,
K
NN,
a
nd
GB
.
T
he
fi
nd
i
ngs
rev
eal
e
d
that
a
ll
ML
models
ach
ie
ve
d
accu
racies
e
xceed
i
ng
80%,
with
S
VM
bei
ng
t
he
m
os
t
pr
eci
se,
reachi
ng
an
accu
rac
y
of
over
90%.
Co
ns
ide
r
ing
these
stu
dies
and
t
heir
li
mit
at
ion
s,
the
current
stu
dy
was
de
sig
ned
t
o
ad
dr
e
ss
thes
e
issues
and en
ha
nce t
he
d
ia
gn
os
is a
nd treat
me
nt
of lum
bar inj
uri
es.
The
main
obje
ct
ive
of
this
art
ic
le
is
to
ex
plore
a
nd
anal
yze
the
eme
rg
i
ng
impact
of
mac
hi
ne
le
ar
nin
g
in
the
ide
ntifi
cat
ion
a
nd
pr
even
ti
on
of
lu
mb
a
r
le
sio
ns.
We
will
hi
gh
l
igh
t
how
the
pro
po
se
d
ad
va
nced
method
ologies,
if
im
plement
ed,
ca
n
sig
nif
ic
antly
im
pro
ve
diag
nosti
c
accu
racy
an
d
opti
mize
trea
tment
protoc
ols.
This
resea
rch
no
t
on
l
y
c
ontrib
ut
es
to
the
me
dical
fiel
d
but
al
so
pro
vid
es
a
cl
ear
f
rame
w
ork
f
or
fu
t
ur
e
re
searc
h
and
cl
inica
l
de
velo
pm
e
nts.
I
n
ad
diti
on
t
o
e
mphasiz
in
g
the
pr
e
valence
a
nd
impact
o
f
lo
w
bac
k
injur
ie
s,
this
s
tudy
will
iden
ti
fy
s
pecific
a
reas
that
requi
re
imp
r
ov
e
me
nt,
s
uch
as
re
du
ci
ng
false
posit
ive
diag
noses
on
M
RI
scan
s
a
nd
opti
mizi
ng
str
at
egies
f
or
the
mana
geme
nt
of
c
hron
ic
lo
w
back
pain.
Gi
ve
n
t
hat
M
L
has
prove
n
to
be
a
highly
ef
f
ect
ive
to
ol
in
data
anal
ysi
s,
it
is
cru
ci
a
l
to
thoro
ughl
y
evaluate
the
va
rio
us
avail
able
al
gor
it
hm
s
to
sel
ect
the
m
os
t
a
pp
ropr
ia
te
on
e
f
or
l
ow
ba
ck
inju
ry
dia
gnos
i
s
an
d
treat
m
e
nt.
T
he
innov
at
io
n
bro
ught
by
M
L
c
an
acce
le
rate
the
ti
me
to
dia
gnos
is
a
nd
th
us
t
he
init
ia
ti
on
of
treat
me
nt,
wh
ic
h
reduces t
he wa
it
ing
ti
me fo
r p
at
ie
nts and
pr
e
ven
ts
their
heal
th from
deterio
rati
ng
[
6]
,
[
15]
,
[
23]
.
The
a
rtic
le
is
orga
nized
a
s
f
ollows:
sect
io
n
2
outl
ines
t
he
meth
odolog
y,
desc
ribi
ng
the
a
ppr
oach
us
e
d
.
Sect
io
n
3
prese
nts
the
resu
lt
s
obta
ine
d
.
Sect
io
n
4
pr
ov
i
des
a
disc
ussi
on
that
a
naly
zes
and
i
nter
pret
s
the
fin
dings
.
Finall
y,
s
ect
ion 5
conclu
des wit
h
t
he
stu
dy’s
c
oncl
us
io
ns
.
2.
METHO
D
The
project
pr
esented
e
ntail
s
ap
plied
re
sea
rch
at
a
pr
e
di
ct
ive
le
vel,
ai
ming
to
s
olv
e
a
pro
blem
thr
ough
c
ompr
ehensi
ve,
orga
nized,
an
d
s
ys
t
emat
ic
app
li
ca
ti
on
of
a
cq
uire
d
knowle
dge
t
o
fi
nd
a
s
olu
ti
on.
I
n
this
rese
arc
h, we
ha
ve
de
velop
e
d
a
c
ompre
hensi
ve
so
l
ut
io
n
for
dia
gnos
in
g
a
nd
treat
i
ng
lum
bar
le
sio
ns
,
o
ptin
g
for
a
pre
-
e
xpe
rimental
de
sig
n.
T
his
ch
oice
is
justi
fied
by
it
s
capaci
ty
to
identif
y
a
nd
a
ddres
s
po
te
ntial
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
omp E
ng,
V
ol.
15
, No
.
1
,
Febr
uary
20
25
:
894
-
907
896
te
chn
ic
al
a
nd
op
e
rati
onal
iss
ues
be
fore
imp
act
ing
act
ual
pa
ti
ents
[
9]
.
A
ddit
ion
al
ly
,
t
he
qu
a
ntit
at
ive
a
ppr
oac
h
facil
it
at
es
the
i
te
rati
ve
de
velo
pm
e
nt
of
t
he
s
olu
ti
on,
mi
nim
iz
ing
ris
ks
an
d
ens
ur
in
g
it
s
sa
f
et
y
a
nd
reli
abili
ty
as
it
undergo
e
s
re
fineme
nt.
T
hus
,
this
de
sig
n
al
lows
f
or
a
djust
ments
a
nd
imp
roveme
nts
bas
ed
on
te
st
resu
l
ts
and
feedbac
k
[6]
.
T
he
p
r
oj
e
c
t
i
s
m
e
t
i
c
ul
ou
s
l
y
s
t
r
uc
t
ur
e
d
i
nt
o
f
ou
r
e
s
s
e
nt
i
a
l
ph
a
s
e
s
f
ol
l
ow
i
n
g
t
he
c
r
os
s
-
i
n
d
us
t
r
y
s
t
a
n
da
r
d
pr
oc
e
s
s
fo
r
da
t
a
m
i
ni
ng
(
C
R
I
S
P
-
DM
)
m
e
t
h
o
do
l
o
gy
,
t
he
m
o
s
t
w
i
de
l
y
us
e
d r
e
f
e
r
e
nc
e
m
od
e
l
f
or
de
ve
l
op
i
n
g
da
t
a
m
i
ni
ng
pr
oj
e
c
t
s
[
3
1]
.
