C
om
p
u
t
e
r
S
c
ie
n
c
e
an
d
I
n
f
or
m
at
io
n
T
e
c
h
n
ol
ogi
e
s
V
ol
.
6
, N
o.
3
,
N
ove
m
be
r
20
25
, pp.
245
~
252
I
S
S
N
:
2722
-
3221
,
D
O
I
:
10.11591/cs
it
.
v
6
i
3
.
pp
245
-
252
245
Jou
r
n
al
h
om
e
page
:
ht
tp
:
//
ia
e
s
pr
ime
.c
om
/i
nde
x
.php/c
s
it
A
d
u
al
-
m
o
d
e
l
m
ac
h
i
n
e
l
e
ar
n
i
n
g ap
p
r
oac
h
t
o m
e
d
i
c
ar
e
f
r
au
d
d
e
t
e
c
t
i
on
:
c
om
b
i
n
i
n
g u
n
su
p
e
r
vi
se
d
an
om
al
y
d
e
t
e
c
t
i
on
w
i
t
h
su
p
e
r
vi
se
d
l
e
ar
n
i
n
g
Je
s
u
M
ar
c
u
s
I
m
m
an
u
ve
l
A
r
oc
k
ia
s
a
m
y, G
ow
r
i
s
h
an
k
ar
B
h
oop
at
h
i
L
e
a
di
ng H
e
a
l
t
hc
a
r
e
C
om
pa
ny,
R
i
c
hm
ond, V
i
r
gi
ni
a
A
r
t
ic
le
I
n
f
o
A
B
S
T
R
A
C
T
A
r
ti
c
le
h
is
to
r
y
:
R
e
c
e
iv
e
d
A
pr
3
,
2025
R
e
vi
s
e
d
M
a
y
27
,
2025
A
c
c
e
pt
e
d
J
un
13
,
2025
Medicare
fraud,
costing
$54.35
billion
in
imprope
r
payments
in
2024,
undermines
U.S.
healthcare
by
draining
resources
meant
for
vul
nerable
populations.
Traditional
detection
methods
struggle
with
reactive
d
esigns,
high
false
positives,
and
reliance
on
scarce
labeled
data,
exacerbated
by
a
0.017%
fraud
prevalence.
This
paper
proposes
a
dual
-
model
m
achine
learning
framewo
rk
to
tackle
these
challenge
s.
Unsupervise
d
an
omaly
detection
uses
cluster
-
based
local
outlier
factor
(
CBLOF
)
and
empirical
cumulati
ve
outli
er
detection
(
ECOD
)
to
identify
novel
fraud
patterns
across
37
million
records.
These
findings
are
validated
by
the
list
of
ex
cluded
individuals/entitie
s
(
LEIE
)
.
Supervised
classification,
with
C4.5
d
ecision
trees
and
logistic
regression
,
refines
these
anomalies
using
an
80:20
balanced
dataset,
reducing
false
positives
by
63%.
Key
innovations
include
hybrid
sampling
to
address
class
imbalance,
LEIE
integration
for
l
abeled
validation,
and
parallelized
processing
of
2.1
million
claims
hourly.
Achieving
an
are
a
under
the
curve
(
AUC
)
,
a
measure
of
model
accur
acy,
of
88.3%,
this
approach
outperforms
single
-
model
systems
by
24%,
bl
ending
explorato
ry
detection
with
actionabl
e
precision
.
This
scalable,
interp
retable
framework
potential
ly
advances
fraud
detection,
safeguarding
public
funds
and
Medicare’s
integrity
with
a
practical,
adaptable
solution
for
ev
olving
threats.
K
e
y
w
o
r
d
s
:
A
r
ti
f
ic
ia
l
in
te
ll
ig
e
nc
e
C
lu
s
te
r
-
ba
s
e
d l
oc
a
l
out
li
e
r
f
a
c
to
r
E
m
pi
r
ic
a
l
c
um
ul
a
ti
ve
out
li
e
r
de
te
c
ti
on
M
a
c
hi
ne
l
e
a
r
ni
ng
M
e
di
c
a
r
e
f
r
a
ud
U
ns
upe
r
vi
s
e
d l
e
a
r
ni
ng
This is an
open
acce
ss artic
le unde
r the
CC BY
-
SA
license.
C
or
r
e
s
pon
di
n
g A
u
th
or
:
J
e
s
u
M
a
r
c
us
I
m
m
a
nuve
l
A
r
oc
ki
a
s
a
m
y
L
e
a
di
ng
H
e
a
lt
hc
a
r
e
C
om
pa
ny
R
ic
hm
ond, Vir
gi
ni
a
, U
ni
te
d S
ta
te
s
of
A
m
e
r
ic
a
E
m
a
il
:
je
s
um
a
r
c
us
@
gm
a
il
.c
om
1.
I
N
T
R
O
D
U
C
T
I
O
N
M
e
di
c
a
r
e
f
r
a
ud,
c
os
ti
ng
$54.35
bi
ll
io
n
in
i
m
pr
ope
r
pa
ym
e
nt
s
in
2024
[
1
]
,
unde
r
m
in
e
s
U
.S
.
he
a
lt
hc
a
r
e
by
dr
a
in
in
g
r
e
s
our
c
e
s
f
or
vul
ne
r
a
bl
e
popula
ti
ons
.
T
r
a
di
ti
ona
l
f
r
a
ud
de
te
c
ti
on
s
ys
te
m
s
,
r
e
ly
in
g
on
r
ul
e
-
ba
s
e
d
a
udi
ts
or
s
upe
r
vi
s
e
d
m
a
c
hi
ne
le
a
r
ni
ng,
f
a
c
e
c
r
it
ic
a
l
li
m
it
a
ti
ons
.
B
r
e
nna
n
[
2]
hi
ghl
ig
ht
e
d
th
e
c
la
s
s
im
ba
la
nc
e
c
r
is
is
, w
it
h f
r
a
udul
e
nt
c
a
s
e
s
c
om
pr
is
in
g only 0.017%
of
c
la
im
s
, l
e
a
di
ng t
o hi
gh f
a
ls
e
ne
ga
ti
ve
r
a
te
s
(
ove
r
40%
)
in
s
upe
r
vi
s
e
d
m
ode
ls
.
S
ta
ti
s
ti
c
a
l
m
e
th
ods
,
a
s
not
e
d
by
B
ol
to
n
a
nd
H
a
nd
[
3]
,
s
tr
uggl
e
to
a
da
pt
to
e
vol
vi
ng f
r
a
ud pa
tt
e
r
ns
, m
is
s
in
g nove
l
s
c
h
e
m
e
s
.
R
e
c
e
nt
uns
up
e
r
vi
s
e
d
a
ppr
oa
c
h
e
s
,
s
uc
h
a
s
G
r
e
s
oi
e
t
al
.
[
4]
,
la
c
k
la
be
le
d
va
li
da
ti
on,
r
e
s
ul
ti
ng
in
hi
gh
f
a
ls
e
pos
it
iv
e
s
[
5]
,
w
hi
le
s
c
a
la
bi
li
ty
is
s
ue
s
hi
nde
r
pr
oc
e
s
s
in
g
la
r
ge
da
ta
s
e
ts
li
ke
th
e
37
m
il
li
on
M
e
di
c
a
r
e
c
la
im
s
[
6]
.
O
ur
dua
l
-
m
ode
l
f
r
a
m
e
w
or
k
a
ddr
e
s
s
e
s
th
e
s
e
ga
ps
b
y
in
te
gr
a
ti
ng
uns
upe
r
vi
s
e
d
a
nom
a
ly
de
te
c
ti
on
(
c
lu
s
te
r
-
ba
s
e
d
lo
c
a
l
out
li
e
r
f
a
c
to
r
(
C
B
L
O
F
)
a
nd
e
m
pi
r
ic
a
l
c
um
ul
a
ti
ve
out
li
e
r
de
te
c
ti
on
(
E
C
O
D
)
)
w
it
h
s
upe
r
vi
s
e
d
c
la
s
s
if
ic
a
ti
on
(
C
4.5
de
c
is
io
n
tr
e
e
s
a
nd
lo
gi
s
ti
c
r
e
gr
e
s
s
io
n
)
,
le
ve
r
a
gi
ng
li
s
t
of
e
xc
lu
d
e
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2722
-
3221
C
om
put
S
c
i
I
nf
T
e
c
hnol
,
V
ol
.
6
, N
o.
3
,
N
ove
m
be
r
20
25
:
245
-
252
246
in
di
vi
dua
ls
/e
nt
it
ie
s
(
L
E
I
E
)
va
li
da
ti
on
[
7]
a
nd
hyb
r
id
s
a
m
pl
in
g
to
m
it
ig
a
te
c
la
s
s
im
ba
la
nc
e
.
T
hi
s
a
ppr
oa
c
h
r
e
duc
e
s
f
a
ls
e
pos
it
iv
e
s
by
63%
a
nd
pr
oc
e
s
s
e
s
2.1
m
il
li
on
c
la
im
s
hour
ly
,
of
f
e
r
in
g
a
s
c
a
la
bl
e
,
in
te
r
pr
e
ta
bl
e
s
ol
ut
io
n f
or
r
e
a
l
-
w
or
ld
de
pl
oym
e
nt
.
L
im
it
at
io
n
s
of
t
r
ad
it
io
n
al
d
e
t
e
c
t
io
n
m
e
t
h
od
s
T
r
a
di
ti
ona
l
f
r
a
ud
de
te
c
ti
on
s
ys
te
m
s
a
r
e
w
id
e
ly
us
e
d
but
ha
ve
th
e
ir
li
m
it
a
ti
ons
.
T
he
y
r
e
ly
on
r
ul
e
-
ba
s
e
d a
udi
ts
or
s
up
e
r
vi
s
e
d m
a
c
hi
ne
l
e
a
r
ni
ng, whic
h c
a
n l
e
a
d t
o t
hr
e
e
c
r
it
ic
a
l
f
la
w
s
:
−
R
e
a
c
ti
ve
de
s
ig
n:
r
e
a
c
ti
ve
de
s
ig
n f
oc
us
e
s
on known f
r
a
ud pa
tt
e
r
ns
but
m
is
s
e
s
ne
w
s
c
he
m
e
s
.
−
H
ig
h
f
a
ls
e
pos
it
iv
e
s
:
ove
r
70%
of
f
la
gge
d c
la
im
s
a
r
e
l
e
gi
ti
m
a
te
,
w
a
s
ti
ng i
nve
s
ti
ga
ti
ve
r
e
s
our
c
e
s
.
−
L
a
be
l
de
pe
nde
nc
y:
s
upe
r
vi
s
e
d
m
a
c
hi
ne
l
e
a
r
ni
ng r
e
qui
r
e
s
c
os
tl
y
, s
c
a
r
c
e
l
a
be
ll
e
d d
a
ta
.
W
hi
le
r
e
c
e
nt
s
tu
di
e
s
de
m
ons
tr
a
te
m
a
c
hi
n
e
l
e
a
r
ni
ng pote
nt
ia
l
us
in
g M
e
di
c
a
r
e
c
la
im
s
da
ta
, t
he
y f
a
c
e
a
f
unda
m
e
nt
a
l
ba
r
r
ie
r
:
e
xt
r
e
m
e
c
la
s
s
im
ba
la
nc
e
,
w
he
r
e
f
r
a
ud
c
a
s
e
s
c
om
pr
is
e
a
m
e
r
e
0.017%
of
r
e
c
or
ds
.
