Int
ern
at
i
onal
Journ
al of Ele
ctrical
an
d
Co
mput
er
En
gin
eeri
ng
(IJ
E
C
E)
Vo
l.
9
, No
.
5
,
Octo
ber
201
9
, pp.
3786
~379
7
IS
S
N: 20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v9
i
5
.
pp
3786
-
37
97
3786
Journ
al h
om
e
page
:
http:
//
ia
es
core
.c
om/
journa
ls
/i
ndex.
ph
p/IJECE
Fra
m
ework to
p
redict
NP
A/Willfu
l default
s in corp
orate l
oans:
a
big
data
approa
ch
Giri
ja
Attiger
i
,
M
anoh
ara P
ai M M
,
Radh
ika
M
P
ai
Depa
rtment
o
f
I
nform
at
ion
and
Com
m
unic
at
ion Te
chno
log
y
,
M
a
nipa
l
Instit
u
te of
Technol
og
y
,
Manipa
l
Aca
d
e
m
y
of
High
er Ed
uca
t
ion, Indi
a
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Ma
r 6
, 2
01
9
Re
vised
A
pr 9
,
201
9
Accepte
d
Apr
1
8
, 201
9
Grow
th
and
deve
lopment
of
t
he
ec
onom
y
is
depe
ndent
on
the
banki
ng
s
y
stem.
Bad
lo
a
ns
which
are
Non
-
Perform
ing
Asset
s
(NP
A)
are
the
m
ea
sure
for
assess
ing
the
fina
nc
ia
l
health
of
the
b
ank.
I
t
i
s
ver
y
important
to
cont
ro
l
NP
A
as
it
aff
ect
s
the
profitabilit
y
,
and
d
et
er
iora
t
es
the
qua
li
t
y
o
f
assets
of
the
bank.
It
is
observe
d
tha
t
the
r
e
is
a
signifi
ca
nt
ris
e
in
the
num
ber
of
will
ful
def
aulter
s.
Hen
ce
s
y
stemat
ic
i
dent
ifica
ti
on,
a
ware
ness
and
a
ss
ess
m
ent
of
par
amete
rs
is
e
ss
ent
ia
l
for
e
ar
l
y
pr
edi
c
ti
on
o
f
will
ful
def
ault
beha
vior
.
The
m
ai
n
obj
ec
t
ive
of
th
e
pape
r
is
to
ide
nt
if
y
e
xhausti
ve
l
ist
of
par
amete
r
s
essenti
a
l
for
pre
dicting
whether
the
loa
n
will
bec
om
e
NP
A
and
the
r
e
b
y
will
ful
d
efa
u
lt
.
Thi
s
proc
ess
in
cl
udes
unde
rsta
nding
of
ex
isti
n
g
s
y
stem
to
che
ck
NP
As
and
ide
nt
if
y
ing
the
c
ritical
par
amete
rs.
Also
pro
pose
a
fra
m
ework
for
NP
A/W
il
lful
def
aul
t
ide
n
ti
fi
catio
n.
Th
e
fr
amework
class
ifi
es
the
data
comprising
of
struct
ure
d
a
nd
unstruct
ure
d
p
ar
amete
rs
as
NP
A/W
il
lful
def
aul
t
or
not
.
In
or
der
to
sel
ec
t
th
e
best
c
la
ss
ifi
c
at
i
o
n
m
odel
in
the
fra
m
ework
a
n
expe
riment
at
i
on
is
conduc
te
d
on
loa
n
dataset
on
big
data
pla
tform.
Since
the
loa
n
data
is
struct
ure
d,
unstruct
ure
d
co
m
ponent
is
inc
orpora
te
d
b
y
gene
ra
ti
ng
s
y
nt
het
i
c
data.
The
result
s
indi
c
ate
tha
t
ne
ura
l
net
work m
odel g
ive
s be
st
a
cc
ur
acy
and
h
ence con
sidere
d
in the
fr
a
m
ework.
Ke
yw
or
d
s
:
Bi
g
Data
Corporat
e L
oa
n
Finan
ci
al
F
ra
ud
Ma
chine
Le
ar
ni
ng
Non Per
form
ing
Assets
Param
et
erizat
i
on
W
il
lf
ul
def
a
ult
Copyright
©
201
9
Instit
ut
e
o
f Ad
vanc
ed
Engi
n
ee
r
ing
and
S
cienc
e
.
Al
l
rights re
serv
ed
.
Corres
pond
in
g
Aut
h
or
:
Ma
nohar
a
Pai
M M
,
Dep
a
rtm
ent o
f Info
rm
at
ion
a
nd C
omm
un
ic
ation
Tech
nolo
gy
,
Ma
nip
al
Insti
tu
te
o
f
Tech
nolo
gy, Ma
nip
al
A
cadem
y of
H
i
gher
E
ducat
io
n,
Ma
nip
al
, 5
76104,
India.
Em
a
il
:
m
mm
.p
ai
@m
anipal.edu
1.
INTROD
U
CTION
Ba
nk
i
ng
syst
em
with
integra
ti
on
of
ad
ve
nt
te
ch
no
l
og
y
he
lps
to
f
os
te
r
th
e
eco
nom
ic
dev
el
opm
ent.