T
hi
s
m
e
t
ho
d
ol
og
y
i
s
r
e
no
w
ne
d
f
o
r
i
t
s
s
t
r
uc
t
ur
e
d
a
n
d
s
ys
t
e
m
a
t
i
c
da
t
a
a
na
l
ys
i
s
a
nd
kn
ow
l
e
dg
e
e
xt
r
a
c
t
i
on
a
pp
r
oa
c
h,
m
a
ki
n
g
i
t
a
pp
l
i
c
a
bl
e
t
o
v
a
r
i
ou
s
pr
oj
e
c
t
s
[
3
2]
.
T
he
pr
o
c
e
s
s
c
om
m
e
nc
e
s
w
i
t
h
un
de
r
s
t
a
nd
i
n
g
t
he
bu
s
i
ne
s
s
a
nd
da
t
a
,
w
hi
c
h
i
s
c
r
uc
i
a
l
f
or
t
he
c
om
pa
n
y
a
s
it
e
nh
a
nc
e
s
t
he
l
i
ke
li
ho
o
d
of
s
uc
c
e
s
s
f
or
t
he
i
r
da
t
a
m
i
ni
ng
e
nd
e
a
v
or
s
[
3
3]
.
G
i
ve
n
t
he
a
bo
ve
,
t
he
f
ou
r
ph
a
s
e
s
o
f
t
he
c
ur
r
e
n
t
r
e
s
e
a
r
c
h
on
a
p
r
e
di
c
t
i
ve
m
od
e
l
ba
s
e
d
o
n
ML
a
l
go
r
i
t
h
m
s
t
o
i
de
nt
i
f
y
a
nd
pr
e
ve
nt
l
um
ba
r
i
nj
ur
i
e
s
a
r
e
de
t
a
i
l
e
d
be
l
ow
.
2.1. D
ata
pre
-
processin
g
Dataset
pr
e
-
processin
g
is
not
just
an
init
ia
l
ste
p
but
a
cr
uc
ia
l
on
e
in
our
r
esearch
.
At
thi
s
sta
ge,
our
main
ob
j
ect
ive
is
to
ad
dr
e
ss
missi
ng
an
d
null
values
that
cou
l
d
disrupt
our
predict
io
ns
.
The
pr
ima
ry
f
ocus
is
cl
eaning
t
he
da
ta
an
d
co
nver
ti
ng
al
l
feat
ur
e
data
int
o
nu
merical
values
,
e
nab
li
ng
the
al
gorithm
to
operate
eff
ect
ivel
y.
Y
our
r
ole
in
this
process
is
sig
ni
ficant.
T
o
ac
hi
eve
this,
we
c
onduct
ed
rig
or
ou
s
pr
e
-
proce
s
sing
base
d
on
datas
et
s
obta
ine
d
f
r
om
Kaggle,
ex
tract
ing
the
m
os
t
releva
nt
va
lues
a
nd
ad
dr
e
ssing
iss
ues
rel
at
ed
t
o
the
lu
mb
a
r
re
gi
on
.
On
e
of
t
he
dataset
s,
na
med
“
col
umn
_3C_wek
a.
csv
,
”
re
ferred
to
as
“
C
olumn
”
he
nc
eforth,
include
s
dia
gnos
ti
cs
relat
ed
t
o
the
s
pin
e
a
nd
is
based
on
six
val
ues
re
pres
enting
a
ngle
s
of
the
s
pin
e's
es
sentia
l
par
ts.
T
he
othe
r
dataset
,
na
med
“
D
ata
set
_s
pi
ne.
csv
,
”
he
reafter
re
ferre
d
to
as
“
Spine
”
f
or
brevit
y,
con
ta
in
s
data relat
ed
to
abno
rmal
or
no
rmal s
pin
es
and
12 v
a
riables
represe
nting di
ff
e
ren
t
par
ts
of
the s
pin
e
.
It
is
esse
ntial
to
note
that
t
he
qual
it
y
of
bo
t
h
dataset
s
sign
ific
a
ntly
infl
uen
ces
t
he
M
L
m
odel
's
eff
ect
ive
ness.
Access
to
t
hes
e
on
li
ne
data
s
ources
was
fac
il
it
at
ed
thro
ug
h
an
i
nternet
connecti
on,
al
lo
wing
us
to
ob
ta
in
rele
va
nt
me
dical
in
f
ormat
ion.
Our
com
pr
e
he
ns
ive
pre
-
processi
ng
proce
dure
in
vo
l
ved
meti
cul
ou
sl
y
app
l
ying
Dum
my
c
odin
g
a
nd labeli
ng for ca
te
gorical
v
a
riables
based o
n meanin
gful
dat
aset
s su
c
h
as
K
agg
le
.
F
i
g
u
r
e
s
1
a
n
d
2
d
e
p
i
c
t
t
h
e
d
a
t
a
s
e
t
s
C
o
l
u
m
n
a
n
d
S
p
i
n
e
,
r
e
s
p
e
c
t
i
v
e
l
y
.
F
o
r
t
he
d
e
f
i
n
i
t
i
o
n
o
f
t
h
e
c
o
l
u
m
n
c
o
d
e
,
w
e
a
s
s
i
gn
e
d
t
h
e
v
a
l
u
e
1
f
o
r
a
h
e
r
n
i
a
a
n
d
0
f
o
r
n
o
h
e
r
n
i
a
,
a
s
w
e
l
l
a
s
1
f
o
r
a
n
o
r
m
a
l
c
o
n
d
i
t
i
o
n
a
n
d
0
f
o
r
a
n
a
b
n
o
r
m
a
l
c
o
n
d
i
t
i
o
n
.