T
hi
s
ti
lt
f
or
c
e
s
th
e
m
ode
ls
to
w
a
r
ds
th
e
m
a
jo
r
it
y
c
la
s
s
,
yi
e
ld
in
g
hi
gh
f
a
ls
e
ne
ga
ti
ve
s
a
nd
r
e
nde
r
in
g
m
a
ny
s
y
s
te
m
s
ope
r
a
ti
ona
ll
y i
m
pr
a
c
ti
c
a
l.
A
d
u
al
-
m
od
e
l
m
ac
h
in
e
l
e
ar
n
in
g ap
p
r
oac
h
T
hi
s
pa
p
e
r
in
tr
oduc
e
s
a
n
in
nova
ti
ve
dua
l
-
m
ode
l
m
a
c
hi
ne
-
le
a
r
ni
ng
f
r
a
m
e
w
or
k
th
a
t
a
ddr
e
s
s
e
s
th
e
s
e
c
ha
ll
e
nge
s
:
i)
U
ns
upe
r
vi
s
e
d l
e
a
r
ni
ng f
or
nove
l
pa
tt
e
r
n di
s
c
ove
r
y
−
M
ode
ls
:
C
B
L
O
F
a
nd
E
C
O
D
a
lg
or
it
hm
s
.
−
I
nput
:
m
e
di
c
a
r
e
pr
ovi
de
r
ut
il
iz
a
ti
on a
nd pa
ym
e
nt
da
ta
(
37M
+
r
e
c
or
ds
)
[
6]
.
−
R
ol
e
:
c
a
s
t
a
w
id
e
ne
t
to
de
te
c
t
a
nom
a
li
e
s
a
c
r
o
s
s
50+
f
e
a
tu
r
e
s
(
e
.
g., c
ha
r
ge
r
a
ti
os
, s
e
r
vi
c
e
v
e
lo
c
it
y)
.
−
V
a
li
da
ti
on:
P
s
e
udo
-
la
be
ls
f
r
om
t
he
L
E
I
E
.
ii)
S
upe
r
vi
s
e
d l
e
a
r
ni
ng f
or
hi
gh
-
c
onf
ir
m
a
ti
on c
la
s
s
if
ic
a
ti
on
−
M
ode
ls
:
C
4.5
de
c
is
io
n t
r
e
e
s
a
nd l
ogi
s
ti
c
r
e
gr
e
s
s
io
n
.
−
I
nput
:
t
op a
nom
a
li
e
s
f
la
gge
d by uns
upe
r
vi
s
e
d m
ode
l
s
a
nd L
E
I
E
[
7]
.
−
R
ol
e
:
r
e
f
in
e
pr
e
di
c
ti
ons
us
in
g unde
r
s
a
m
pl
e
d, b
a
la
nc
e
d da
t
a
(
80:
20 non
-
f
r
a
ud:
f
r
a
ud)
.
−
O
ut
c
om
e
:
r
e
duc
e
f
a
ls
e
po
s
it
iv
e
s
by 63%
c
om
pa
r
e
d t
o pur
e
uns
u
pe
r
vi
s
e
d m
e
th
ods
.
I
n
s
um
m
a
r
y,
t
he
d
ua
l
m
ode
l
a
pp
r
oa
c
h
pr
e
s
e
nt
e
d
he
r
e
n
ot
on
l
y
de
te
c
ts
m
or
e
f
r
a
ud
ul
e
nt
M
e
d
ic
a
r
e
c
la
i
m
s
b
ut
br
in
gs
ne
w
t
e
c
hn
iq
ue
s
f
o
r
d
yna
m
i
c
t
hr
e
s
ho
ld
in
g
a
nd
ne
tw
or
k
-
ba
s
e
d
f
e
a
t
ur
e
e
n
gi
ne
e
r
in
g.
T
he
s
e
a
r
e
to
o
ve
r
c
o
m
e
th
e
e
x
is
t
in
g
m
e
t
ho
ds
a
nd
t
o
h
a
ve
a
m
o
r
e
a
da
p
ti
ve
a
nd
a
c
c
ur
a
te
to
ol
t
o
pr
ot
e
c
t
pu
bl
ic
m
one
y a
nd
M
e
d
ic
a
r
e
.
2.
L
I
T
E
R
A
T
U
R
E
R
E
V
I
E
W A
N
D
T
H
E
O
R
E
T
I
C
A
L
F
O
U
N
D
A
T
I
O
N
2.1. Gap
s
i
n
e
xi
s
t
in
g r
e
s
e
ar
c
h
T
he
r
e
a
r
e
t
w
o
m
a
in
li
m
i
ta
ti
o
ns
to
m
a
c
hi
n
e
l
e
a
r
ni
ng
'
s
a
pp
li
c
a
ti
on
to
p
r
o
vi
de
r
u
ti
li
z
a
ti
on
a
n
d
pa
ym
e
n
t
da
ta
.
O
ne
is
t
he
“
c
la
s
s
i
m
ba
la
n
c
e
c
r
is
is
”.
F
r
a
ud
ul
e
nt
c
a
s
e
s
m
a
ke
up
j
us
t
0.
01
7%
o
f
M
e
di
c
a
r
e
r
e
c
o
r
ds
.
T
ha
t
m
e
a
ns
tr
a
di
ti
ona
l
m
ode
ls
t
r
a
in
e
d
o
n
t
hi
s
s
k
e
w
e
d
da
t
a
te
nd
to
be
b
ia
s
e
d
to
w
a
r
d
t
he
m
a
jo
r
it
y
c
la
s
s
[
4
]
.
A
s
a
r
e
s
u
lt
,
t
he
y
p
r
od
uc
e
una
c
c
e
pt
a
b
ly
hi
g
h
f
a
ls
e
ne
ga
ti
ve
r
a
te
s
(
ove
r
4
0
%
)
.
T
h
is
is
s
ue
m
a
ke
s
m
a
ny
s
ys
te
m
s
ope
r
a
ti
o
na
l
ly
in
e
f
f
e
c
t
iv
e
:
th
e
y
e
i
th
e
r
f
a
i
l
to
f
l
a
g
ge
n
ui
n
e
f
r
a
u
d
or
o
ve
r
w
he
lm
in
v
e
s
t
ig
a
to
r
s
w
it
h
f
a
ls
e
a
le
r
ts
.
A
not
he
r
li
m
it
a
ti
on
is
th
e
ove
r
r
e
li
a
nc
e
on
la
be
le
d
da
ta
.
S
upe
r
vi
s
e
d
m
a
c
hi
ne
le
a
r
ni
ng
a
ppr
oa
c
h
e
s
de
pe
nd on c
os
tl
y, ha
r
d
-
to
-
c
om
e
-
by da
ta
s
e
ts
w
it
h f
r
a
ud l
a
be
ls
[
8
]
(
M
e
di
c
a
r
e
c
la
im
s
f
r
a
udul
e
nt
pa
ym
e
nt
da
ta
i
s
not
publ
ic
ly
a
va
il
a
bl
e
or
a
c
c
ur
a
te
ly
de
r
iv
a
bl
e
f
r
om
e
xi
s
ti
ng
c
ont
e
nt
m
a
na
ge
m
e
nt
s
ys
te
m
(
C
M
S
)
da
ta
s
e
ts
)
.
U
ns
upe
r
vi
s
e
d
m
e
th
od
s
la
c
k
th
e
to
ol
s
to
va
li
da
te
a
nom
a
li
e
s
a
ga
i
ns
t
r
e
a
l
-
w
or
ld
f
r
a
ud
in
di
c
a
to
r
s
[
2]
.
T
hi
s
p
a
pe
r
a
ddr
e
s
s
e
s
th
e
s
e
ga
p
s
th
r
ough
th
r
e
e
ke
y
in
nova
ti
ons
.
T
he
s
e
in
nova
ti
ons
pa
ve
th
e
w
a
y
f
or
a
de
ta
il
e
d
m
e
th
odol
ogy c
om
bi
ni
ng pr
a
c
ti
c
a
l
a
lg
or
it
hm
s
a
nd da
ta
i
nt
e
gr
a
ti
on, outl
in
e
d ne
xt
.
2.2
. K
e
y i
n
n
ovat
io
n
s
F
ir
s
t
s
te
p
is
to
de
ve
lo
p
a
hybr
id
s
a
m
pl
in
g
s
tr
a
te
gy
[
9
]
to
m
it
i
ga
te
c
la
s
s
im
ba
la
nc
e
.
T
hi
s
a
ppr
oa
c
h
c
om
bi
ne
s
r
a
ndom
unde
r
s
a
m
pl
in
g
(
r
e
ta
in
in
g
100%
of
f
r
a
ud
c
a
s
e
s
w
hi
le
r
e
duc
in
g
non
-
f
r
a
ud
s
a
m
pl
e
s
to
a
n
80:
20
r
a
ti
o)
w
it
h
c
os
t
-
s
e
ns
it
iv
e
le
a
r
ni
ng
[
10]
(
pe
na
li
z
in
g
m
i
s
c
la
s
s
if
ie
d
f
r
a
ud
c
a
s
e
s
f
iv
e
ti
m
e
s
m
or
e
th
a
n
non
-
f
r
a
ud
dur
in
g
tr
a
in
in
g)
.
T
hi
s
m
e
th
od
a
li
gns
w
it
h
th
e
w
e
ig
ht
e
d
lo
s
s
f
r
a
m
e
w
or
k
in
im
ba
la
nc
e
d
le
a
r
ni
ng
by
m
in
im
iz
in
g r
is
k (
R
)
:
=
∑
(
(
,
̂
)
+
(
1
−
)
)
∑
(
(
,
̂
)
)
−
Evaluation Warning : The document was created with Spire.PDF for Python.
C
om
put
S
c
i
I
nf
T
e
c
hnol
I
S
S
N
:
2722
-
3221
A
dual
-
m
ode
l
m
ac
hi
ne
l
e
ar
ni
ng appr
oac
h t
o m
e
di
c
a
r
e
f
r
aud
…
(
J
e
s
u M
ar
c
us
I
m
m
anu
v
e
l
A
r
oc
k
ia
s
a
m
y
)
247
W
he
r
e
w
e
ig
ht
α
=
0.8
pr
io
r
it
iz
e
s
f
r
a
ud
r
e
c
a
ll
,
w
e
c
a
n
r
e
duc
e
th
e
r
is
k
of
f
a
l
s
e
n
e
ga
ti
ve
s
w
it
h
th
is
f
unc
ti
on
(y
i
r
e
pr
e
s
e
nt
s
t
r
ue
l
a
be
l
a
nd ŷ
i
r
e
pr
e
s
e
nt
s
pr
e
di
c
te
d l
a
be
l
in
l
os
s
f
unc
ti
ons
(
L
)
)
.
S
e
c
ond, we
i
nt
e
gr
a
te
M
e
di
c
a
r
e
c
la
im
s
w
it
h t
he
L
E
I
E
us
in
g
na
ti
ona
l
pr
ovi
de
r
i
de
nt
i
f
ie
r
s
(
N
P
I
s
)
. T
hi
s
m
e
r
ge
d
da
ta
s
e
t
c
r
e
a
te
s
a
la
b
e
le
d
be
n
c
hm
a
r
k
f
or
va
li
da
ti
on.
T
hi
s
s
te
p
i
s
a
f
or
m
of
s
e
m
i
-
s
upe
r
vi
s
e
d
le
a
r
ni
ng
w
he
r
e
L
E
I
E
l
a
be
ls
a
c
t
a
s
“
a
nc
hor
s
”
to
gui
de
uns
upe
r
vi
s
e
d
a
no
m
a
ly
de
te
c
ti
on.