They
perform
m
ai
nly
two
im
portant
f
un
ct
io
ns
.
O
ne
is
m
ob
il
iz
ing
de
po
si
ts
by
pro
vid
in
g
at
tract
ive
int
erest
rates
to
c
onve
r
t
inert
savi
ngs
into
act
ive
ca
pi
ta
l
and
s
ec
ond
is
distrib
utin
g
these
de
posit
s
thr
ough
loa
ns
t
o
th
e
corp
or
at
es
t
o
gro
w
furthe
r
t
ha
t
directl
y
hel
ps
i
n
ec
onom
i
c
de
velo
pm
ent.
A
vaili
ng
loa
n
has
bec
om
e
an
easy
process
in
I
nd
i
a
with
the
c
re
di
t
and
che
que
s
et
tl
e
m
ents.
Ba
nk
i
ng
as
well
a
s
N
on
-
Ba
nk
i
ng
Fina
n
ce
C
om
pan
ie
s
offer
dif
fer
e
nt
ty
pes
of
l
oans
accor
ding
to
requirem
ents
of
c
orp
or
at
es
[1,
2].
T
he
r
equ
i
rem
ents
can
be
purc
hase
of
in
ven
t
or
y,
paym
ent
of
l
ong
un
paid
bill
s,
bu
il
ding
of
inf
rastr
uctu
re,
purc
ha
se
of
e
qu
i
pm
e
nt,
loa
n
rep
ay
m
ents
an
d
s
o
on
[
3].
Ba
sed
on
the
re
qu
i
rem
ents
loans
a
re
br
oad
ly
cl
assifi
ed
as
Pers
on
al
L
oa
n
,
Credit
Ca
rd
L
oa
n
,
H
om
e
Loan
,
Ve
hicle
Loa
n
,
E
du
cat
io
n
L
oa
n
,
Loa
n
a
gainst
the
I
ns
ura
nce
Schem
es/
FD
/M
utu
al
fun
ds
,
a
nd
B
usi
ness
L
oan
t
o
Corporat
es
.
T
he
re
are
se
veral
bu
si
ness
l
oans
possible
su
c
h
as
Work
i
ng
capit
al
loan
t
o
us
e
i
n
day
to
day
act
ivit
ie
s
,
Re
al
Esta
te
loan
t
o
bu
y
a
pro
pe
rty
for
pro
du
ct
io
n
,
Ve
nture
loa
n
t
o
s
t
art
up
bu
si
ness
,
Line
of
c
red
it
loa
n
for
certai
n
fin
ancial
assist
an
ce
per
i
od
ic
al
ly
,
Eq
uip
m
ent
loan
t
o
assist
buyi
ng
asset
req
u
i
rem
ents
,
Term
loan
t
o
acq
uire
long
te
rm
fixed
asset
s
,
Loa
n
against
pr
operty
fo
r
s
uppo
rtin
g
bu
si
ness
by
prov
i
ding
sec
ur
it
y
to
the
c
orp
orat
es
,
Ca
sh
C
re
dit
facil
it
y
as
ov
er
dr
a
ft
ag
ai
nst
the
secu
rity
of
the
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Framew
or
k
to pre
dict N
PA/W
il
lf
ul d
ef
au
lt
s i
n
c
orpo
r
ate lo
an
s
: a
big d
ata a
pp
r
oac
h
(
Gir
ij
a
Att
igeri
)
3787
stock
by
pled
gi
ng
t
he
c
urre
nt
asset
s
a
nd
L
et
te
r
of
cre
dit
with
wh
ic
h
t
he
ba
nk
gua
ran
te
es
that
the
sel
le
r
will
receive
paym
e
nt on
ce
rtai
n
c
onditi
ons.
I
n
this
pap
e
r,
fo
c
us
is
on
th
e
pr
oc
ess
of
m
a
nag
i
ng
c
orp
or
a
te
loan,
as
the
recovery
of
th
ese
loans
is
te
dio
us
ta
sk
a
nd
it
af
fects
econom
y
of
the
country
hea
vily
.
Healt
hy
bankin
g
syst
e
m
rep
rese
nts
healt
hy
econom
y
of
th
e
natio
n.
How
ever
t
her
e
are
hindra
nces
t
o
achieve
t
he
re
qu
i
red
set
up
f
or
t
he
sam
e.
All
the
corp
or
at
e
loa
ns
bor
rowe
d
do
no
t
e
nd
up
a
s
asset
s
f
or
t
he
ba
nk.
T
her
e
are
two
outc
om
es
of
the
l
oan
s:
On
e
is
perform
ing
asset
s
and
a
no
t
he
r
one
is
N
on
-
P
erfor
m
ing
Asse
ts
(N
P
A).
NPAs
are
the
l
oa
ns
w
hich
ge
ne
rate
the
loss
in
capit
al
of
banks
an
d
a
re
not
easi
ly
r
ecov
e
ra
ble
by
the
banks
.
Thi
s
is
the
m
os
t
t
edio
us
chall
en
ge
f
or
bankin
g
sect
or
as
it
im
pacts
the
perfor
m
ance
by
decli
ning
th
e
prof
it
s.
I
t
has
bec
om
e
m
ajo
r
pr
ob
le
m
f
or
al
l
public
sect
or
a
nd
pri
vate
sect
or
ba
nks.