A
d
d
i
t
i
o
n
a
l
l
y
,
1
w
a
s
a
s
s
i
g
n
e
d
t
o
i
n
d
i
c
a
t
e
t
h
e
p
r
e
s
e
n
c
e
o
f
s
p
o
n
d
y
l
o
l
i
s
t
h
e
s
i
s
a
n
d
0
f
o
r
i
t
s
a
b
s
e
n
c
e
.
R
e
g
a
r
d
i
n
g
t
h
e
s
p
i
n
e
c
o
d
e
,
1
s
i
g
n
i
f
i
e
s
a
n
o
r
m
a
l
c
o
n
d
i
t
i
o
n
,
w
h
i
l
e
0
d
e
n
o
t
e
s
a
n
a
b
n
o
r
m
a
l
c
o
n
d
i
t
i
o
n
.
Figure
1.
Datas
et
Colu
mn
: T
he
f
irst si
x
c
olumns
r
e
pr
ese
nt
cru
ci
al
s
pin
e
m
easur
e
s, w
hile t
he
la
st col
umn
represe
nts the
t
arg
et
for
eac
h pati
ent
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
omp E
ng
IS
S
N:
20
88
-
8708
Compari
son
of
mach
i
ne
le
arn
ing
algorit
hms
to ide
ntif
y and
pr
eve
nt
…
(
C
hri
sti
an
Ova
ll
e
Pa
ulino
)
897
Figure
2.
Datas
et
Sp
i
ne
:
T
he
first ele
ve
n
c
olumns
r
e
pr
ese
nt
var
ia
bles r
e
pre
sentin
g diff
e
re
nt p
a
rts
of the
sp
ine
,
wh
il
e the
last
c
olu
m
n
cl
assi
fies the s
pin
e
as a
bnormal
or
nor
mal
We
meti
c
ulous
ly
f
oc
us
ed
on
cru
ci
al
s
pin
e
measu
res,
incl
ud
i
ng
pel
vic
i
ncide
nce,
pel
vic
ti
lt
,
lumb
a
r
lor
do
sis
a
ngle
,
sacral
ti
lt
,
pelvic
ra
diu
s,
de
gree
of
s
pondyl
olist
hesis,
pelv
ic
slop
e,
direct
ti
lt
,
tho
racic
s
lop
e,
cerv
ic
al
ti
lt
,
sacral
an
gle,
a
nd sco
li
oti
c
slo
pe.
This meti
culo
us
a
ppr
oach
to d
at
a
pre
par
at
io
n
insti
ll
s
co
nf
i
den
ce
in
the
th
oro
ughn
e
ss
of
our
r
esearch
.
It
la
ys
the
f
oundat
io
n
for
a
detai
le
d
a
nd
acc
ur
at
e
anal
ys
is
of
lo
w
bac
k
pro
blems in
th
e subse
qu
e
nt st
eps.
2.2. Cl
as
sific
ati
on met
hods
Six
mac
hi
ne
le
arn
i
ng
m
od
el
s
wer
e
selec
te
d f
or co
ns
ide
rati
on in
ou
r
re
sear
ch:
−
Lo
gisti
c
regre
ssion
(LR
):
E
valuates
t
he
c
onnecti
on
between
the
cat
e
gorical
de
pe
ndent
va
riable
(to
be
pr
e
dicte
d) an
d on
e
or m
ore in
dep
e
ndent
va
riables (
w
hich
i
nfl
ue
nce th
e
f
ormer) b
y
est
ima
ti
ng
probabil
it
ie
s
us
in
g
a
lo
gisti
c f
un
ct
io
n
[34
]
.
−
Suppor
t
vecto
r
machi
ne
(
SVM):
It
is
an
al
gorithm
su
it
ab
le
for
antic
ipat
ing
urba
n
lo
gi
sti
cs
dema
nd.
It
offer
s
s
pecific
ben
e
fits
i
n
s
ol
ving
pro
blems
with
li
mit
ed
da
ta
set
s
an
d
no
nlinear
functi
ons
an
d
ide
ntifyi
ng
patte
rn
s
in m
ulti
dimens
io
nal s
paces
[35
]
.
−
K
-
near
e
st
nei
ghbo
r
(KNN):
It
is
a
cl
assi
ficat
ion
meth
od
t
ha
t
determi
nes
a
data
po
i
nt's
cl
a
ss
by
exa
minin
g
it
s
cl
os
est
neig
hbors'
cl
asse
s.
These
nei
ghbo
rs
a
re
ide
ntifie
d
base
d
on
the
ir
distance
f
rom
the
data
point
,
commo
nly
cal
culat
ed usin
g E
uclidean
d
ist
a
nc
e
[
36]
.
−
Conv
olu
ti
onal
neural
net
wor
k
(CN
N)
:
It
is
a
sp
eci
al
iz
ed
de
ep
ne
ural
netw
ork
c
omprisin
g
in
pu
t,
hi
dd
e
n,
and
ou
t
pu
t
la
yer
s
.
T
he
hi
dden
la
ye
rs
are
pa
rtic
ularly
nota
ble
f
or
inc
orp
or
at
in
g
c
on
vo
l
ution
al
la
ye
rs
sp
eci
fical
ly
de
sign
e
d
t
o per
form
c
onvoluti
on
operati
ons
[
37]
.
−
Decisi
on
tree
(
DT)
:
It
is
a
si
mp
le
a
nd
easi
l
y
unde
rstan
da
bl
e
ap
proach
in
machine
le
ar
ni
ng
that
is
ap
plied
acro
s
s
var
io
us
discipli
nes
.
It
is
pract
ic
al
,
r
equ
i
res
le
ss
da
ta
,
an
d
pro
vide
s
inter
pr
et
a
bili
ty.
Decisi
on
tre
e
gen
e
rates m
od
el
s o
r
ga
nized
i
n
the
str
ucture
of
a
tree
a
nd ca
n be
us
ed
for b
oth
regressi
on
and cla
ssific
at
ion
pro
blems
[38]
.
−
Extreme
gradi
ent
boost
ing
(
XG
B
oost):
It
i
s
base
d
on
D
T
an
d
em
ploy
s
a
serial
trai
ni
ng
process
wi
th
da
ta
set
s
to
co
mb
ine
weak
e
r
predict
ors
int
o
str
onge
r
pre
dictors
.