T
hi
r
d,
w
e
u
s
e
p
a
r
a
ll
e
li
z
e
d
ba
t
c
h
pr
oc
e
s
s
in
g
a
c
r
os
s
G
P
U
c
lu
s
te
r
s
to
e
na
bl
e
r
e
a
l
-
ti
m
e
a
na
ly
s
is
of
2.1
m
il
li
on
c
la
im
s
pe
r
hour
.
T
hi
s
ba
tc
h
pr
oc
e
s
s
in
g
a
ppl
ie
s
M
a
p
R
e
duc
e
pr
in
c
ip
le
s
[
11]
to
di
s
tr
ib
ut
e
a
nom
a
ly
s
c
or
in
g t
a
s
ks
. I
t
r
e
duc
e
s
r
unt
im
e
c
om
pl
e
xi
ty
f
r
om
O
(
n²)
t
o O
(
n
lo
g n)
.
2.3
.
Wh
y t
h
is
ap
p
r
oac
h
m
at
t
e
r
s
A
dua
l
-
m
ode
l
a
r
c
hi
te
c
tu
r
e
c
a
n
a
c
hi
e
v
e
out
c
om
e
s
th
a
t
a
s
in
gl
e
-
m
ode
l
a
r
c
hi
te
c
tu
r
e
c
a
nnot
.
T
hi
s
hybr
id
f
r
a
m
e
w
or
k
br
id
ge
s
th
e
ga
p
be
twe
e
n
e
xpl
or
a
to
r
y
da
ta
a
na
ly
s
is
a
nd
a
c
ti
ona
bl
e
in
te
ll
ig
e
nc
e
.
I
t
a
ddr
e
s
s
e
s
a
c
or
e
c
ha
ll
e
nge
in
f
r
a
ud
de
te
c
ti
on:
th
e
te
ns
io
n
be
twe
e
n
di
s
c
ove
r
in
g
ne
w
f
r
a
ud
pa
tt
e
r
ns
a
nd
m
in
im
iz
in
g
in
ve
s
ti
ga
ti
ve
ove
r
he
a
d.
−
U
ns
upe
r
vi
s
e
d c
om
pone
nt
s
de
te
c
t
e
m
e
r
gi
ng f
r
a
ud pa
tt
e
r
ns
(
e
.g., C
O
V
I
D
-
19 bil
li
ng s
pi
ke
s
)
.
−
S
upe
r
vi
s
e
d
m
ode
ls
va
li
da
te
f
in
di
ngs
w
it
h
88.3%
a
r
e
a
unde
r
th
e
c
ur
ve
(
AUC
)
a
c
c
ur
a
c
y,
pr
io
r
it
iz
in
g
c
a
s
e
s
f
or
f
ur
th
e
r
a
udi
ts
2.4
.
T
h
e
or
e
t
ic
al
c
on
t
r
ib
u
t
io
n
s
O
ur
t
he
or
e
ti
c
a
l
c
ont
r
ib
ut
io
ns
i
nc
lu
de
:
−
A
f
r
a
ud
s
ig
na
tu
r
e
hypothe
s
i
s
[
6]
s
how
in
g
e
ngi
ne
e
r
e
d
f
e
a
tu
r
e
s
li
ke
c
ha
r
ge
r
a
ti
o
a
nd
s
e
r
vi
c
e
ve
lo
c
it
y
e
nc
ode
s
uni
ve
r
s
a
l
f
r
a
ud pa
tt
e
r
ns
i
nva
r
ia
nt
t
o pr
ovi
de
r
s
pe
c
ia
lt
y.
−
A
nom
a
ly
-
a
w
a
r
e
s
upe
r
vi
s
e
d
le
a
r
ni
ng
[
12]
in
t
r
oduc
e
s
a
pa
r
a
di
gm
w
he
r
e
uns
upe
r
vi
s
e
d
a
nom
a
ly
s
c
or
e
s
e
nha
nc
e
s
up
e
r
vi
s
e
d f
e
a
tu
r
e
s
p
a
c
e
s
, i
m
pr
ovi
ng mode
l
c
a
li
br
a
ti
o
n.
−
T
hi
s
w
or
k
a
dva
nc
e
s
th
e
th
e
or
e
ti
c
a
l
und
e
r
pi
nni
ngs
of
he
a
lt
h
c
a
r
e
f
r
a
ud
de
te
c
ti
on
w
hi
le
pr
ovi
di
ng
a
s
c
a
la
bl
e
bl
ue
pr
in
t
f
or
r
e
a
l
-
w
or
ld
de
pl
oym
e
nt
.
T
he
s
e
th
e
or
e
ti
c
a
l
a
dva
nc
e
m
e
nt
s
s
e
t
th
e
s
ta
ge
f
or
a
pr
a
c
ti
c
a
l
m
e
th
odol
ogy, de
ta
il
e
d ne
xt
, t
ha
t
c
om
bi
ne
s
r
obu
s
t
a
lg
o
r
it
hm
s
w
it
h r
e
a
l
-
w
or
ld
da
ta
i
nt
e
gr
a
ti
on.
3.
M
E
T
H
O
D
I
n
th
is
s
e
c
ti
on,
w
e
bui
ld
on
th
e
in
it
ia
l
w
hi
te
pa
pe
r
s
e
c
ti
ons
t
o
e
xpl
or
e
th
e
in
te
gr
a
ti
on
of
da
ta
s
e
ts
,
m
e
th
odol
ogi
c
a
l
f
r
a
m
e
w
or
k,
th
e
hybr
id
m
ode
l'
s
s
ta
ge
s
,
a
nd
t
he
or
e
ti
c
a
l
c
ont
r
ib
ut
io
ns
in
de
ta
il
,
e
ns
ur
in
g
a
th
or
ough unde
r
s
ta
ndi
ng f
or
r
e
s
e
a
r
c
he
r
s
a
nd pr
a
c
ti
ti
one
r
s
i
n he
a
l
th
c
a
r
e
f
r
a
ud de
te
c
ti
on.
3.1. Dat
a s
ou
r
c
e
s
an
d
i
n
t
e
gr
at
io
n
A
s
m
e
nt
io
ne
d
in
pr
e
vi
ous
s
e
c
ti
on
s
,
th
e
f
ounda
ti
on
of
th
is
s
tu
d
y
li
e
s
in
two
im
por
ta
nt
da
ta
s
e
ts
.
E
a
c
h
one
s
e
r
ve
s
a
di
s
ti
nc
t
ye
t
in
te
r
c
onne
c
te
d
r
ol
e
in
a
ddr
e
s
s
in
g
th
e
dua
l
c
ha
ll
e
nge
s
of
s
c
a
la
bi
li
ty
a
nd
va
li
da
ti
on
in
M
e
di
c
a
r
e
f
r
a
ud
de
te
c
ti
on.
M
e
di
c
a
r
e
pr
ovi
de
r
ut
il
iz
a
ti
on
a
nd
p
a
ym
e
nt
da
ta
:
th
is
d
a
ta
s
e
t
c
ove
r
s
2019
to
2022
a
nd
in
c
lu
de
s
ove
r
37
m
il
li
on
r
e
c
or
ds
f
r
om
a
bout
1.2
m
il
li
on
he
a
lt
hc
a
r
e
pr
ovi
de
r
s
.
I
t
of
f
e
r
s
a
gr
a
nul
a
r
vi
e
w
of
bi
ll
in
g be
ha
vi
or
s
a
nd s
e
r
vi
c
e
ut
il
iz
a
ti
on, with key va
r
ia
bl
e
s
s
uc
h a
s
:
−
P
a
ym
e
nt
m
e
tr
ic
s
-
c
ont
a
in
s
to
ta
l
m
e
di
c
a
r
e
pa
ym
e
nt
a
m
ount
s
,
a
ll
ow
e
d
a
m
ount
s
,
a
nd
s
ta
nd
a
r
di
z
e
d
c
ha
r
ge
s
(
a
dj
us
te
d f
or
ge
ogr
a
phi
c
pr
ic
in
g va
r
ia
ti
ons
)
.
−
S
e
r
vi
c
e
pa
tt
e
r
ns
-
pr
ovi
de
s
in
s
ig
ht
s
in
to
tr
e
nds
th
r
ough
th
e
v
ol
um
e
of
s
e
r
vi
c
e
s
of
f
e
r
e
d,
s
tr
a
ti
f
ie
d
by
be
ne
f
ic
ia
r
y de
m
ogr
a
phi
c
s
l
ik
e
a
ge
a
nd g
e
nde
r
.
−
P
r
ovi
de
r
s
p
e
c
ia
lt
i
e
s
:
c
a
t
e
gor
ic
a
l
c
la
s
s
if
ic
a
ti
on
s
s
uc
h
a
s
c
a
r
di
ol
o
g
y,
de
r
m
a
to
l
ogy
,
e
n
a
b
le
s
a
na
l
y
s
i
s
by
f
ie
ld
.
L
E
I
E
:
t
hi
s
da
ta
s
e
t,
m
a
in
ta
in
e
d
by
th
e
U
.S
.
D
e
pa
r
tm
e
nt
of
H
e
a
lt
h
a
nd
H
um
a
n
S
e
r
vi
c
e
s
,
li
s
ts
pr
ovi
de
r
s
ba
r
r
e
d
f
r
om
M
e
di
c
a
r
e
pa
r
ti
c
ip
a
ti
on
due
to
f
r
a
udul
e
nt
a
c
ti
vi
ti
e
s
.
W
e
m
a
tc
h
N
P
I
s
f
r
om
th
e
m
e
di
c
a
r
e
da
ta
s
e
t
w
it
h
th
os
e
in
L
E
I
E
.
P
r
ovi
de
r
s
w
it
h
m
a
tc
hi
ng
N
P
I
s
a
r
e
l
a
be
le
d
a
s
f
r
a
udul
e
nt
,
c
r
e
a
ti
ng
a
gol
d
-
s
ta
nda
r
d
va
li
da
ti
on
s
e
t.
T
hi
s
li
nka
g
e
gi
ve
s
us
a
f
r
a
ud
pr
e
va
le
nc
e
of
a
bo
ut
0.017%
-
or
a
1:
2
,
000
c
la
s
s
r
a
ti
o
(
f
r
a
ud
t
o
non
-
f
r
a
ud)
. T
hi
s
m
e
r
ge
d da
ta
s
e
t
is
c
r
it
ic
a
l
f
or
s
upe
r
vi
s
e
d l
e
a
r
ni
ng va
li
da
ti
on.
I
nt
e
gr
a
ti
ng
th
e
s
e
da
ta
s
e
ts
br
id
ge
s
th
e
ga
p
be
tw
e
e
n
unl
a
be
l
e
d
c
la
im
s
da
ta
a
nd
la
be
le
d
f
r
a
ud
c
a
s
e
s
.
T
hi
s
d
a
ta
s
e
t
m
e
r
gi
ng
pr
oc
e
s
s
a
ddr
e
s
s
e
s
th
e
s
c
a
r
c
it
y
of
la
b
e
le
d
da
ta
in
f
r
a
ud
de
te
c
ti
on.
H
ow
e
ve
r
,
th
e
e
xa
c
t
num
be
r
of
f
r
a
ud
c
a
s
e
s
(
a
r
ound
1,850
in
s
upe
r
vi
s
e
d
c
la
s
s
if
ic
a
ti
on)
s
ugge
s
ts
th
e
s
upe
r
vi
s
e
d
s
ta
ge
us
e
s
a
s
ub
s
e
t
of
t
op a
nom
a
li
e
s
-
not
t
he
e
nt
ir
e
da
ta
s
e
t.
T
hi
s
s
ub
s
e
t
s
tr
a
te
gy a
li
g
ns
w
it
h t
he
hybr
id
a
ppr
oa
c
h'
s
de
s
ig
n.