Acc
ordin
g
to
RB
I
repor
t
2016,
to
ta
l
gr
oss
NPA
am
ou
nt
was
6
la
kh
crores.
By
2017
there
was
inc
rease
of
1
la
kh
cro
re
[7.31
Cr
]
.
Loss
es
are
over
f
our
t
im
es
m
or
e
than
the
prof
it
s
ind
ic
at
in
g NP
A'
s p
owe
r
to
tr
ap
the
eco
nom
y of t
he
c
ountr
y i
n
vicio
us de
bt cycl
e.
Ba
nks
pro
vide
bi
g
loa
ns
to
co
rporat
es
in
orde
r
t
o
ac
hieve
higher
prof
it
s
[4
]
.
Com
pan
ie
s
sta
r
t
beh
a
ving
a
s
a
def
a
ulter
by
showi
ng
losses
in
c
om
pan
y'
s
finan
ce
.
S
om
e
deliberatel
y
don'
t
rep
ay
even
if
they
hav
e
suffici
e
nt
fina
ncial
res
ources
to
pay.
Com
pan
ie
s
pro
po
s
e
to
pay
th
ese
loa
ns
by
ta
king
oth
e
r
loa
ns
from
m
ul
ti
ple
banks
.
U
nr
ec
overa
bl
e
loan
s
put
c
om
pan
y
to
ba
nkruptcy
sta
tus.
Du
e
to
t
he
will
fu
l
def
a
u
lt
be
ha
viour
,
m
any
genuine
com
pan
ie
s
do
no
t
ge
t
econo
m
ic
su
ppor
t
at
the
tim
e
of
need
an
d
m
ay
end
up
i
n
the
f
ai
lure.
Hen
ce
,
su
c
h
c
om
pan
ie
s
m
ay
no
t
be
able
to
pay
existi
ng
loan
s,
an
d
get
def
a
ulter
ta
g.
Wh
e
n
an
in
di
vi
du
al
or
bu
si
ness
ente
r
pr
ise
decli
ne
s
to
fu
lfil
l
paym
ent
co
m
m
itm
ents
with
finan
ci
al
insti
tuti
on
s
eve
n
w
he
n
it
has
su
f
fici
ent
capa
ci
ty
fo
r
rep
ay
m
ent,
su
ch
a
bor
rowe
r
un
it
is
con
side
re
d
as
will
fu
l
def
a
ult.
More
ba
d
loa
ns
get
gen
e
rated
beca
us
e
of
e
xisti
ng
ba
d
loa
ns
.
I
n
order
to
pr
e
ve
nt
the
fail
ing
eco
nom
y
the
ba
nks
deci
de
to
le
nd
m
on
ey
to
save
the
com
pan
ie
s
go
i
ng
ba
nkr
up
t.
T
hey
ta
ke
adv
a
ntage
of
the
sit
uation
le
ading
t
o
incr
ease
d
will
fu
l
def
a
ults, brin
ging t
he
e
conom
y back
t
o
the
sam
e statu
s.
It
has
bee
n
obs
erv
e
d
that
t
here
is
a
risin
g
tre
nd
i
n
NPAs,
es
pecial
ly
in
publ
ic
sect
or
ba
nk
s.
The
re
a
re
sever
al
cau
ses
for
this
and
th
ere
is
a
stron
g
eviden
ce
f
or
def
i
ning
the
re
la
ti
on
betwee
n
fr
au
d
an
d
NPA
[5
]
.
RB
I
data
obta
ined
t
hro
ugh
R
TI
re
quest
ind
i
cat
e
that
8670
loan
fr
a
ud
ca
s
es
am
ou
nting
Rs.
612.6
bill
ion
a
re
recorde
d ov
e
r
la
st five f
ina
nci
al
yea
rs.
Th
ese
f
ra
ud
s a
re r
e
fe
rr
in
g
to case
s w
he
re bo
rrow
e
r
delibe
ratel
y tr
ie
s t
o
deceive t
he ba
nk and
does
no
t rep
ay
t
he
loa
n
am
ou
nt
[
6].
NPA
is
a
m
ajor
issue
face
d
by
al
l
co
m
m
erci
al
banks
as
the
ba
nk
s
are
c
on
siderin
g
l
o
an
a
dv
a
nces
a
s
rev
e
nue
gen
e
ra
ti
ng
asset
.
Qu
a
li
ty
of
this
a
ss
et
nee
d
to
be
t
aken
care
to
im
pr
ov
e
t
he
prof
it
abili
ty
of
fi
nan
ci
al
insti
tuti
on
s,
a
nd u
lt
im
at
e
ly
f
i
nan
ci
al
cli
m
a
te
of
t
he
eco
nom
y as a whole. In th
is re
ga
rd
e
arly
d
et
ect
ion
of
N
P
A
is
a
big
reli
ef
f
or
s
uch
a
m
ajor
pro
blem
face
d
by
al
l
finan
c
ia
l
insti
tuti
on
s.
Wh
il
e
analy
zi
ng
the
sce
na
rio,
it
is
no
t
just
t
he
poor
eco
nom
ic
c
onditi
ons
that
r
esulte
d
i
n
NPA,
but
delibera
te
def
a
ults
have
al
so
res
ulted
in
hu
ge
piled
up
of
N
P
A.