It
it
er
at
ively
op
ti
mi
zes
the
obje
ct
ive
functi
on
un
ti
l r
eachin
g
it
s lo
w
est
v
al
ue
, at
w
hich p
oin
t t
he
t
rainin
g process
sto
ps
[
36]
.
This
ch
oice
w
as
gu
i
ded
by
pr
i
or
re
searc
h,
creati
ng
a
s
ol
id
an
d
evi
den
c
e
-
bac
ke
d
strat
egy
[4]
,
[8]
.
Fu
rt
hermo
re,
we
inc
orp
or
at
e
d
12
c
riti
cal
va
riables
int
o
th
e
model
de
vel
opment,
inclu
di
ng
pelvic
i
nciden
ce
,
pelvic
ti
lt
,
an
d
the
lum
ba
r
lo
r
do
sis
an
gle.
W
e
fo
ll
owe
d
the
sta
nd
a
rd
pra
ct
ic
e
of
di
vid
in
g
t
he
data
into
80
%
f
or
trai
ning
a
nd
20%
f
or
te
sti
ng,
wh
ic
h
ref
i
ne
d
our
meth
odol
ogy
a
nd
facil
it
at
ed
a
more
t
hor
ough
a
naly
sis
of
lumb
a
r
iss
ues.
Each
al
gorith
m
was
ta
il
or
e
d
with
s
pecific
set
ti
ng
s.
For
i
nst
ance,
l
og
ist
ic
regressi
on
wa
s
co
nf
i
gu
re
d
with
de
fau
lt
s
et
ti
ng
s,
s
up
port
vecto
r
mac
hin
e
with
a
li
near
kernel,
K
-
near
e
st
neig
hbor
with
E
uc
li
dean
distance
a
nd
10
nei
ghbors
i
n
the
Colu
m
n
da
ta
set
and
20
ne
ighbors
in
t
he
Sp
i
ne
dataset
,
conv
olu
ti
onal
neural
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
omp E
ng,
V
ol.
15
, No
.
1
,
Febr
uary
20
25
:
894
-
907
898
netw
ork
with
3
la
yer
s
of
32,
64,
a
nd
64
ne
uron
s
pe
r
la
ye
r,
and
10
0
ep
oc
hs.
Decisi
on
t
re
e
an
d
X
GB
oo
s
t
wer
e
config
ur
e
d usi
ng d
e
fa
ult set
ti
ng
s
.
2.3. Ev
alu
at
io
n met
ri
cs
In
t
his
sect
io
n,
we
def
i
ne
the
ap
propriat
e
e
valuati
on
met
r
ic
s
for
asses
sing
the
perform
ance
of
eac
h
al
gorithm
in
di
agnosi
ng
a
nd
treat
ing
lo
w
ba
ck
pro
blems.
T
he
sel
ect
e
d
me
tric
s
pr
ov
i
de
a
com
pr
e
he
ns
ive
vie
w
of
each
m
odel
's
ef
fecti
veness,
ens
uri
ng
it
me
et
s
t
he
cl
ie
nt's
sta
nd
a
rds.
We
evaluate
t
he
re
le
van
t
metri
cs
us
in
g
a co
nfusion
ma
trix
[
39]
. T
he m
et
rics ch
os
e
n f
or
e
valuati
on
are as
f
ollow
s:
−
Pr
eci
sio
n (PR)
Pr
eci
sio
n
mea
su
res
t
he
perf
ormance
of
a
n
ML
al
gorith
m
by
dete
rmi
ning
the
rati
o
of
c
orrectl
y
pr
e
dicte
d
posit
ive
case
s
t
o
t
he
total
predict
e
d
posit
ive
case
s.
W
her
e
re
presents
act
ual
posit
ive
va
lue
s,
and
re
pr
ese
nts
false p
os
it
ive
valu
es. I
t i
s
calc
ula
te
d
as
(1)
:
(
)
=
+
(1)
−
Re
cal
l (RC
)
Re
cal
l
ind
ic
at
es
how
well
the
m
od
el
i
de
ntifie
d
al
l
c
onfirme
d
cases
in
a
giv
e
n
cl
a
ss.
T
his
is
par
ti
cula
rly
rel
evan
t
in
the
medical
c
onte
xt
wh
e
re
accu
rate
detect
ion
of
al
l
lu
mb
a
r
le
sion
s
is
c
riti
cal
f
or
eff
ect
ive
treat
ment. It
is cal
c
ulate
d
as
(
2)
:
(
)
=
+
(2)
−
F1
-
Score
The
F1
-
sc
ore
is
a
ke
y
metr
ic
f
or
asse
ssin
g
l
umbar
i
njury
diag
nosti
c
models.
It
re
presents
the
harmo
nic
mea
n
of
pr
eci
si
on
and
recall
,
pro
vid
in
g
a
more
balance
d
performa
nce
meas
ur
e
[40
]
.
It
is
co
mpute
d
as
(
3)
:
1
−
=
2
.
+
(3)
Fu
rt
hermo
re,
the
a
rea
under
the
cu
r
ve
(
A
U
C)
is
util
iz
ed
t
o
e
valuate
th
e
discriminati
on
abili
ty
of
a
bin
a
ry
cl
assifi
cat
ion
model.
A
higher
AU
C
val
ue
i
nd
ic
at
e
s
a
higher
acc
ur
ac
y
of
th
e
M
L
al
gorith
m.
A
UC
is
commo
nly
us
e
d
i
n
the
c
on
te
xt
of
the
R
OC
cu
rv
e
,
w
hich
rep
rese
nts
the
cl
assifi
cat
ion
predict
io
n
res
ults
in
a
two
-
dime
ns
io
na
l plane
[
41]
.
These
m
et
rics
are
ty
pical
ly
analyze
d
us
ing
sta
ti
sti
cal
appr
oach
es
.
A
lt
ho
ug
h
creati
ng
r
ules
f
or
pr
e
dicti
ng
res
ul
ts
may
in
vo
l
ve
rou
gh
-
set
te
chnolo
gy.