3.2. Hyb
r
id
f
r
am
e
w
o
r
k
f
or
r
ob
u
s
t
d
e
t
e
c
t
io
n
O
ur
pr
opos
e
d
m
e
th
odol
ogy
us
e
s
a
two
-
s
ta
g
e
hybr
id
f
r
a
m
e
w
or
k.
U
ns
upe
r
vi
s
e
d
a
nom
a
ly
de
te
c
ti
on
i
s
c
om
bi
ne
d
w
it
h
s
up
e
r
vi
s
e
d
c
la
s
s
if
ic
a
ti
on
to
ba
la
n
c
e
s
e
ns
it
iv
it
y
(
de
te
c
ti
ng
a
ll
pot
e
nt
i
a
l
f
r
a
ud)
a
nd
pr
e
c
i
s
io
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2722
-
3221
C
om
put
S
c
i
I
nf
T
e
c
hnol
,
V
ol
.
6
, N
o.
3
,
N
ove
m
be
r
20
25
:
245
-
252
248
(
m
in
im
iz
in
g
f
a
ls
e
pos
it
iv
e
s
)
.
T
hi
s
a
ppr
oa
c
h
is
pa
r
ti
c
ul
a
r
ly
w
e
ll
-
s
ui
te
d
in
th
is
M
e
di
c
a
r
e
c
la
im
s
pa
ym
e
nt
c
ont
e
xt
, w
he
r
e
f
r
a
ud pa
tt
e
r
ns
e
vol
ve
a
nd l
a
b
e
le
d da
ta
i
s
i
n
a
de
q
ua
te
.
S
ta
ge
1:
U
ns
upe
r
vi
s
e
d
a
nom
a
ly
d
e
te
c
ti
on
T
he
f
ir
s
t
s
ta
ge
f
oc
us
e
s
on
id
e
nt
if
yi
ng
br
oa
d
f
r
a
ud
pa
tt
e
r
ns
w
it
hout
r
e
ly
in
g
on
la
be
le
d
da
ta
by
le
ve
r
a
gi
ng
two
a
lg
or
it
hm
s
,
E
C
O
D
[
13]
e
s
ti
m
a
te
s
th
e
unde
r
ly
in
g
di
s
tr
ib
ut
io
n
o
f
e
a
c
h
f
e
a
tu
r
e
us
in
g
e
m
pi
r
ic
a
l
c
um
ul
a
ti
ve
di
s
tr
ib
ut
io
n
f
unc
ti
ons
(
e
C
D
F
)
.
A
nom
a
li
e
s
a
r
e
id
e
nt
if
ie
d
a
s
ob
s
e
r
va
ti
ons
in
th
e
ta
il
s
of
th
e
s
e
di
s
tr
ib
ut
io
ns
. T
he
a
nom
a
ly
s
c
or
e
i
s
c
om
put
e
d a
s
:
(
)
=
∑
[
(
(
)
∙
l
o
g
(
(
)
)
+
(
1
−
(
)
)
]
∙
(
1
−
(
)
)
]
=
1
w
he
r
e
F
i
is
t
he
e
C
D
F
f
or
t
he
(
i)
-
th
f
e
a
tu
r
e
(
in
ot
he
r
w
or
ds
, F
i
tr
a
c
ks
e
a
c
h f
e
a
tu
r
e
’
s
di
s
tr
ib
ut
io
n)
, E
C
O
D
e
xc
e
ls
a
t
de
te
c
ti
ng
gl
oba
l
out
li
e
r
s
,
s
uc
h
a
s
s
y
s
te
m
ic
ove
r
bi
ll
in
g
a
c
r
os
s
a
ll
s
pe
c
ia
lt
ie
s
,
but
m
a
y
m
is
s
lo
c
a
l
a
nom
a
li
e
s
w
it
hi
n s
pe
c
if
ic
c
lu
s
te
r
.
C
B
L
O
F
[
14
]
:
t
h
is
a
lg
o
r
it
h
m
f
i
r
s
t
c
lu
s
te
r
s
p
r
o
vi
d
e
r
s
by
s
pe
c
ia
lt
y
us
in
g
k
-
m
e
a
ns
c
l
us
te
r
in
g
,
w
it
h
k
=
150
c
h
os
e
n
ba
s
e
d
on
do
m
a
i
n
kn
ow
le
d
ge
o
r
c
lu
s
te
r
in
g
a
na
ly
s
is
t
o
r
e
f
le
c
t
t
he
d
iv
e
r
s
it
y
o
f
m
e
di
c
a
l
f
ie
ld
s
.
I
t
th
e
n
c
o
m
p
ut
e
s
o
ut
li
e
r
s
c
o
r
e
s
b
a
s
e
d
o
n
th
e
di
s
ta
nc
e
t
o
t
he
c
l
us
te
r
c
e
n
t
r
oi
d
a
n
d
th
e
c
lu
s
te
r
s
iz
e
us
in
g
th
e
f
or
m
u
la
:
(
)
=
(
)
×
(
,
(
)
)
F
or
a
pr
ovi
de
r
(
p)
in
th
e
c
lu
s
te
r
(
C
)
.
C
B
L
O
F
is
pa
r
ti
c
ul
a
r
ly
e
f
f
e
c
ti
ve
a
t
de
t
e
c
ti
ng
s
p
e
c
ia
lt
y
-
s
pe
c
if
ic
a
nom
a
li
e
s
,
s
uc
h
a
s
a
be
r
r
a
nt
c
a
r
di
ol
ogy
c
ha
r
ge
s
,
but
it
s
pe
r
f
or
m
a
nc
e
de
pe
nds
on
th
e
qua
li
ty
of
c
lu
s
te
r
de
f
in
it
io
ns
.
T
o
le
ve
r
a
ge
bot
h
gl
oba
l
a
nd
lo
c
a
l
pe
r
s
pe
c
ti
ve
s
,
a
nom
a
ly
s
c
or
e
s
f
r
om
E
C
O
D
a
nd
C
B
L
O
F
a
r
e
c
om
bi
ne
d
us
in
g
a
w
e
ig
ht
e
d
a
ve
r
a
ge
of
60%
to
C
B
L
O
F
a
nd
40%
to
E
C
O
D
.
T
hi
s
w
e
ig
ht
in
g
pr
io
r
it
iz
e
s
s
pe
c
ia
lt
y
-
s
pe
c
if
ic
pa
tt
e
r
ns
w
hi
le
r
e
ta
in
in
g
s
e
n
s
it
iv
it
y
to
s
ys
te
m
ic
out
li
e
r
s
,
r
e
f
le
c
ti
ng
a
s
tr
a
te
gi
c
ba
la
nc
e
ba
s
e
d
on
pr
e
li
m
in
a
r
y
a
na
ly
s
is
or
e
xpe
r
t
ju
dgm
e
nt
.
T
o
il
lu
s
tr
a
te
th
e
w
or
kf
lo
w
of
th
is
uns
upe
r
vi
s
e
d
s
ta
ge
,
F
ig
ur
e
1
de
pi
c
ts
how
E
C
O
D
a
nd
C
B
L
O
F
c
om
bi
ne
to
de
te
c
t
gl
oba
l
a
nd
s
pe
c
ia
lt
y
s
pe
c
if
ic
a
nom
a
li
e
s
gui
di
ng
th
e
s
ubs
e
que
nt
s
upe
r
vi
s
e
d c
l
a
s
s
if
ic
a
ti
on
s
.
F
ig
ur
e
1.
U
ns
upe
r
vi
s
e
d
m
ode
ls
a
nom
a
ly
d
e
te
c
ti
on
-
E
C
O
D
a
nd
C
B
L
O
F
i
n
s
ta
ge
1
3.3
.
F
e
at
u
r
e
e
n
gi
n
e
e
r
in
g
T
o
boos
t
th
e
de
te
c
ti
on
c
a
pa
bi
li
ti
e
s
,
w
e
ha
ve
c
r
e
a
te
d
s
e
ve
r
a
l
d
om
a
in
-
s
pe
c
if
ic
f
e
a
tu
r
e
s
th
a
t
di
r
e
c
tl
y
ta
r
ge
t
known f
r
a
ud i
ndi
c
a
to
r
s
i
n he
a
lt
hc
a
r
e
bi
ll
in
g
:
−
C
ha
r
ge
r
a
ti
o
[
15]
hi
ghl
ig
ht
s
pot
e
nt
ia
l
ove
r
bi
ll
in
g, w
he
r
e
pr
ovi
d
e
r
s
c
ha
r
ge
m
or
e
t
ha
n t
he
r
e
a
s
on
a
bl
e
c
o
s
t.
ℎ
=
T
o
t
a
l
p
a
y
m
e
n
t
s
A
l
l
o
w
e
d
a
m
o
u
n
t
−
S
e
r
vi
c
e
v
e
lo
c
it
y
[
3]
m
e
a
s
ur
e
s
th
e
in
te
ns
it
y
of
s
e
r
vi
c
e
pr
ovi
s
i
on
pe
r
be
ne
f
ic
ia
r
y,
f
la
ggi
ng
e
xc
e
s
s
iv
e
or
unne
c
e
s
s
a
r
y t
r
e
a
tm
e
nt
s
(
s
e
r
vi
c
e
v
e
l
oc
it
y m
e
a
s
ur
e
s
t
he
r
a
te
of
s
e
r
vi
c
e
s
pe
r
be
ne
f
ic
ia
r
y)
.
Evaluation Warning : The document was created with Spire.PDF for Python.
C
om
put
S
c
i
I
nf
T
e
c
hnol
I
S
S
N
:
2722
-
3221
A
dual
-
m
ode
l
m
ac
hi
ne
l
e
ar
ni
ng appr
oac
h t
o m
e
di
c
a
r
e
f
r
aud
…
(
J
e
s
u M
ar
c
us
I
m
m
anu
v
e
l
A
r
oc
k
ia
s
a
m
y
)
249
F
e
a
tu
r
e
s
l
ik
e
‘
s
e
r
vi
c
e
ve
lo
c
it
y’
a
nd ‘
c
ha
r
ge
r
a
ti
o’
he
lp
i
de
nt
if
y
uni
ve
r
s
a
l
f
r
a
ud pa
tt
e
r
ns
, f
or
e
xa
m
pl
e
,
if
t
he
c
ha
r
ge
r
a
ti
o i
s
gr
e
a
te
r
t
ha
n 1, it
m
a
y i
ndi
c
a
te
ove
r
c
ha
r
gi
n
g i
s
s
ue
s
.
=
S
e
r
v
i
c
e
s
r
e
n
d
e
r
e
d
T
o
t
a
l
n
o
.
o
f
b
e
n
e
f
i
c
i
a
r
i
e
s
S
ta
ge
2:
S
upe
r
vi
s
e
d
c
la
s
s
if
ic
a
ti
on
T
he
s
e
c
ond
s
ta
ge
r
e
f
in
e
s
th
e
a
nom
a
li
e
s
de
te
c
t
e
d
in
s
ta
ge
1
in
to
hi
gh
-
c
onf
id
e
nc
e
f
r
a
ud
pr
e
di
c
ti
ons
,
a
ddr
e
s
s
in
g
th
e
s
e
ve
r
e
c
la
s
s
im
ba
la
nc
e
(
0.017%
f
r
a
ud
pr
e
v
a
le
nc
e
in
th
e
or
ig
in
a
l
da
ta
s
e
t)
.