He
nce
,
m
or
e
e
m
ph
asi
s
on
identif
yi
ng
wil
lful
de
fau
lt
is
the
nee
d
of
the
hour.
T
he
obj
e
ct
ive
of
t
his
pap
e
r
i
s
to
bu
il
d
a
da
ta
m
od
el
f
or
e
arly
detect
ion
of
t
he
will
fu
l
def
a
ult.
T
he
proces
s
of
cl
ass
ify
ing
will
fu
l
def
a
ult
involves
co
ns
i
der
i
ng
va
rio
us
cat
egories
of
par
am
et
ers
su
c
h
as
fina
ncia
l,
per
s
onal
,
s
ocia
l
et
c.
The
data
to
ca
pture
these
pa
r
a
m
et
ers
cou
l
d
be
str
uctu
re
d,
un
st
ru
ct
ur
e
d
a
nd
nee
ds
co
ntinuo
us
a
naly
sis.
Als
o,
the
data
nee
d
t
o
be
ca
ptured
from
var
io
us
s
ources
w
hich
are
cal
le
d
as
Heter
og
e
ne
ou
s
sources
.
Hen
c
e,
Bi
g
data
te
chnolo
gy
is
us
ed
to
de
sign
a
data
m
od
el
w
her
e
the
s
iz
e,
var
ie
ty
an
d
com
plexity
of
data
can
be
ha
nd
l
e
d
eff
ect
ively
[7
]
.
Var
i
ou
s
cl
as
sific
at
ion
al
go
rithm
s
are
design
e
d
us
i
ng
bi
g
data
ap
pro
ach
to
eval
uate
the
cl
assifi
cat
ion
da
ta
m
od
el
fo
r p
red
ic
ti
on
of
ear
ly
stages of
NP
As.
The
rest
of
the
pa
per
is
orga
ni
zed
as
f
ollo
ws
.
Sect
io
n
2
dis
cusses
a
bout
t
he
bac
kgrou
nd
of
t
he
work
with
re
sp
ect
to
natio
nal
an
d
inter
na
ti
on
al
scen
ari
o
a
nd
li
te
rart
ur
e
rev
ie
w.
Sect
ion
3
descr
i
bes
par
am
et
erizat
i
on
process
an
d
data
m
od
el
fo
r
np
a/
will
fu
l
defualt
inde
ntific
at
ion
.
S
ect
ion
4
ex
pla
ins
the
fr
am
ewo
r
k
for
NPA/wil
lf
ul
def
a
ult
ide
ntific
at
ion
.
The
va
li
dation
proces
s
of
fr
am
ewo
r
k
a
nd
eval
uation
of
pr
e
dicti
on m
od
el
is ex
plained
in Secti
on
5.
S
eci
on
6 d
rw
as
con
cl
us
io
n.
2.
BACKG
ROU
ND
Ther
e
are
sev
eral
poli
ci
es,
s
chem
es
wh
ic
h
the
la
w
a
ge
nc
ie
s
hav
e
set
up
to
deal
wit
h
the
fall
ing
econom
y
of
th
e
co
un
t
ry.
T
he
exten
sive
li
te
r
at
ur
e
s
urvey
ha
s
bee
n
ca
rr
ie
d
out
f
or
stu
dying
the
NPA
sc
enar
i
o
in
India
as
w
el
l
as
in
var
io
us
ot
her
c
ount
ries
[8
-
10
]
.
Also
,
the
st
udy
has
bee
n
done
to
identify
va
rio
us
par
am
et
ers
w
hich
a
re
us
ef
ul for NP
A
i
de
ntific
at
ion
pr
ocess
.
2.1.
Int
er
natio
na
l
scen
ario
China
rece
ntly
has
250
m
il
l
i
on
do
ll
ar
s
of
bad
de
bt.
The
se
are
m
a
inly
the
loans
t
hat
are
directl
y
relat
ed
to
real
est
at
e,
us
ed
t
o
dev
el
op
t
he
inf
rastr
uctu
re.
The
sta
te
of
t
he
NPA
is
du
e
to
poli
ti
cal
and
so
ci
al
i
m
plica
ti
on
s,
le
gal
im
ped
i
m
e
nts,
bank
ruptc
y
la
ws,
r
eal
est
at
e.
Ital
y
al
so
f
aced
207
bill
ion
do
ll
ar
ba
d
de
bt
du
e
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
9
, N
o.
5
,
Oct
ober
20
19
:
3
7
8
6
-
3
7
9
7
3788
to
real
est
at
e.
Ba
d
de
bts
al
on
g
with
natio
nal
deb
ts
inc
urre
d
huge
fi
nan
c
ia
l
crisi
s
for
the
c
ountry.
Howe
ve
r
as
it
is p
art
of the
Euroz
on
e
, it w
as sav
e
d by
bailout
fun
ds
pro
vi
ded
t
o recapit
al
iz
e b
an
ks
.
Russia
acco
unts
up
to
9.1
6%
NP
A
rati
o
of
the
total
lo
ans.
It
is
beca
us
e
the
co
untr
y
is
m
os
tly
dep
e
ndent
on
oil
and
gas
ex
ports
.