When
in
formati
on
about
gr
a
nu
le
s
is
pro
vid
e
d
in
th
e
pr
e
dicti
on pr
oc
ess, a
n
inte
rpre
ta
ti
on
of t
he
“
fu
zzi
ness
”
ba
se
d on r
ough sets
can be ac
hie
ve
d
[42]
.
2.
4.
Web
en
viron
me
nt
in
te
gr
at
i
on
At
t
his
project
sta
ge
,
we
fo
c
us
on
inte
gr
at
ing
t
he
most
e
f
fecti
ve
m
od
el
into
a
we
b
en
vi
ronme
nt
to
ens
ur
e
t
he
de
ve
lop
me
nt
of
a
so
l
ution
co
nd
ucive
to
a
dopt
ing
t
hese
te
c
hnol
og
ie
s
in
on
li
ne
me
dical
pract
ic
e
[23]
.
We
ha
ve
impleme
nted
an
inte
rf
ace
w
it
h
G
rad
i
o,
a
n
op
e
n
-
sou
rce
P
ython
li
br
a
r
y,
to
achie
ve
t
his
goal
.
Gr
a
dio
plays
a
cru
ci
al
r
ole
in
impleme
nting
a
m
odel
in
an
acce
ssibl
e
web
e
nviro
nm
e
nt,
sig
nifi
cantl
y
enh
a
ncin
g
t
he
sp
ee
d
a
nd
e
f
fecti
ven
es
s
of
decisi
on
-
maki
ng,
pa
rtic
ularl
y
in
healt
h
for
disease
co
nt
ro
l
a
nd
pr
e
ve
ntion pla
ns
[43]
.
Gr
a
dio
is
buil
t
upon
a
we
b
int
erf
ace
t
hat
faci
li
ta
te
s
interact
i
on
with
an
ML
sy
ste
m
us
i
ng
models
a
nd
al
gorithms.
It
al
lows
f
or
i
nputti
ng
e
xtensi
ve
data
for
ear
ly
disease
dia
gnos
is
[
44]
.
H
ence,
Gr
a
dio
i
s
an
excell
ent
c
ho
i
ce
f
or
integ
rati
ng
the
sel
ect
ed
m
odel
int
o
a
we
b
e
nviro
nme
nt,
ena
blin
g
the
rap
i
d
c
rea
ti
on
of
us
er
inter
faces
for
le
ar
ning
m
od
el
s
.
This
de
ta
il
ed
a
ppr
oach,
ma
rked
by
t
hor
ough
a
nd
preci
se
evaluati
on,
has
bee
n
im
pleme
nted
i
n
the
project
.
It
ha
s
offe
red
a
c
omplet
e
a
nd
ac
cur
at
e
a
ssessm
ent
of
the
model'
s
performa
nce
i
n
the
m
ul
ti
cl
ass
cl
assifi
cat
ion
t
ask,
c
onfi
rming
it
s
ef
fecti
ve
ne
ss
f
or
dia
gnosi
ng
a
nd
treat
in
g
lu
mb
a
r
le
sio
ns
.
These
pr
oc
edures
are il
lustrate
d i
n
the
sc
hemati
c d
ia
gram
pres
ented
i
n
Fi
gure
3
.
Th
us
,
a
proces
s
flo
w
dia
gr
a
m
il
lustrati
ng
the
seq
ue
nce
of
ta
s
ks
a
nd
a
ct
ivit
ie
s
was
con
si
der
e
d,
a
s
dep
ic
te
d
in
Figure
4.
The
pr
ocess
i
niti
at
es
wh
e
n
t
he
us
er
enters
t
he
s
ys
t
em
an
d
i
nputs
the
patie
nt's
da
ta
for
evaluati
on.
S
ubse
qu
e
ntly,
the
syst
em
validat
es
the
acc
ur
ac
y
of
the
e
ntere
d
data.
I
n
c
ase
of
a
ny
i
nacc
uraci
es,
the
s
ys
te
m
will
flag
a
n
e
rror.
Fo
ll
owin
g
validat
ion,
t
he
s
ys
t
em
processes
the
e
nter
e
d
data
thr
ough
t
he
t
r
ai
ned
model,
pro
vid
i
ng the
us
e
r wit
h
the
r
es
ults.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
omp E
ng
IS
S
N:
20
88
-
8708
Compari
son
of
mach
i
ne
le
arn
ing
algorit
hms
to ide
ntif
y and
pr
eve
nt
…
(
C
hri
sti
an
Ova
ll
e
Pa
ulino
)
899
Figure
3. Sc
he
mati
c d
ia
gram
of predict
ive
m
od
el
us
i
ng in
t
he
trai
ning m
odel
s
Figure
4. Proce
ss
flo
wc
har
t
3.
RESU
LT
S
AND DI
SCUS
S
ION
In
this
sect
io
n,
we
present
the
resu
lt
s
obta
ined
t
o
f
urt
her
ex
plore
t
he
ef
fecti
ven
e
ss
of
t
hese
al
gorithms
in
healt
hcar
e
set
ti
ng
s
.
The
se
t
ools
not
only
im
pro
ve
acc
urac
y
i
n
i
de
ntify
in
g
a
nd
preve
nting
lo
w
back
inju
ries
but
al
so
ha
ve
t
he
pote
ntial
to
offer
ne
w
pers
pecti
ves
i
n
me
dical
care.
T
his
ins
piring
pote
ntial
pro
vid
es a
s
olid
fou
nd
at
io
n f
or futu
re r
e
sear
ch
a
nd cli
nical
dev
el
opment
.
3.1. Disp
ersi
on wit
h da
ta se
pa
r
at
i
on
As
sta
te
d
in
se
ct
ion
2.2,
we
a
dopted
the
sta
nd
a
r
d
80
-
20
ra
ti
o
to
par
ti
ti
on
the
data.
S
ubs
equ
e
ntly,
to
ob
ta
in
detai
le
d
i
ns
ig
hts
i
nto
our
va
riables
an
d
cl
asses
,
Fig
ures
5(
a
)
t
o
5
(c)
de
pict
the
pl
ot
te
d
Col
umn
dat
aset
s.