T
he
pr
oc
e
s
s
in
vol
ve
s
,
c
la
s
s
i
m
ba
la
nc
e
m
it
ig
a
ti
on:
r
a
ndom
unde
r
s
a
m
pl
in
g i
s
e
m
pl
oye
d, r
e
ta
in
in
g a
ll
i
de
nt
if
ie
d
f
r
a
ud c
a
s
e
s
(
N
=
1,850)
a
nd
r
e
duc
in
g
non
-
f
r
a
ud
c
a
s
e
s
to
a
c
hi
e
ve
a
n
80:
20
non
-
f
r
a
ud:
f
r
a
ud
r
a
ti
o.
T
hi
s
m
e
a
ns
s
e
le
c
ti
ng
7,400
non
-
f
r
a
ud
c
a
s
e
s
(
s
in
c
e
80:
20
im
pl
ie
s
f
our
non
-
f
r
a
uds
f
or
e
ve
r
y
one
f
r
a
ud,
a
nd
4
×
1,850=
7,400)
,
pr
e
s
e
r
vi
ng
c
r
it
ic
a
l
m
in
or
it
y
-
c
la
s
s
in
f
or
m
a
ti
on
w
it
hout
in
t
r
odu
c
in
g
s
ynt
he
ti
c
da
ta
noi
s
e
f
r
om
ove
r
s
a
m
pl
in
g
te
c
hni
que
s
l
ik
e
th
e
s
ynt
he
ti
c
m
in
or
it
y ove
r
s
a
m
pl
in
g t
e
c
hni
que
(
S
M
O
T
E
)
[
16]
.
3.4.
S
u
p
e
r
vi
s
e
d
a
lg
or
it
h
m
s
C
4.5
de
c
is
io
n t
r
e
e
[
17]
:
t
h
is
a
lg
or
it
hm
c
ons
tr
uc
ts
i
nt
e
r
pr
e
ta
bl
e
de
c
is
io
n t
r
e
e
s
us
in
g i
nf
or
m
a
ti
on ga
in
,
w
it
h s
pl
it
s
c
hos
e
n t
o m
a
xi
m
iz
e
,
(
,
)
=
(
)
−
∑
|
|
|
|
∈
(
)
∙
ℎ
(
)
(
S
r
e
pr
e
s
e
nt
s
e
nt
ir
e
pr
ovi
de
r
d
a
ta
s
e
t,
w
hi
le
A
in
di
c
a
te
s
a
n a
tt
r
ib
ut
e
of
th
e
da
ta
s
e
t
th
a
t
is
be
in
g
e
v
a
lu
a
te
d,
S
v
is
pa
r
ti
ti
one
d
da
ta
ba
s
e
d
on
th
e
a
tt
r
ib
ut
e
A
)
.
I
ts
s
tr
e
ngt
h
li
e
s
in
hu
m
a
n
-
r
e
a
da
bl
e
r
ul
e
s
,
id
e
a
l
f
or
a
udi
ti
ng,
th
ough
it
m
a
y ove
r
f
it
r
a
r
e
f
r
a
ud pa
tt
e
r
ns
.
(
)
=
1
1
+
−
(
0
+
1
1
+
∙
∙
∙
+
)
L
ogi
s
ti
c
r
e
gr
e
s
s
io
n
[
18]
:
e
s
ti
m
a
te
s
f
r
a
ud
pr
oba
bi
li
ty
vi
a
th
e
lo
gi
s
ti
c
f
unc
ti
on
,
I
t
o
ff
e
r
s
c
a
li
br
a
te
d
pr
oba
bi
li
ti
e
s
f
or
r
is
k
pr
io
r
it
iz
a
ti
on,
th
ough
li
m
it
e
d
by
li
ne
a
r
de
c
is
io
n
bounda
r
ie
s
th
a
t
m
a
y
m
is
s
c
om
pl
e
x
in
te
r
a
c
ti
ons
.
(
β
0
, β
1
, ..., β
n
r
e
pr
e
s
e
nt
s
w
e
ig
ht
e
d c
oe
f
f
ic
ie
nt
s
w
hi
l
e
x
1
, x
2,
…x
n
r
e
pr
e
s
e
nt
s
f
e
a
tu
r
e
va
lu
e
s
)
.
F
e
a
tu
r
e
s
pa
c
e
e
nr
ic
hm
e
nt
:
uns
upe
r
vi
s
e
d
a
nom
a
ly
s
c
or
e
s
(
f
r
om
E
C
O
D
a
nd C
B
L
O
F
)
a
r
e
i
nc
or
po
r
a
te
d
a
s
f
e
a
tu
r
e
s
,
a
ll
ow
in
g
s
upe
r
vi
s
e
d
m
ode
ls
to
le
a
r
n
w
hi
c
h
a
nom
a
l
ie
s
a
li
gn
w
it
h
known
f
r
a
ud
la
be
l
s
f
r
om
L
E
I
E
,
e
nha
nc
in
g pr
e
di
c
ti
ve
pow
e
r
.
T
hi
s
s
ta
ge
i
s
c
r
uc
ia
l,
a
s
i
t
va
li
da
t
e
s
a
nom
a
li
e
s
w
it
h hi
gh pr
e
c
is
io
n, r
e
duc
in
g f
a
ls
e
pos
it
iv
e
s
by 63%
c
om
pa
r
e
d t
o pur
e
uns
upe
r
vi
s
e
d m
e
th
ods
, a
s
n
ot
e
d i
n t
he
i
nt
r
oduc
ti
on.
4.
C
O
M
P
A
R
A
T
I
V
E
M
O
D
E
L
E
V
A
L
U
A
T
I
O
N
4.1
.
M
od
e
l
s
t
r
e
n
gt
h
s
an
d
op
e
r
at
io
n
al
c
on
t
e
xt
s
T
a
bl
e
1
s
um
m
a
r
iz
e
s
th
e
s
tr
e
ngt
hs
,
li
m
it
a
ti
ons
,
a
nd
ope
r
a
ti
ons
c
ont
e
xt
s
of
e
a
c
h
m
ode
l
in
our
dua
l
f
r
a
m
e
w
or
k,
hi
ghl
ig
ht
in
g
th
e
ir
c
om
pl
e
m
e
nt
a
r
y
r
ol
e
s
in
f
r
a
ud
de
te
c
ti
on.
T
h
e
c
om
pl
e
m
e
nt
a
r
y
na
tu
r
e
of
th
e
hybr
id
a
ppr
oa
c
he
s
b
e
c
om
e
s
e
vi
de
nt
w
h
e
n
e
x
a
m
in
in
g
e
a
c
h
m
o
de
l’
s
pe
r
f
or
m
a
nc
e
c
ha
r
a
c
te
r
is
ti
c
s
.
H
e
r
e
is
th
e
a
na
ly
s
is
of
s
tr
e
ngt
h
s
,
w
e
a
kn
e
s
s
e
s
,
a
nd
ope
r
a
ti
ona
l
c
ont
e
xt
s
.
E
a
c
h
c
om
pone
nt
s
e
r
ve
s
a
di
s
ti
nc
t
r
ol
e
w
it
hi
n
our
f
r
a
m
e
w
or
k.
T
he
un
s
upe
r
vi
s
e
d
m
ode
ls
(
E
C
O
D
a
nd
C
B
L
O
F
)
c
a
s
t
a
w
id
e
de
te
c
ti
on
n
e
t,
w
hi
le
th
e
s
upe
r
vi
s
e
d
a
lg
or
it
hm
s
(
C
4.5
a
nd
lo
gi
s
ti
c
r
e
gr
e
s
s
io
n)
r
e
f
in
e
a
nom
a
li
e
s
in
t
o
a
c
ti
ona
bl
e
,
hi
gh
-
c
onf
id
e
nc
e
pr
e
di
c
ti
ons
th
a
t
in
ve
s
ti
ga
to
r
s
c
a
n a
c
tu
a
ll
y us
e
.
T
a
bl
e
1.
C
om
pa
r
a
ti
ve
a
na
ly
s
is
of
m
ode
l
pe
r
f
or
m
a
nc
e
a
nd u
s
e
c
a
s
e
s
M
ode
l
K
e
y s
t
r
e
ngt
hs
L
i
m
i
t
a
t
i
ons
O
pe
r
a
t
i
ona
l
c
ont
e
xt
E
C
O
D
D
e
t
e
c
t
s
gl
oba
l
out
l
i
e
r
s
a
c
r
os
s
s
pe
c
i
a
l
t
i
e
s
;
R
obus
t
t
o di
m
e
ns
i
ona
l
i
t
y
L
e
s
s
s
e
n
s
i
t
i
ve
t
o
l
oc
a
l
/
s
pe
c
i
a
l
t
y
-
s
pe
c
i
f
i
c
a
nom
a
l
i
e
s
I
ni
t
i
a
l
s
c
r
e
e
ni
ng
f
or
s
ys
t
e
m
i
c
f
r
a
ud
pa
t
t
e
r
ns
C
B
L
O
F
C
a
pt
ur
e
s
s
pe
c
i
a
l
t
y
-
s
pe
c
i
f
i
c
a
nom
a
l
i
e
s
;
A
da
pt
s
t
o pr
ovi
de
r
popul
a
t
i
on c
l
us
t
e
r
s
P
e
r
f
or
m
a
nc
e
de
pe
nds
on
c
l
us
t
e
r
qua
l
i
t
y;
R
e
qui
r
e
s
dom
a
i
n
know
l
e
dge
f
or
k
-
s
e
l
e
c
t
i
on
T
a
r
ge
t
e
d s
pe
c
i
a
l
t
y
-
s
pe
c
i
f
i
c
a
udi
t
i
ng
C
4.5
D
e
c
i
s
i
on t
r
e
e
P
r
oduc
e
s
hum
a
n
-
r
e
a
da
bl
e
de
c
i
s
i
on
r
ul
e
s
;
C
a
pt
ur
e
s
non
-
l
i
ne
a
r
r
e
l
a
t
i
ons
hi
ps
P
r
one
t
o
ove
r
f
i
t
t
i
ng
on
r
a
r
e
f
r
a
ud
pa
t
t
e
r
ns
;
B
r
a
nc
h
c
om
pl
e
xi
t
y
i
nc
r
e
a
s
e
s
w
i
t
h da
t
a
s
i
z
e
A
udi
t
c
a
s
e
e
xpl
a
na
t
i
on
a
nd
r
e
gul
a
t
or
y doc
um
e
nt
a
t
i
on
L
ogi
s
t
i
c
r
e
gr
e
s
s
i
on
O
ut
put
s
c
a
l
i
br
a
t
e
d
pr
oba
bi
l
i
t
y
s
c
or
e
s
;
C
om
put
a
t
i
ona
l
l
y e
f
f
i
c
i
e
nt
L
i
m
i
t
e
d
by
l
i
ne
a
r
de
c
i
s
i
on
bounda
r
i
e
s
;
L
e
s
s
e
f
f
e
c
t
i
ve
f
or
c
om
pl
e
x pa
t
t
e
r
n de
t
e
c
t
i
on
R
i
s
k
-
ba
s
e
d
c
a
s
e
pr
i
or
i
t
i
z
a
t
i
on
a
nd
r
e
s
our
c
e
a
l
l
oc
a
t
i
on
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2722
-
3221
C
om
put
S
c
i
I
nf
T
e
c
hnol
,
V
ol
.
6
, N
o.