Wh
e
n
gl
ob
al
oil
pri
ces
crash
e
d
do
wn,
it
m
ark
ed
col
la
ps
e
of
R
us
si
a’s
oil
and
ga
s
in
dust
ries.
In
tu
rn,
ba
nks
a
pprove
d
loan
s
to
re
sc
ue
th
e
ec
onomy
.
H
ow
e
ve
r
th
e
sancti
one
d
m
on
ey
nev
e
r
cam
e
ba
ck
le
adin
g
to
the
fina
ncial
cr
isi
s.
Adding
to
it
s
econ
om
ic
sancti
on
s
i
m
posed
by
Am
erica
and
oth
e
r
E
uro
pean co
un
t
ries cau
s
ed
sl
owd
own
i
n
ec
onom
ic
g
r
ow
t
h.
Sp
ai
n
face
d
de
bt/housi
ng
cris
is
of
unpaid
lo
ans
i
n
2008.
But
G
over
nm
ent
pro
vid
e
d
fix
t
o
t
he
iss
ue
qu
ic
kly
with
r
e
m
edial
m
eas
ur
es
.
As
a
res
ult,
the
ba
d
de
bt
decr
ea
sed
fr
om
6.
09%
in
2016
to
5.
7%
in
Ju
ne
20
17.
I
r
el
and
is
facin
g
the
NPA
is
su
e
du
e
to
ec
onom
ic
slow
dow
n.
It
has
s
et
up
Nati
ona
l
Asset
Ma
nag
em
ent
Ag
e
ncy
for
in
so
lve
ncy
ser
vices
to
su
pp
or
t
real
est
at
e
an
d
hous
i
ng
debt
or
s.
T
hese
re
m
edial
m
easur
es
hav
e
droppe
d
c
ount
ry’
s
ba
d
loa
n
rati
o
f
ro
m
27%
in
20
13
t
o
14.2
pe
rcen
t
i
n
20
16.
T
his
tren
d
is
con
t
rib
uting t
o succes
s whic
h ca
n be att
ri
bu
t
ed
to
s
uccessfu
l deb
t
restr
uctu
rin
g program
s.
2.2.
India
n
sce
na
r
io
In
Indian
fina
ncial
sect
or
duri
ng
2017,
t
he
m
os
t
discusse
d
to
pics
are
GS
T
,
dem
onet
iz
at
ion
an
d
NPAs.
NPA
ha
s
le
d
to
al
m
os
t
10
%
of
t
he
loans
im
pacti
ng
ar
ound
9
la
kh
cr
or
e
,
af
fecti
ng
India
n
eco
no
m
y
neg
at
ively
.
RB
I
is
the
I
nd
ia
n
bankin
g
i
ns
ti
tu
te
wh
ic
h
c
oor
di
n
at
es
an
d
regulat
es
the
act
iv
it
ie
s
of
the
ba
nks
i
n
Indian
eco
nom
y.
Ma
j
or c
halle
ng
es f
ace
d by RB
I
are
m
entione
d
as
foll
ows [3
]
:
a.
NPA:
As
e
xpla
ined
ea
rlie
r,
NPA
is
the
i
ndic
at
or
to
i
den
ti
fy
the
sta
tus
of
the
co
rpo
rate
loans
w
hich
a
r
e
no
t
r
eg
ularly
s
et
tl
ed
and
c
re
at
es
finan
ci
al
crisi
s
for
ba
nk
s
as
well
as
f
or
c
om
pan
y.
NPAs
af
fect
the
fina
ncial
growt
h
of
the
bank
and
he
nce
de
cl
ines
the
eco
no
m
ic
con
diti
on
of
the
c
ountry.
I
n
orde
r
to
ta
ckle this,
I
ndia
n
go
vernm
ent h
as ta
ken m
a
ny init
ia
ti
ves.
1)
Tem
po
rar
y reli
ef
of
s
eve
ral th
ou
s
an
d
c
rores
2)
Sp
eci
al
cou
rts
to d
eal
with
com
pan
ie
s h
avi
ng
bad loa
ns
3)
Re
du
ce
d
i
ntere
st rates
4)
Me
rg
e
r of
ba
nks t
o
re
duce
burd
e
n of ba
d
l
oa
ns
b.
Ba
nk
Fr
a
uds
a
nd
Cy
ber
T
hreat
s:
Ba
nk
fr
a
ud
s
a
nd
cy
ber
-
at
ta
cks
on
fi
na
ncial
tran
sact
ion
s
are
il
le
ga
l
m
eans
of
obta
inin
g
the
m
on
e
y
or
asset
s
es
pe
ci
al
ly
fr
om
ban
k.
O
ne
way
to
obta
in
m
on
e
y
fr
om
a
bank
is
to
ta
ke
out
a
lo
an,
w
hic
h
ba
nk
ers
are
m
or
e
th
an
will
ing
t
o
e
ncou
rag
e
if
t
he
y
hav
e g
oo
d
re
aso
n
to
belie
ve
that
the
m
on
ey
will
be
rep
a
id
in
fu
ll
with
interest
.
A
fraudulent
loa
n,
ho
w
eve
r,
is
one
in
wh
ic
h
th
e
borro
wer
is
a
bu
si
ness
e
ntit
y
con
tr
olled
by
a
dishon
e
st
ba
nk
officer
or
a
n
accom
plice
.