It
is
obse
rv
e
d
that
the
C
olumn
dataset
s
e
xhibit
an
asce
ndin
g
tre
nd.
Simi
la
rly,
Fig
ur
e
6
il
lustrate
s
the
data
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
omp E
ng,
V
ol.
15
, No
.
1
,
Febr
uary
20
25
:
894
-
907
900
div
isi
on
of
the
Spin
e
dataset
,
re
veali
ng
a
l
ow
c
orrelat
ion
am
ong
it
s
va
riables,
a
te
sta
ment
to
the
m
od
el
's
abili
ty to ha
nd
l
e comple
x d
at
a w
it
h
c
onfi
dence.
(a)
(b)
(c)
Figure
5. Scat
te
r plots
of
Col
umn
dataset
of
(a)
he
rn
ia
a
nd
no h
e
r
nia, (b
) normal a
nd a
bnormal
,
an
d
(c)
spo
ndylo
li
s
thesis
an
d n
o
s
pondyl
olist
hesis
Figure
6. Scat
te
r plot
of
t
he
Spine
dataset
of
normal a
nd a
bnormal
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
omp E
ng
IS
S
N:
20
88
-
8708
Compari
son
of
mach
i
ne
le
arn
ing
algorit
hms
to ide
ntif
y and
pr
eve
nt
…
(
C
hri
sti
an
Ova
ll
e
Pa
ulino
)
901
3.2. M
od
el
tr
aining
Fo
r
t
he
trai
ni
ng
of
models
on
the
C
olumn
dataset
,
th
e
lo
gisti
c
r
eg
r
ession
model,
trai
ne
d
with
var
ia
bles
s
uch
as
_
,
_
,
_
_
,
_
,
_
,
and
_
ℎ
,
pro
vid
e
d
a
c
omp
rehensive
diag
nosis
f
or
eac
h
case,
cl
assi
fy
i
ng
t
he
conditi
ons
as
he
rn
ia
,
s
pond
ylo
li
sthesis,
or
st
and
a
r
d
bac
k.
F
igures
7(a)
t
o
7(
c
)
is
pr
e
sent
ed
bel
ow
t
o
vi
su
al
iz
e
the
trai
ning
s
pe
ci
fi
cs
of
each
model
in
the
Colu
mn
datase
t.
Conve
rsely,
within
t
he
Spi
ne
dataset
,
the
LR
model
was
tr
ai
ned
with
va
riables
suc
h
a
s
_
,
_
,
ℎ
_
,
_
,
_
,
a
n
d
_
,
yieldi
ng
a
m
ore
ge
ner
al
iz
ed
ou
tc
om
e
by
cl
assi
fy
i
ng
the
s
pine
as
“abno
rmal”
or
“normal.”
Fi
gure
8
is
pro
vid
e
d
belo
w
to
off
er
a
detai
le
d
i
nsi
gh
t
into
the
i
nput
of
eac
h
model
within t
he
Sp
i
ne
dataset
.
It
is
es
sentia
l
t
o
highli
ght
tha
t
the
R
OC
c
urve
was
em
ployed
f
or
trai
ning
meas
ur
eme
nt
,
ex
hib
it
in
g
sat
isfact
ory
ou
tc
om
es,
pa
rtic
ularly
f
or
LR
i
n
t
he
C
olumn
dataset
,
with
c
urves
of
0.70
i
n
Fi
gure
7(a)
,
0.90
in
curve Figure
7(b),
an
d
0.9
9
i
n
Fig
ur
e
7(c)
. In
c
ontrast
,
a
l
ower
trai
ni
ng
sc
or
e
of
0.4
8
is
observe
d
for
the
Spine
dataset
,
as
depi
ct
ed
in
Fig
ure
8.
H
ow
e
ver,
oth
e
r
metri
cs
will
be
util
iz
ed
to
as
sess
it
s
pr
e
dicti
ve
ca
pa
bili
ty
adequate
ly
.
(a)
(b)
(c)
Figure
7. Trai
ni
ng
plo
ts
of
C
ol
umn
dataset
for
(a
) her
nia a
nd no her
nia,
(b) normal
a
nd a
bnormal
,
an
d
(c)
spo
ndylo
li
s
thesis
an
d n
o
s
pondyl
olist
hesis
R
e
g
a
r
d
i
n
g
t
h
e
p
e
r
f
o
r
m
a
n
c
e
m
e
t
r
i
c
s
,
a
c
o
m
p
a
r
a
t
i
v
e
a
n
a
l
y
s
i
s
o
f
t
h
e
p
e
r
f
o
r
m
a
n
c
e
o
f
t
h
e
6
s
e
l
e
c
t
e
d
ML
m
o
d
e
l
s
i
s
p
r
e
s
e
n
t
e
d
i
n
T
a
b
l
e
1
,
w
h
e
r
e
t
h
e
y
w
e
r
e
e
v
a
l
u
a
t
e
d
us
i
n
g
t
h
e
s
e
l
e
c
t
e
d
m
e
t
r
i
c
s
.
F
i
r
s
t
l
y
,
t
h
e
m
e
t
r
i
c
s
v
a
l
u
e
s
f
o
r
t
h
e
C
o
l
u
m
n
d
a
t
a
s
e
t
a
r
e
p
r
e
s
e
n
t
e
d
,
f
o
l
l
o
w
e
d
b
y
t
h
o
s
e
f
o
r
t
h
e
S
p
i
n
e
d
a
t
a
s
e
t
f
o
r
e
a
c
h
m
od
e
l
;
i
t
i
s
n
o
t
e
w
o
r
t
h
y
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
omp E
ng,
V
ol.
15
, No
.
1
,
Febr
uary
20
25
:
894
-
907
902
t
h
a
t
t
h
e
o
r
d
e
r
o
f
t
h
e
d
a
t
a
s
e
t
s
r
e
m
a
i
n
s
c
o
n
s
i
s
t
e
n
t
(
r
e
f
e
r
t
o
T
a
b
l
e
1
)
.