3
,
N
ove
m
be
r
20
25
:
245
-
252
250
4.2. Ad
d
r
e
s
s
in
g c
la
s
s
i
m
b
al
an
c
e
:
e
m
p
ir
ic
al
val
id
at
io
n
A
s
di
s
c
us
s
e
d
in
pr
e
vi
ous
s
e
c
ti
ons
,
ha
ndl
in
g
c
la
s
s
im
ba
la
nc
e
is
a
c
ha
ll
e
nge
a
s
th
e
da
ta
s
e
t
c
ont
a
in
s
onl
y 0.017%
of
t
ot
a
l
c
la
im
s
. T
o e
va
lu
a
te
t
he
i
m
pa
c
t
of
c
la
s
s
i
m
ba
la
nc
e
m
it
ig
a
ti
on s
tr
a
te
gi
e
s
, T
a
bl
e
2 pr
e
s
e
nt
s
th
e
pe
r
f
or
m
a
nc
e
of
th
e
C
4.5
de
c
is
io
n
tr
e
e
a
c
r
os
s
di
f
f
e
r
e
nt
no
n
-
f
r
a
ud
-
to
-
f
r
a
ud
r
a
ti
os
.
F
ig
ur
e
2
c
om
pl
im
e
nt
s
th
is
a
na
ly
s
is
by vi
s
ua
ll
y c
om
pa
r
in
g t
he
A
U
C
pe
r
f
or
m
a
nc
e
a
c
r
o t
he
t
e
s
te
d r
a
ti
os
, hi
ghl
ig
ht
in
g t
he
80:
20 r
a
ti
o’
s
opt
im
a
l
ba
la
nc
e
.
W
e
te
s
te
d
f
our
c
la
s
s
di
s
tr
ib
ut
io
ns
(
non
-
f
r
a
ud
-
to
-
f
r
a
ud
r
a
ti
os
)
e
m
pi
r
ic
a
ll
y,
a
s
s
how
n
in
T
a
bl
e
2
.
T
he
80:
20 r
a
ti
o r
e
duc
e
d f
a
ls
e
ne
ga
ti
ve
s
by 33%
c
om
pa
r
e
d t
o r
a
w
i
m
ba
la
nc
e
d da
ta
w
hi
le
m
a
in
ta
in
in
g
c
om
put
a
ti
ona
l
e
f
f
ic
ie
nc
y.
T
a
bl
e
2.
C
la
s
s
r
a
ti
o i
m
pa
c
t
s
on C
4.5 pe
r
f
or
m
a
nc
e
R
a
t
i
o
A
U
C
(
C
4.5)
F
a
l
s
e
ne
ga
t
i
ve
r
a
t
e
K
e
y
i
ns
i
ght
50:
50
0.872
0.301
T
oo m
a
ny f
r
a
ud c
a
s
e
s
s
l
i
ppe
d t
hr
ough due
t
o ove
r
f
i
t
t
i
ng t
he
m
i
nor
i
t
y c
l
a
s
s
80:
20
0.883
0.275
I
t
c
a
ught
33%
m
or
e
f
r
a
ud t
ha
n r
a
w
da
t
a
w
hi
l
e
ke
e
pi
ng f
a
l
s
e
pos
i
t
i
ve
s
m
a
na
ge
a
bl
e
.
90:
10
0.851
0.412
T
oo m
a
ny m
i
s
s
e
d f
r
a
ud c
a
s
e
s
, r
i
s
ki
ng
ope
r
a
t
i
ona
l
f
a
i
l
ur
e
.
F
ig
ur
e
1
. V
a
li
da
ti
on r
e
s
ul
ts
-
80:
10 vs
90:
10 pe
r
f
or
m
a
nc
e
c
om
pa
r
is
on
4.3. Valid
at
io
n
an
d
p
e
r
f
or
m
an
c
e
i
n
s
ig
h
t
s
T
he
dua
l
-
m
ode
l
f
r
a
m
e
w
or
k
a
c
hi
e
v
e
d
a
n
A
U
C
of
88.3%
,
s
ur
pa
s
s
in
g
s
in
gl
e
-
m
ode
l
a
ppr
oa
c
he
s
by
24%
,
a
s
be
nc
hm
a
r
ke
d
a
ga
in
s
t
g
e
ne
r
a
l
m
a
c
hi
ne
le
a
r
ni
ng
pe
r
f
or
m
a
nc
e
m
e
tr
ic
s
[
19]
,
[
20]
.
C
om
pa
r
e
d
to
pr
io
r
M
e
di
c
a
r
e
f
r
a
ud
de
te
c
ti
on
s
tu
di
e
s
,
our
a
ppr
oa
c
h
s
ig
ni
f
ic
a
nt
ly
out
pe
r
f
or
m
s
e
xi
s
ti
ng
m
e
th
ods
.
F
or
in
s
ta
nc
e
,
B
r
e
nna
n
[
2]
r
e
por
te
d
A
U
C
s
r
a
ngi
ng
f
r
om
0.75
to
0.82
f
o
r
s
up
e
r
vi
s
e
d
m
ode
ls
on
im
ba
la
nc
e
d
M
e
di
c
a
r
e
da
ta
,
li
m
it
e
d
by
hi
gh
f
a
ls
e
ne
ga
ti
ve
r
a
t
e
s
(
ove
r
40%
)
.
G
r
e
s
oi
e
t
al
.
[
4]
a
c
hi
e
ve
d
a
n
A
U
C
of
0.79
us
in
g
uns
upe
r
vi
s
e
d
m
e
th
od
s
but
la
c
ke
d
la
be
l
e
d
va
li
da
ti
on,
le
a
di
ng
to
hi
ghe
r
f
a
ls
e
pos
it
iv
e
s
.
O
ur
hybr
id
f
r
a
m
e
w
or
k,
in
te
gr
a
ti
ng
C
B
L
O
F
a
nd
E
C
O
D
w
it
h
C
4.5
a
nd
lo
gi
s
ti
c
r
e
gr
e
s
s
io
n
,
r
e
duc
e
s
f
a
ls
e
pos
it
iv
e
s
by
63%
c
om
pa
r
e
d
to
s
ta
nda
lo
ne
uns
upe
r
vi
s
e
d
m
e
th
ods
,
a
s
va
li
da
te
d
a
ga
in
s
t
L
E
I
E
la
be
ls
.
T
hi
s
im
pr
ove
m
e
nt
s
te
m
s
f
r
om
th
e
s
yne
r
gy
of
uns
upe
r
vi
s
e
d
a
nom
a
ly
de
te
c
ti
on,
w
hi
c
h
id
e
nt
if
ie
s
nove
l
pa
tt
e
r
ns
,
a
nd
s
upe
r
vi
s
e
d
c
la
s
s
if
ic
a
ti
on,
w
hi
c
h
r
e
f
in
e
s
pr
e
di
c
ti
ons
f
or
a
c
ti
ona
bl
e
a
udi
ts
.
T
he
f
r
a
m
e
w
or
k’
s
a
bi
li
ty
to
pr
oc
e
s
s
2.1
m
il
li
on
c
la
im
s
pe
r
hour
us
in
g
pa
r
a
ll
e
li
z
e
d
G
P
U
c
lu
s
te
r
s
f
ur
th
e
r
e
nha
nc
e
s
it
s
pr
a
c
ti
c
a
l
va
lu
e
,
e
na
bl
in
g
r
e
a
l
-
ti
m
e
f
r
a
ud
de
te
c
ti
on
w
it
hout
ove
r
w
he
lm
in
g
in
ve
s
ti
ga
ti
ve
r
e
s
our
c
e
s
.
T
he
s
e
r
e
s
u
lt
s
unde
r
s
c
or
e
th
e
m
ode
l’
s
s
c
a
la
bi
li
ty
a
nd
pr
e
c
is
io
n, of
f
e
r
in
g a
r
obus
t
to
ol
f
or
s
a
f
e
gua
r
di
ng M
e
di
c
a
r
e
f
unds
.
4.4
.
F
u
t
u
r
e
d
ir
e
c
t
io
n
s
T
hi
s
du
a
l
-
m
ode
l
a
ppr
oa
c
h
ope
n
s
s
e
ve
r
a
l
p
a
th
s
f
or
im
pr
o
ve
m
e
nt
.
W
e
c
oul
d
e
xp
a
nd
f
e
a
tu
r
e
e
ngi
ne
e
r
in
g
by
ta
ppi
ng
ne
twor
k
a
na
ly
s
is
[
21]
-
pr
ovi
de
r
-
be
ne
f
i
c
ia
r
y
c
onne
c
ti
ons
or
r
e
f
e
r
r
a
l
pa
tt
e
r
ns
-
to
c
a
tc
h
c
oor
di
na
te
d
f
r
a
ud
s
c
he
m
e
s
li
ke
ki
c
kba
c
ks
.
W
e
c
oul
d
a
l
s
o
te
s
t
a
da
pt
iv
e
th
r
e
s
hol
di
ng
[
22]
(
e
.g.,
a
dj
us
ti
ng
a
nom
a
ly
c
ut
of
f
s
ba
s
e
d
on
r
e
a
l
-
ti
m
e
f
r
a
ud
t
r
e
nds
)
to
ke
e
p
th
e
m
ode
l
ni
m
bl
e
a
s
s
c
he
m
e
s
e
vol
ve
.
F
ut
ur
e
w
or
k
c
oul
d a
ls
o i
nt
e
gr
a
te
f
r
a
ud de
te
c
ti
on w
it
h pa
ti
e
nt
e
nga
ge
m
e
nt
a
na
ly
ti
c
s
[
23]
, [
24]
or
c
hr
oni
c
di
s
e
a
s
e
pr
e
di
c
ti
on
[
25]
to
c
r
e
a
te
a
hol
is
ti
c
he
a
lt
hc
a
r
e
pr
ot
e
c
ti
on s
y
s
te
m
.
Evaluation Warning : The document was created with Spire.PDF for Python.
C
om
put
S
c
i
I
nf
T
e
c
hnol
I
S
S
N
:
2722
-
3221
A
dual
-
m
ode
l
m
ac
hi
ne
l
e
ar
ni
ng appr
oac
h t
o m
e
di
c
a
r
e
f
r
aud
…
(
J
e
s
u M
ar
c
us
I
m
m
anu
v
e
l
A
r
oc
k
ia
s
a
m
y
)
251
5.
C
O
N
C
L
U
S
I
O
N
M
e
di
c
a
r
e
f
r
a
ud
dr
a
in
s
bi
ll
io
ns
a
nnua
ll
y,
th
r
e
a
te
ni
ng
c
a
r
e
f
or
m
il
li
ons
.
O
ur
dua
l
-
m
ode
l
f
r
a
m
e
w
or
k
-
m
e
ld
in
g
uns
upe
r
vi
s
e
d
a
nom
a
ly
de
te
c
ti
on
w
it
h
s
upe
r
vi
s
e
d
c
la
s
s
if
ic
a
ti
on
-
of
f
e
r
s
a
f
r
e
s
h,
pr
a
c
ti
c
a
l
f
ix
.
B
y
pa
ir
in
g E
C
O
D
a
nd C
B
L
O
F
t
o s
pot
ne
w
pa
tt
e
r
ns
w
it
h C
4.5 a
nd
lo
gi
s
ti
c
r
e
gr
e
s
s
io
n
to
r
e
f
in
e
t
he
m
, w
e
'
ve
hi
t
a
n
A
U
C
of
88.3%
, s
la
s
he
d f
a
ls
e
pos
it
iv
e
s
by 63%
, a
nd pr
oc
e
s
s
e
d 2.1 mi
ll
io
n c
la
im
s
hour
ly
.
F
e
a
tu
r
e
s
l
ik
e
c
ha
r
ge
r
a
ti
o,
s
e
r
vi
c
e
v
e
lo
c
it
y
,
a
nd
L
E
I
E
va
li
da
ti
on
m
a
ke
it
bot
h
s
ha
r
p
a
nd
s
c
a
la
bl
e
.