The
"b
orr
ow
e
r
"
then
declares
ba
nkr
up
tc
y
or
va
nish
es
a
nd
th
e
m
on
ey
is
gone.
T
he
bor
r
ower
m
ay
even
be
a
non
-
existe
nt
entit
y
and
the
loan
m
erely
an
arti
fice
to
con
c
eal
a
theft
of
a
la
rg
e
su
m
of
m
on
ey
f
ro
m
the
bank.
T
his
can
al
so
be
see
n
as
a co
m
ponen
t
within m
or
tga
ge
f
ra
ud.
To
d
ay
'
s
ro
bb
e
rs
are
doin
g
rob
ber
y
be
hind
the
inter
net
us
ing
ta
r
gete
d
an
d
soph
ist
ic
at
ed
cy
ber
crim
e
tacti
cs.
So
m
e
of
the
e
xam
ple
at
ta
cks
are
phishi
ng,
Ca
r
ba
nak
m
al
war
e,
SQ
L
i
nj
ect
i
on
at
ta
cks,
a
nd
at
ta
cks
on
bank
database,
cre
dit
card
s
an
d
on
on
li
ne
fi
na
ncial
tr
ansacti
on
s
.
IT
te
am
s
at
banks
have
inc
reased
protect
ion
of
c
us
t
om
er
data
an
d
lim
ited
cre
dit
card
f
raud,
bu
t
the
s
ecur
it
y
of
m
os
t
banks'
inter
nal
syst
e
m
s sti
l
l need
to
b
e
im
pr
oved
.
c.
In
c
rease
in
e
xc
ess
li
qu
i
dity
:
The
inc
rease
of
pe
nalty
rate
will
increase
the
interest
rates
an
d
exce
s
s
reserve
owne
d
by
ba
nks.
T
her
e
fore,
t
he
total
li
qu
idit
y
in
eco
nom
y
will
increase
rap
i
dly
without
involvin
g
poli
cy
rate
reducti
on
m
echan
ism
(lo
os
e
m
on
et
ary
poli
cy
),
just
wh
e
n
t
he
li
qu
idit
y
sh
ould
be
restrict
ed.
T
he
reason
be
hind
in
crease
in
ex
c
ess
li
qu
idit
y
in
bank
is
the
econom
ic
con
diti
on
wh
ic
h
is
in
li
qu
idit
y
trap.
Liqu
idit
y
trap
is
a
con
diti
on
wh
e
re
retu
rn
f
ro
m
ban
ki
ng
loa
n
is
too
s
m
al
l
to
cov
e
r
interm
ediat
ion
co
st and
banks
g
et
h
igh
e
r
yi
eld
in
reserves th
an
giv
i
ng
loa
ns. In
thi
s cond
it
i
on, expan
si
ve
m
on
et
ary
po
li
cy
will
on
ly
cau
se
increase
i
n
e
xcess
reserves.
Du
e
t
o
inc
reas
e
in
li
qu
i
dity
,
f
inancial
crisi
s
are
inc
reasin
g
i
n
the
ba
nk
s
wh
ic
h
le
ads
t
o
we
aken
i
ng the
dom
est
ic
cu
rr
en
c
y wit
h
res
pect t
o
inter
natio
nal
currencies
.
2.3.
Li
tera
tu
re
re
view
In
Indian
fina
ncial
sect
or
duri
ng
2017,
t
he
m
os
t
discusse
d
to
pics
are
GS
T
,
dem
onet
iz
at
ion
an
d
NPAs.
NPA
ha
s
le
d
to
al
m
os
t
10
%
of
t
he
loans
im
pacti
ng
ar
ound
9
la
kh
cr
or
e
,
af
fecti
ng
India
n
eco
no
m
y
neg
at
ively
.
RB
I
is
the
I
nd
ia
n
bankin
g
i
ns
ti
tu
te
w
hic
h
c
oor
di
nates
an
d
regulat
es
the
act
iv
it
ie
s
of
the
ba
nks
i
n
Indian
eco
no
m
y.
Charan
a
nd
Bra
r
[
11]
hav
e
prese
nted
a
st
udy
on
stressed
asset
s
in
India.
T
he
y
hav
e
m
entioned
a
bo
ut
identific
at
io
n
of
num
ber
of
facto
rs
that
le
ad
to
this
sit
uation.
T
hey
ha
ve
ide
ntifie
d
broa
d
cat
egories
of
t
he
reas
ons
su
c
h
as
stress
f
or
global
slow
down,
gove
r
nan
c
e
relat
ed
iss
ue
s,
po
li
ti
cal
factor
s
as
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Framew
or
k
to pre
dict N
PA/W
il
lf
ul d
ef
au
lt
s i
n
c
orpo
r
ate lo
an
s
: a
big d
ata a
pp
r
oac
h
(
Gir
ij
a
Att
igeri
)
3789
well
as
m
a
li
nt
entions
an
d
m
i
sco
nduct.
T
he
y
al
so
em
ph
asi
ze
need
f
or
e
xtensiv
e
researc
h
into
t
he
fact
or
s
t
hat
cause
deteri
or
a
ti
ng
asset
qual
it
y i
n
public sec
tor ba
nk
s
.