A
d
d
i
t
i
o
n
a
l
l
y
,
i
t
c
a
n
b
e
o
b
s
e
r
v
e
d
t
h
a
t
t
h
e
L
R
m
o
d
e
l
e
x
h
i
b
i
t
e
d
e
x
c
e
p
t
i
o
n
a
l
pe
r
f
o
r
m
a
n
c
e
i
n
b
o
t
h
d
a
t
a
s
e
t
s
.
F
o
r
t
h
e
S
p
i
n
e
da
t
a
s
e
t
,
i
t
a
t
t
a
i
n
e
d
a
p
r
e
c
i
s
i
o
n
o
f
8
6
%
,
r
e
c
a
l
l
o
f
8
5
%
,
a
n
d
F1
-
s
c
o
r
e
o
f
8
6
%
,
i
n
d
i
c
a
t
i
v
e
o
f
i
t
s
h
i
g
h
a
c
c
u
r
a
c
y
i
n
i
d
e
nt
i
f
y
i
n
g
p
o
s
i
t
i
v
e
a
n
d
n
e
g
a
t
i
v
e
c
a
s
e
s
,
p
r
o
v
i
d
i
n
g
a
d
e
t
a
i
l
e
d
d
i
a
g
n
o
s
i
s
b
y
c
a
t
e
g
o
r
i
z
i
n
g
t
h
e
s
p
i
n
e
i
n
t
o
s
p
e
c
i
f
i
c
c
o
n
d
i
t
i
on
s
s
u
c
h
a
s
h
e
r
n
i
a
,
s
p
o
n
d
y
l
o
l
i
s
t
h
e
s
i
s
,
o
r
n
o
r
m
a
l
ba
c
k
.
F
u
r
t
h
e
r
m
o
r
e
,
o
n
t
h
e
S
p
i
ne
d
a
t
a
s
e
t
,
t
h
e
L
R
m
o
d
e
l
a
c
h
i
e
v
e
d
a
p
r
e
c
i
s
i
on
o
f
7
0
%
,
a
r
e
c
a
l
l
o
f
7
1
%
,
a
n
d
a
n
F1
-
s
c
o
r
e
o
f
6
3
%
.
T
h
o
u
g
h
s
t
i
l
l
c
o
m
m
e
n
d
a
b
l
e
,
t
h
e
m
o
d
e
l
b
i
n
a
r
i
l
y
c
l
a
s
s
i
f
i
e
s
t
he
s
p
i
n
e
a
s
“
a
b
n
o
r
m
a
l
”
o
r
“
n
o
r
m
a
l
,
”
o
f
f
e
r
i
n
g
a
m
o
r
e
g
e
n
e
r
a
l
i
z
e
d
d
i
a
g
n
o
s
i
s
i
n
t
h
i
s
d
a
t
a
s
e
t
.
T
h
e
s
e
f
i
n
d
i
n
g
s
u
n
d
e
r
s
c
o
r
e
t
h
e
L
R
m
o
d
e
l
'
s
a
d
a
p
t
a
b
i
l
i
t
y
t
o
d
i
f
f
e
r
e
n
t
d
a
t
a
s
e
t
s
a
n
d
t
h
e
s
i
g
n
i
f
i
c
a
n
t
r
o
l
e
o
f
s
e
l
e
c
t
i
n
g
p
e
r
t
i
n
e
n
t
v
a
r
i
a
b
l
e
s
f
o
r
e
a
c
h
d
i
a
g
n
o
s
t
i
c
a
s
s
e
s
s
m
e
nt
c
o
n
t
e
x
t
,
h
i
g
h
l
i
g
h
t
i
ng
t
h
e
i
m
p
o
r
t
a
n
c
e
o
f
y
o
u
r
w
o
r
k
i
n
t
h
i
s
f
i
e
l
d
.
Figure
9
disp
l
ays
a
heatma
p
il
lustrati
ng
t
he
co
rr
el
at
io
n
be
tween
t
he
vari
ables
in
t
he
da
ta
set
.
This
visu
al
iz
at
ion
ai
ds
in
ide
ntif
yin
g
patte
r
ns
an
d
tren
ds
within
the
data
by
s
howca
sin
g
their
relat
ion
s
hip
s
.
In
this
heatmap
,
re
d
denotes
a
pos
it
ive
correla
ti
on,
with
m
ore
intense
s
ha
des
in
dicat
ing
a
stronge
r
posit
ive
correla
ti
on
;
bl
ue
si
gn
i
fies
a
ne
gative
co
r
relat
ion
,
with
da
rk
e
r
hues
re
pr
ese
ntin
g
a
str
onge
r
ne
gative
correla
ti
on. W
hite i
nd
ic
at
es t
he
a
bs
e
nce
of
c
orrelat
ion
.
Figure
8
.
Tr
ai
ni
ng
plo
t
of the
Sp
i
ne
dataset
for n
ormal
and
abno
rmal
Table
1.
T
raini
ng r
es
ults
of
ML m
od
el
s
A
B
C
D
Mod
els
Metr
ics
Precisio
n
Recall
F
l
-
sco
re
LR
0
.86
0
.85
0
.86
0
.70
0
.71
0
.63
SVM
0
.85
0
.85
0
.85
0
.79
0
.69
0
.57
CNN
0
.80
0
.81
0
.80
0
.62
0
.53
0
.56
K
-
NN
0
.72
0
.71
0
.71
0
.60
0
.65
0
.53
DT
0
.66
0
.66
0
.66
0
.56
0
.45
0
.49
XGBo
o
st
0
.80
0
.81
0
.80
0
.67
0
.69
0
.68
3.3. Implem
e
nt
ati
on i
n
G
ra
dio sof
tware
T
h
e
s
y
s
t
e
m
'
s
gr
a
p
h
i
c
a
l
i
n
t
e
r
f
a
c
e
o
f
f
e
r
s
a
u
s
e
r
-
f
r
i
e
n
d
l
y
e
x
p
e
r
i
e
n
c
e
,
f
e
a
t
u
r
i
n
g
c
l
e
a
r
l
y
d
e
f
i
n
e
d
d
a
t
a
e
n
t
r
y
s
e
c
t
i
o
n
s
f
o
r
i
n
p
u
t
t
i
n
g
r
e
l
e
v
a
nt
i
n
f
o
r
m
a
t
i
o
n
a
b
o
u
t
l
o
w
b
a
c
k
i
n
j
u
r
i
e
s
.