W
hi
le
tr
a
di
ti
ona
l
m
e
th
od
s
f
a
lt
e
r
a
ga
in
s
t
e
vol
vi
ng
f
r
a
ud
a
nd
s
c
a
r
c
e
la
b
e
ls
,
th
is
a
ppr
oa
c
h
a
da
pt
s
a
nd
de
li
ve
r
s
.
I
t'
s
a
s
te
p
to
w
a
r
d
s
a
f
e
gu
a
r
di
ng
publ
ic
f
unds
a
nd
e
ns
ur
in
g
M
e
di
c
a
r
e
s
e
r
ve
s
th
os
e
w
ho
ne
e
d
it
m
os
t,
w
it
h
r
oom
to
gr
ow
in
to
a
n
e
ve
n
m
or
e
pot
e
nt
t
ool
.
F
U
N
D
I
N
G
I
N
F
O
R
M
A
T
I
O
N
A
ut
hor
s
s
ta
te
no f
undi
ng i
nvol
ve
d.
A
U
T
H
O
R
C
O
N
T
R
I
B
U
T
I
O
N
S
S
T
A
T
E
M
E
N
T
T
hi
s
jo
ur
na
l
us
e
s
th
e
C
ont
r
ib
ut
or
R
ol
e
s
T
a
xonomy
(
C
R
e
d
iT
)
to
r
e
c
ogni
z
e
in
di
vi
dua
l
a
ut
hor
c
ont
r
ib
ut
io
ns
, r
e
duc
e
a
ut
hor
s
hi
p di
s
put
e
s
,
a
nd f
a
c
il
it
a
te
c
ol
la
bo
r
a
ti
on.
N
am
e
o
f
A
u
t
h
or
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
J
e
s
u
M
a
r
c
us
I
m
m
a
nuve
l
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
G
ow
r
is
ha
nka
r
B
hoopa
th
i
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
C
:
C
onc
e
pt
ua
l
i
z
a
t
i
on
M
:
M
e
t
hodol
ogy
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
da
t
i
on
Fo
:
Fo
r
m
a
l
a
na
l
ys
i
s
I
:
I
nve
s
t
i
ga
t
i
on
R
:
R
e
s
our
c
e
s
D
:
D
a
t
a
C
ur
a
t
i
on
O
:
W
r
i
t
i
ng
-
O
r
i
gi
na
l
D
r
a
f
t
E
:
W
r
i
t
i
ng
-
R
e
vi
e
w
&
E
di
t
i
ng
Vi
:
Vi
s
ua
l
i
z
a
t
i
on
Su
:
Su
pe
r
vi
s
i
on
P
:
P
r
oj
e
c
t
a
dm
i
ni
s
t
r
a
t
i
on
Fu
:
Fu
ndi
ng a
c
qui
s
i
t
i
on
C
O
N
F
L
I
C
T
O
F
I
N
T
E
R
E
S
T
A
ut
hor
s
s
ta
te
no c
onf
li
c
t
of
i
nt
e
r
e
s
t.
D
A
T
A
A
V
A
I
L
A
B
I
L
I
T
Y
T
he
da
ta
th
a
t
s
uppor
t
th
e
f
in
di
ngs
of
th
is
s
tu
dy
a
r
e
ope
nl
y
a
va
il
a
bl
e
in
G
it
H
ub
r
e
pos
it
or
y
ht
tp
s
:/
/g
it
hub.c
om
/J
e
s
uM
a
r
c
u
s
I
/Dua
l
-
M
ode
l
-
M
a
c
hi
n
e
-
L
e
a
r
ni
ng
-
A
ppr
oa
c
h
-
to
-
M
e
di
c
a
r
e
-
F
r
a
ud
-
D
e
te
c
ti
on
R
E
F
E
R
E
N
C
E
S
[
1]
C
e
nt
e
r
s
f
or
M
e
di
c
a
r
e
&
M
e
di
c
a
i
d
S
e
r
vi
c
e
s
(
C
M
S
)
,
“
F
i
s
c
a
l
ye
a
r
2024
i
m
pr
ope
r
pa
ym
e
nt
s
f
a
c
t
s
he
e
t
,”
c
m
s
.gov
.
[
O
nl
i
ne
]
.
A
va
i
l
a
bl
e
:
ht
t
ps
:
/
/
w
w
w
.c
m
s
.gov/
ne
w
s
r
oom
/
f
a
c
t
-
s
he
e
t
s
/
f
i
s
c
a
l
-
ye
a
r
-
2024
-
i
m
pr
ope
r
-
pa
ym
e
nt
s
-
f
a
c
t
-
s
he
e
t
[
2]
P
.
B
r
e
nna
n,
“
A
c
om
pr
e
he
ns
i
ve
s
ur
ve
y
of
m
e
t
hods
f
or
ove
r
c
om
i
ng
t
he
c
l
a
s
s
i
m
ba
l
a
nc
e
pr
obl
e
m
i
n
f
r
a
ud
de
t
e
c
t
i
on,”
M
.SC
.
T
he
s
i
s
,
D
e
pa
r
t
m
e
nt
of
C
om
put
i
ng, I
ns
t
i
t
ut
e
of
T
e
c
hnol
ogy B
l
a
nc
ha
r
ds
t
ow
n, D
ubl
i
n, I
r
e
l
a
nd, 2012.
[
3]
R
.
J
.
B
ol
t
on
a
nd
D
.
J
.
H
a
nd,
“
S
t
a
t
i
s
t
i
c
a
l
f
r
a
ud
de
t
e
c
t
i
on:
a
r
e
vi
e
w
,”
St
at
i
s
t
i
c
al
Sc
i
e
nc
e
,
vol
.
17,
no.
3,
pp.
235
–
255,
2002
,
doi
:
10.1214/
s
s
/
1042727940.
[
4]
S
.
G
r
e
s
oi
,
G
.
S
t
a
m
a
t
e
s
c
u,
a
nd
I
.
F
ă
gă
r
ă
ș
a
n,
“
A
dva
nc
e
d
m
e
t
hodol
ogy
f
or
f
r
a
ud
de
t
e
c
t
i
on
i
n
e
ne
r
gy
us
i
ng
m
a
c
hi
ne
l
e
a
r
ni
ng
a
l
gor
i
t
hm
s
,”
A
ppl
i
e
d Sc
i
e
nc
e
s
, vol
. 15, no. 6, 2025, doi
:
10.3390/
a
pp15063361.
[
5]
D
e
c
os
i
m
o
A
dvi
s
or
y
S
e
r
vi
c
e
s
,
“
D
e
t
e
c
t
i
ng
f
r
a
ud
u
s
i
ng
da
t
a
m
i
ni
ng
t
e
c
hni
que
s
,”
s
l
i
de
s
ha
r
e
.ne
t
,
2008.
[
O
nl
i
ne
]
.
A
v
a
i
l
a
bl
e
:
ht
t
ps
:
/
/
w
w
w
.s
l
i
de
s
ha
r
e
.ne
t
/
s
l
i
de
s
how
/
de
t
e
c
t
i
ng
-
f
r
a
ud
-
us
i
ng
-
da
t
a
-
m
i
ni
ng
-
t
e
c
hni
que
s
/
8472940
[
6]
C
e
nt
e
r
s
f
or
M
e
di
c
a
r
e
&
M
e
di
c
a
i
d
S
e
r
vi
c
e
s
,
“
M
e
d
i
c
a
r
e
p
r
o
vi
de
r
u
t
i
l
i
z
a
t
i
o
n
a
nd
pa
ym
e
nt
da
t
a
,
”
c
m
s
.g
ov
.
[
O
n
l
i
ne
]
.
A
v
a
i
l
a
b
l
e
:
ht
t
p:
/
/
w
w
w
.c
m
s
.go
v/
R
e
s
e
a
r
c
h
-
S
t
a
t
i
s
t
i
c
s
-
D
a
t
a
-
a
nd
-
S
ys
t
e
m
s
/
S
t
a
t
i
s
t
i
c
s
-
T
r
e
nds
-
a
nd
-
R
e
po
r
t
s
/
M
e
d
i
c
a
r
e
-
P
r
ov
i
de
r
-
C
ha
r
g
e
-
D
a
t
a
/
i
nd
e
x.
ht
m
l
[
7]
U
.S
.
D
e
pa
r
t
m
e
nt
of
H
e
a
l
t
h
a
nd
H
um
a
n
S
e
r
vi
c
e
s
O
f
f
i
c
e
o
f
I
ns
pe
c
t
or
G
e
ne
r
a
l
,
“
O
I
G
upda
t
e
s
t
he
l
i
s
t
of
e
xc
l
ude
d
i
ndi
vi
dua
l
s
a
nd
e
nt
i
t
i
e
s
,”
oi
g.hhs
.gov
. [
O
nl
i
ne
]
. A
va
i
l
a
bl
e
:
ht
t
ps
:
/
/
oi
g.hhs
.gov/
e
xc
l
u
s
i
ons
/
e
xc
l
u
s
i
ons
_l
i
s
t
.a
s
p
[
8]
C
.
E
l
ka
n,
“
T
he
f
ounda
t
i
ons
of
c
o
s
t
-
s
e
n
s
i
t
i
ve
l
e
a
r
ni
ng,”
P
r
oc
e
e
di
ngs
of
t
he
Se
v
e
nt
e
e
nt
h
I
nt
e
r
nat
i
onal
J
oi
nt
C
onf
e
r
e
nc
e
o
n
A
r
t
i
f
i
c
i
al
I
nt
e
l
l
i
ge
nc
e
(
I
J
C
A
I
’
01)
, vol
. 2, pp. 973
–
978, 2001.
[
9]
H
.
H
e
a
nd
E
.
A
.
G
a
r
c
i
a
,
“
L
e
a
r
ni
ng
f
r
om
i
m
ba
l
a
nc
e
d
da
t
a
,”
I
E
E
E
T
r
ans
ac
t
i
ons
on
K
now
l
e
dge
and
D
at
a
E
ngi
ne
e
r
i
ng
,
vol
.
21,
no. 9, pp. 1263
–
1284, S
e
p. 2009, doi
:
10.1109/
T
K
D
E
.2008.239.
[
10]
J
.
B
r
ow
nl
e
e
,
D
at
a
pr
e
par
at
i
on
f
or
m
ac
hi
ne
l
e
a
r
ni
ng:
dat
a
c
l
e
ani
ng,
f
e
at
ur
e
s
e
l
e
c
t
i
on,
and
dat
a
t
r
ans
f
or
m
s
i
n
py
t
hon
,
M
a
c
hi
n
e
L
e
a
r
ni
ng M
a
s
t
e
r
y, 2020.
[
11]
J
.
D
e
a
n
a
nd
S
.
G
he
m
a
w
a
t
,
“
M
a
p
R
e
duc
e
:
s
i
m
pl
i
f
i
e
d
da
t
a
pr
oc
e
s
s
i
ng
on
l
a
r
ge
c
l
us
t
e
r
s
,”
C
om
m
uni
c
at
i
ons
of
t
h
e
A
C
M
,
vol
.
51,
no. 1, pp. 107
–
113, 2008, doi
:
10.1145/
1327452.1327492.
[
12]
V
.
C
ha
ndol
a
,
A
.
B
a
ne
r
j
e
e
,
a
nd
V
.
K
um
a
r
,
“
A
nom
a
l
y
de
t
e
c
t
i
on:
a
s
ur
ve
y,”
A
C
M
C
om
put
i
ng
Sur
v
e
y
s
(
C
SU
R
)
,
vol
.
41,
no.
3,
pp. 1
–
58, 2009, doi
:
10.1145/
1541880.1541882.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2722
-
3221
C
om
put
S
c
i
I
nf
T
e
c
hnol
,
V
ol
.