In
the
stu
dy
on
fr
a
uds
in
the
In
dia
n
Ba
nkin
g
Ind
us
try
by
Chara
n
et
.
al
.,
they
us
ed
inter
view
base
d
appr
oach
to
id
entify
the
reas
on
s
for
fr
a
uds
in
bankin
g
sec
tor
[
5].
They
m
ention
the
m
ai
n
factors
as
la
ck
of
su
pe
r
visio
n
f
r
om
the
m
ana
gem
ent,
la
ck
of
in
centi
ve
m
echan
ism
s
fo
r
em
plo
ye
es,
non
-
c
oope
rati
ve
sta
f
f,
corp
or
at
e
bor
r
ow
e
rs
a
nd
thi
rd
pa
rty
age
nc
ie
s
et
c.
O
ne
ver
y
im
po
rtant
thin
g
note
d
is
abse
nce
of
strong
regulat
or
y
syst
e
m
an
d
a
bs
e
nc
e of to
ols a
nd techn
i
qu
e
s to
det
ect
early
w
ar
ning si
gn
al
s.
Ba
rd
a
n
a
nd
M
ukhrjee
[
7]
dea
l
with
will
f
ul
de
fau
lt
an
d
it
s
i
m
pl
ic
at
ion
s
f
or
prof
i
ta
bili
ty
a
nd
decisi
on
-
m
aking
proces
s
of
t
he
loa
ns
a
t
banks.
They
exam
ine
the
cases
w
her
e
t
he
borro
wer
def
a
ults
will
fu
ll
y
by
unde
r
repor
ti
ng
it
s
c
ash
flo
w.
I
n
t
he
analy
sis
they
m
ention
it
is
necessa
ry
f
or
t
he
regulat
or
t
o
ch
oose
l
ow
e
r
loa
n
capaci
ty
to
avoi
d
NPA
le
vels
at
the
ban
k
du
e
to
will
fu
l
def
a
ult.
Howe
ver
,
i
t
will
exer
t
sink
in
g
pr
e
ssure
on
the
prof
it
le
vel
of
t
he
bank.
He
nc
e
it
will
face
a
trade
-
off
betw
een
gr
eat
er
inc
idence
of
will
fu
l
de
fa
ult
an
d
highe
r
prof
it
of
t
he
ba
nk.
T
hey
al
so
e
m
ph
asi
ze
that
the
rea
son
for
increasi
ng
wil
lful
de
fa
ult
is
weak
m
on
it
or
i
ng
an
d
su
pe
r
visio
n
syst
e
m
,
po
or
ba
nkr
uptc
y
la
ws
in
de
velo
ping
c
ountries
li
ke
I
nd
ia
.
All
these
giv
e
op
portu
ni
ty
fo
r
the bo
rrow
e
r
t
o wil
lfully
d
e
f
ault t
he
l
oan.
The
researc
h
pa
per
s
[
12
-
16
]
s
how
t
hat
the
ri
sk
s
t
he
banks
f
ace
an
d
def
a
ult
be
hav
i
or
we
re
chall
en
ges
even
tw
o
deca
des
be
fore,
howev
e
r
with
the
new
te
ch
no
l
ogie
s
at
hand
ch
al
le
ng
es
ha
ve
beco
m
e
m
or
e
diff
ic
ult
to
ad
dr
es
s.
As
per
t
he
li
te
ratur
e
st
ud
y
a
nd
national
a
nd
inter
national
sc
enar
i
os
,
t
he
ai
m
of
the
w
ork
is
to
def
i
ne
the
sta
ndar
d
process
to
identify
t
he
pa
ram
et
ers
wh
ic
h
ca
n
be
us
e
d
f
or
e
arly
detect
i
on
of
fr
a
ud
be
hav
i
or
and f
ur
t
her hel
pful fo
r
early
i
den
ti
ficat
io
n of NP
As/
W
il
lf
ul
def
a
ult.
The o
bj
ect
ive
s
of the
pro
pose
d
a
ppr
oach to i
den
ti
fy
NPAs/
W
i
ll
f
ul
def
a
ult are as
foll
ows
a.
Unde
rstan
d
the
loan p
r
ocess
b.
Id
e
ntify data
param
et
ers
fo
r
e
arly
iden
ti
ficat
ion o
f wil
lful
de
fau
lt
ers
or
NPAs
c.
Id
e
ntify s
uitable
technolo
gy a
nd d
e
velo
p
m
od
el
and al
gorithm
s f
or w
il
lful
d
e
fau
lt
ide
ntif
ic
at
ion
3.
PARA
METE
RIZ
ATION
P
ROEC
E
SS
F
OR NP
A/WIL
LFUL DE
FU
ALT IN
DE
N
TIFIC
ATIO
N
The
process
of
loa
n
sa
nctio
ning
a
fter
the
re
qu
est
f
or
l
oan
ti
ll
the
co
m
ple
ti
on
of
it
is
sho
wn
i
n
Figure
1.
Th
e
m
ai
n
i
m
po
rtant
blo
c
k
of
the
loan
pro
cess
is
m
on
it
or
i
ng
the
fina
ncial
healt
h
of
the
corp
or
at
e/
cu
st
om
er
in
orde
r
to
underst
an
d
the
f
ra
ud
or
will
fu
l
beh
a
vi
or.