U
s
e
r
s
c
a
n
p
r
o
v
i
d
e
d
e
t
a
i
l
s
s
u
c
h
a
s
p
a
i
n
l
o
c
a
t
i
o
n
,
i
nt
e
n
s
i
t
y
,
a
s
s
o
c
i
a
t
e
d
s
y
m
p
t
o
m
s
,
a
n
d
o
t
h
e
r
e
s
s
e
n
t
i
a
l
f
a
c
t
o
r
s
.
A
d
d
i
t
i
o
n
a
l
l
y
,
t
h
e
r
e
i
s
a
d
e
d
i
c
a
t
e
d
s
e
c
t
i
o
n
f
o
r
v
i
s
u
a
l
i
z
i
n
g
d
i
a
g
n
o
s
t
i
c
r
e
s
u
l
t
s
,
i
n
c
l
u
d
i
n
g
t
h
e
t
y
p
e
o
f
l
o
w
b
a
c
k
i
n
j
u
r
y
,
t
r
e
a
t
m
e
n
t
r
e
c
o
m
m
e
n
d
a
t
i
o
n
s
,
a
n
d
r
e
h
a
b
i
l
i
t
a
t
i
o
n
s
t
r
a
t
e
g
i
e
s
.
T
hi
s
t
o
o
l
i
s
d
e
s
i
g
n
e
d
t
o
c
a
t
e
r
t
o
b
o
t
h
e
x
p
e
r
i
e
n
c
e
d
a
n
d
i
n
e
x
p
e
r
i
e
n
c
e
d
u
s
e
r
s
,
e
n
s
u
r
i
n
g
e
a
s
e
o
f
u
s
e
a
n
d
i
n
t
e
r
a
c
t
i
o
n
.
I
t
f
a
c
i
l
i
t
a
t
e
s
f
l
u
i
d
n
a
v
i
g
a
t
i
on
b
e
t
w
e
e
n
d
i
f
f
e
r
e
n
t
s
e
c
t
i
o
n
s
a
n
d
f
u
n
c
t
i
o
n
a
l
i
t
i
e
s
,
p
r
e
s
e
n
t
i
n
g
a
c
c
ur
a
t
e
a
n
d
p
e
r
s
o
n
a
l
i
z
e
d
r
e
s
u
l
t
s
a
n
d
r
e
c
o
m
m
e
n
d
a
t
i
o
n
s
r
e
g
a
r
d
i
n
g
l
o
w
e
r
b
a
c
k
i
n
j
u
r
i
e
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
omp E
ng
IS
S
N:
20
88
-
8708
Compari
son
of
mach
i
ne
le
arn
ing
algorit
hms
to ide
ntif
y and
pr
eve
nt
…
(
C
hri
sti
an
Ova
ll
e
Pa
ulino
)
903
I
m
p
l
e
m
e
n
t
e
d
i
n
a
r
o
b
u
s
t
a
nd
e
a
s
i
l
y
a
c
c
e
s
s
i
b
l
e
w
e
b
e
n
v
i
r
o
n
m
e
n
t
,
t
h
e
s
o
l
u
t
i
o
n
e
n
s
u
r
e
s
t
h
a
t
u
s
e
r
s
,
i
n
c
l
u
d
i
n
g
t
h
o
s
e
l
a
c
k
i
n
g
t
e
c
h
n
i
c
a
l
e
x
p
e
r
t
i
s
e
,
c
a
n
e
f
f
e
c
t
i
v
e
l
y
u
t
i
l
i
z
e
i
t
t
o
di
a
g
n
o
s
e
a
n
d
t
r
e
a
t
l
o
w
b
a
c
k
i
nj
u
r
i
e
s
.
L
e
v
e
r
a
g
i
n
g
G
r
a
d
i
o
f
u
r
t
h
e
r
e
n
h
a
n
c
e
s
t
h
e
t
oo
l
'
s
u
t
i
l
i
t
y
,
e
n
a
b
l
i
n
g
p
r
o
f
e
s
s
i
o
n
a
l
s
f
r
o
m
d
i
v
e
r
s
e
d
i
s
c
i
p
l
i
n
e
s
a
n
d
c
l
i
n
i
c
a
l
s
e
t
t
i
n
gs
t
o
b
e
n
e
f
i
t
f
r
o
m
t
h
i
s
i
n
n
o
v
a
t
i
v
e
s
o
l
u
t
i
o
n
i
n
m
a
n
a
g
i
n
g
s
p
i
n
e
h
e
a
l
t
h
-
r
e
l
a
t
e
d
c
a
s
e
s
.
F
i
gu
r
e
1
0
d
e
p
i
c
t
s
t
h
e
s
y
s
t
e
m
'
s
g
r
a
p
h
i
c
a
l
u
s
e
r
i
n
t
e
r
f
a
c
e
,
i
l
l
u
s
t
r
a
t
i
n
g
i
t
s
a
c
c
e
s
s
i
b
i
l
i
t
y
t
o
a
b
r
o
a
d
a
u
d
i
e
n
c
e
.
T
h
i
s
i
n
t
e
r
f
a
c
e
e
m
p
o
w
e
r
s
h
e
a
l
t
h
c
a
r
e
p
r
o
f
e
s
s
i
o
n
a
l
s
a
n
d
o
t
h
e
r
u
s
e
r
s
w
i
t
h
i
nt
u
i
t
i
v
e
a
c
c
e
s
s
t
o
d
i
a
g
n
o
s
e
a
n
d
t
r
e
a
t
l
o
w
b
a
c
k
i
n
j
u
r
i
e
s
.
Figure
9. Heat
correla
ti
on ma
p of t
he varia
bl
es takin
g
i
nto
a
ccount in
the
d
at
aset
Sp
i
ne
Figure
10. Syst
em
gra
ph
ic
al
use
r
in
te
rf
ace
Evaluation Warning : The document was created with Spire.PDF for Python.