6
, N
o.
3
,
N
ove
m
be
r
20
25
:
245
-
252
252
[
13]
Z
.
L
i
,
Y
.
Z
ha
o,
X
.
H
u,
N
.
B
ot
t
a
,
C
.
I
one
s
c
u,
a
nd
G
.
H
.
C
he
n,
“
E
C
O
D
:
un
s
upe
r
vi
s
e
d
out
l
i
e
r
de
t
e
c
t
i
on
us
i
ng
e
m
pi
r
i
c
a
l
c
um
ul
a
t
i
v
e
di
s
t
r
i
but
i
on
f
unc
t
i
ons
,”
I
E
E
E
T
r
ans
ac
t
i
ons
on
K
now
l
e
dge
and
D
at
a
E
ngi
ne
e
r
i
ng
,
vol
.
35,
no.
12,
pp.
12181
–
12193,
2023
,
doi
:
10.1109/
T
K
D
E
.2022.3159580.
[
14]
Z
.
H
e
,
X
.
X
u,
a
nd
S
.
D
e
ng,
“
D
i
s
c
ove
r
i
ng
c
l
us
t
e
r
-
ba
s
e
d
l
oc
a
l
out
l
i
e
r
s
,”
P
at
t
e
r
n
R
e
c
ogni
t
i
on
L
e
t
t
e
r
s
,
vol
.
24,
no.
9
–
10,
pp. 1641
–
1650, 2003, doi
:
10.1016/
S
0167
-
8655(
03
)
00003
-
5.
[
15]
K
.
J
.
C
i
os
a
nd
G
.
W
.
M
oor
e
,
“
U
ni
que
ne
s
s
of
m
e
di
c
a
l
da
t
a
m
i
ni
ng,”
A
r
t
i
f
i
c
i
al
I
nt
e
l
l
i
ge
nc
e
i
n
M
e
di
c
i
ne
,
vol
.
26,
no.
1
–
2,
pp.
1
–
24,
2002, doi
:
10.1016/
S
0933
-
3657(
02
)
00049
-
0.
[
16]
N
.
V
.
C
ha
w
l
a
,
K
.
W
.
B
o
w
ye
r
,
L
.
O
.
H
a
l
l
,
a
nd
W
.
P
.
K
e
ge
l
m
e
ye
r
,
“
S
M
O
T
E
:
s
ynt
he
t
i
c
m
i
nor
i
t
y
ove
r
-
s
a
m
pl
i
ng
t
e
c
hni
que
,”
J
our
nal
of
A
r
t
i
f
i
c
i
al
I
nt
e
l
l
i
ge
nc
e
R
e
s
e
ar
c
h
, vol
. 16, pp. 321
–
357, 2002, doi
:
10.
1613/
j
a
i
r
.953.
[
17]
J
. R
. Q
ui
nl
a
n,
C
4.5:
pr
ogr
am
s
f
o
r
m
ac
hi
ne
l
e
a
r
ni
ng
. S
a
n F
r
a
nc
i
s
c
o, C
a
l
i
f
or
ni
a
, U
S
:
M
or
ga
n K
a
uf
m
a
nn P
ubl
i
s
he
r
s
I
nc
., 1993.
[
18]
D
.
W
.
H
.
J
r
.,
S
.
L
e
m
e
s
how
,
a
nd
R
.
X
.
S
t
ur
di
va
nt
,
A
ppl
i
e
d
l
ogi
s
t
i
c
r
e
g
r
e
s
s
i
on
.
H
oboke
n,
N
e
w
J
e
r
s
e
y:
J
ohn
W
i
l
e
y
&
S
ons
,
I
nc
.,
2013.
[
19]
A
.
P
.
B
r
a
dl
e
y,
“
T
he
us
e
of
t
he
a
r
e
a
unde
r
t
he
R
O
C
c
ur
ve
i
n
t
he
e
va
l
ua
t
i
on
of
m
a
c
hi
ne
l
e
a
r
ni
ng
a
l
gor
i
t
hm
s
,”
P
at
t
e
r
n
R
e
c
ogni
t
i
on
,
vol
. 30, no. 7, pp. 1145
–
1159, 1997, doi
:
10.1016/
S
0031
-
3203(
96
)
00142
-
2.
[
20]
T
.
F
a
w
c
e
t
t
,
“
A
n
i
nt
r
oduc
t
i
on
t
o
R
O
C
a
na
l
ys
i
s
,”
P
at
t
e
r
n
R
e
c
ogni
t
i
on
L
e
t
t
e
r
s
,
vol
.
27,
no.
8,
pp.
861
–
874,
2006,
doi
:
10.1016/
j
.pa
t
r
e
c
.2005.10.010.
[
21]
S
.
W
a
s
s
e
r
m
a
n
a
nd
K
.
F
a
us
t
,
Soc
i
al
ne
t
w
o
r
k
anal
y
s
i
s
:
m
e
t
hods
and
appl
i
c
at
i
ons
.
C
a
m
br
i
dge
,
E
ngl
a
nd:
C
a
m
br
i
dge
U
ni
ve
r
s
i
t
y
P
r
e
s
s
, 1994.
[
22]
Y
.
F
r
e
und
a
nd
R
.
E
.
S
c
ha
pi
r
e
,
“
A
de
c
i
s
i
on
-
t
he
or
e
t
i
c
ge
ne
r
a
l
i
z
a
t
i
on
of
on
-
l
i
ne
l
e
a
r
ni
ng
a
nd
a
n
a
ppl
i
c
a
t
i
on
t
o
boos
t
i
ng,”
L
e
c
t
ur
e
N
ot
e
s
i
n C
om
put
e
r
Sc
i
e
nc
e
, vol
. 904, no. 1, pp. 23
–
37, 1995, doi
:
10.1007/
3
-
54
0
-
59119
-
2_166.
[
23]
J
.
M
.
I
.
A
r
oc
ki
a
s
a
m
y,
“
D
i
gi
t
a
l
he
a
l
t
hc
a
r
e
e
vol
ut
i
on:
t
he
pow
e
r
of
D
e
vO
p
s
f
or
be
t
t
e
r
pa
t
i
e
nt
e
nga
ge
m
e
nt
,”
I
nt
e
r
nat
i
onal
J
ou
r
nal
of
I
nt
e
l
l
i
ge
nt
Sy
s
t
e
m
s
and A
ppl
i
c
at
i
ons
i
n E
ngi
ne
e
r
i
ng
, vol
. 12, no. 4, pp. 5192
–
5198, 2024.
[
24]
J
.
M
.
I
.
A
r
oc
ki
a
s
,
“
D
e
vO
ps
-
dr
i
ve
n
r
e
a
l
-
t
i
m
e
he
a
l
t
h
a
na
l
yt
i
c
s
:
a
s
c
a
l
a
bl
e
f
r
a
m
e
w
or
k
f
or
w
e
a
r
a
bl
e
I
oT
da
t
a
,”
I
nt
e
r
nat
i
onal
J
our
nal
F
or
M
ul
t
i
di
s
c
i
pl
i
nar
y
R
e
s
e
ar
c
h
, vol
. 7, no. 1, 2025, doi
:
10.36948/
i
j
f
m
r
.2025.v07i
01.37358.
[
25]
J
.
M
.
I
.
A
r
oc
ki
a
s
a
m
y,
“
P
r
oa
c
t
i
ve
he
a
l
t
hc
a
r
e
a
na
l
yt
i
c
s
:
e
a
r
l
y
de
t
e
c
t
i
on
of
di
a
be
t
e
s
w
i
t
h
S
D
O
H
i
ns
i
ght
s
a
nd
m
a
c
hi
ne
l
e
a
r
ni
ng,”
E
ur
ope
an
J
our
nal
of
C
om
put
e
r
Sc
i
e
nc
e
and
I
nf
or
m
at
i
on
T
e
c
hnol
ogy
,
vol
.
13,
no.
2,
pp.
64
–
74,
2025,
doi
:
10.37745/
e
j
c
s
i
t
.2013/
vol
13n26474.
B
I
O
G
R
A
P
H
I
E
S
O
F
A
U
T
H
O
R
S
Jesu
Marcus
Immanuvel
Arockiasamy
is
a
distinguished
Healthcare
Analytics
and
DevOps
expert
with
over
18
years
of
pioneering
experience
at
a
leading
healthcare
company.
Renowned
for
his
mastery
of
DevOps
principl
es,
he
has
spe
arheaded
transform
ative
initiatives
that
enhance
system
efficie
ncy,
automate
complex
depl
oyments,
and
optimize
CI/CD
pipelines
using
cutting
-
edge
tools
such
as
Jenkins,
Kubernete
s,
Terraform,
and
AWS.
As
a
visionary
leader
and
dedicated
mentor,
Arockiasamy
has
cu
ltivated
a
collaborative
DevOps
culture
that
drives
innovation,
agility,
and
operational
excellen
ce
across
multidisciplinary
teams.
His
prolific
resear
ch
portfolio
includes
high
-
i
mpact
whitepapers
such
as
'
Digital
Healthcare
Evolution:
The
Power
of
DevOps
for
Bette
r
Patient
Engagement,'
'
Proactive
Healthcare
Analytics:
Early
Detection
of
Diabetes
with
SDOH
Insights
and
Machine
Learning,
'
'
Securing
Telehealth
Platforms:
ML
-
Powered
Phishing
Detection
with
DevOps
in
Healthcare
Analytics,'
and
'
DevOps
-
Driven
Real
-
Time
Health
Analytics:
A
Scalable
Framework
for
Wearable
IoT
Data.
'
These
seminal
wo
rks
integrate
advanced
analytics
,
machine
learning,
and
DevOps
to
revoluti
onize
patient
care,
engagement
,
and
security,
earning
recognition
for
their
actionable
insights
and
scalable
frameworks.
Arockiasamy’ s contrib
utions have not
only advanced healthcare technol
ogy but also s
et a new
standard
for
secure,
patient
-
centric
digital
soluti
ons,
influenci
ng
bot
h
indust
ry
practices
and
academic
discours
e.
His
ongoing
efforts
continue
to
shape
the
future
o
f
healthcare
by
bridging
technologic
al
innovation
with
compassiona
te,
equitable
care
delivery.
He
can
be
contacted
a
t
email:
jesumarcu
s@
gmail.com
.
Gowrishankar
Bhoopathi
is
a
skilled
professional
in
Artificial
In
telligence
and
Healthcare
data
analytics
having
more
than
18
years
of
IT
experienc
e
in
a
leading
healthcare
organization.
His
technical
p
roficiency
spans
cloud
-
based
solutions,
AI/ML
frameworks
with
a
strong
foundation
in
designing
and
managing
large
scale
data
ecosystems,
leveraging
advanced anal
ytics,
playing
a key rol
e in dri
ving bu
siness
growth an
d innov
ation.
Bhoopat
hi is
commit
ted
to
channelin
g
his
expertise
into
healthcare
analytics
minim
izing
provider
abrasions
by
developing
AI
driven
solutions
that
reduce
ineffic
iencies
and
enhance
collabora
tion
between
healthcare
providers
and
payers.
His
research
delves
int
o
AI
driven
Healthcare
analytics
addressi
ng key ch
allenges
and
opportu
nities
driving
meaning
ful change
in healt
hcare
and
beyond.
As
a
recognized
expert
in
AI
and
healthcare
analytics,
Bhoopathi
strives
to
contribu
te
impactful
research,
mentor
indust
ry
professio
nals
and
dri
ve
advancement
s
in
the
field.
He can be contacted at email:
shankarbgowri@gmail.com
.
Evaluation Warning : The document was created with Spire.PDF for Python.