Mo
nito
rin
g
fina
ncial
heal
th
nee
d
var
i
ou
s
pa
ram
et
ers
w
hich
ar
e
cl
os
el
y
ass
oc
ia
te
d
with
th
e
pur
pose.
He
nce
the
re
is
a
requirem
ent
to
de
fine
sta
nd
a
rd
pro
ce
ss
to
i
den
ti
fy
pa
ram
et
ers
wh
ic
h
will
be
us
e
f
ul
to
de
fine
t
he
data
m
od
el
f
or
early
predict
ion
of
fr
a
uds, wil
lful
def
a
ults an
d f
urt
her N
PA
s
.
Fig
ure
1. Loa
n
sancti
on a
nd re
cov
e
ry
proces
s
Param
et
erizat
i
on
pr
ocess
is
the
im
po
rtant
proces
s
of
ide
ntifyi
ng
es
sentia
l
an
d
c
riti
cal
pa
ram
et
ers
for
carryin
g
out
a
par
ti
cula
r
a
naly
ti
cal
ta
sk
and
c
om
ing
ou
t
with
va
luable
ou
tc
ome
.
For
will
fu
l
def
a
ult
identific
at
ion
i
n
the
l
oan
sce
nar
i
o
the
proc
ess
is
def
i
ne
d
and
is
s
how
n
i
n
Fig
ure
2.
T
he
process
sta
rts
with
identify
in
g
s
ources
that
help
to
unde
rstan
d
the
var
i
ous
te
rm
ino
log
ie
s
of
the
l
oan
an
d
causes
of
NPA
a
nd
there
by w
il
lf
ul d
e
fau
lt
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
9
, N
o.
5
,
Oct
ober
20
19
:
3
7
8
6
-
3
7
9
7
3790
The
de
fine
d
pa
ram
et
erization
proce
ss
co
nsi
der
s
the
par
a
m
et
ers
fr
om
diff
e
ren
t
s
ources
su
ch
a
s
RB
I
do
c
um
ent,
li
ter
at
ur
e
,
case
a
naly
sis
repor
ts
,
br
ai
ns
to
rm
ing
sessio
n,
ba
nk
docu
m
ents
and
s
o
on.
These
par
am
et
ers
are
huge
an
d
unstruct
ur
e
d
an
d
he
nce
nee
d
to
be
cl
assifi
ed
into
broa
d
cat
egories
to
further
captu
re
sp
eci
fic
pa
ram
et
ers
for
eac
h
cat
ego
ry
al
on
g
with
the
ra
ng
e
s.
T
he
pro
cess
is
dynam
ic
in
natur
e
,
c
ov
e
rs
par
am
et
ers
rela
te
d
to
fr
a
ud a
nd ide
ntifie
s t
he
ch
a
ng
e
in ra
nges as
per the ca
te
gories.
Figure
2. Para
m
et
erizat
ion
p
r
ocess
Accor
ding
to
RB
I
ci
rcu
la
r
R
BI/201
5
-
16/1
00
[
17
]
a
will
f
ul
def
ault
is
co
ns
ide
red
t
o
oc
cur
in
a
ny
of
the foll
owin
g
f
our
case
s:
a.
Wh
e
n
the
re
is
a
def
a
ult
in
re
paym
ent
ob
li
gations
by
the
bor
rowe
r
unit
to
the
fina
ncial
i
ns
ti
tuti
on
s
e
ve
n
wh
e
n
it
has
th
e
capaci
ty
to
ho
no
r
the
sai
d
obli
gations.
Th
e
re
is
deliberate
intenti
on
of
not
rep
ay
in
g
the
loan.
b.
The
f
unds
are n
ot u
ti
li
zed
for
the
sp
eci
fic
pu
rpose
inte
nded
fo
r
w
hich
fina
nce
wa
s
avail
e
d
but
ha
ve
bee
n
div
e
rted fo
r ot
her p
urp
os
es.
c.
Wh
e
n
t
he
f
un
ds
hav
e
bee
n
t
app
e
d
off
a
nd
no
t
bee
n
util
iz
ed
f
or
the
pur
po
s
e
f
or
w
hic
h
it
was
a
vaile
d.
Fu
rt
her,
no ass
et
s ar
e a
vaila
bl
e whic
h j
us
ti
fy the
us
a
ge of f
unds
.
d.
Asset
bought
by
th
e len
der
s
’ f
unds
ha
ve bee
n
s
old o
ff with
ou
t t
he kno
wledg
e
of t
he
le
nd
er.
Also
in
cases
wh
e
re
a
le
tt
er
of
c
om
fo
rt
or
gu
a
ra
ntees
are
furn
is
he
d
by
gro
up
com
pan
ie
s
of
will
f
ully
def
a
ulti
ng
unit
s,
these
obli
ga
ti
on
s
a
re
not
hono
red
w
hen
t
hey
are
i
nvok
ed
by
t
he
le
nd
er,
the
n
s
uc
h
gro
up
com
pan
ie
s ar
e
al
so
c
on
si
der
e
d
to
b
e
w
il
lf
ul
def
a
ulters.
RB
I
sug
gests
i
n
it
s
doc
um
ent
on
data
sta
ndar
dizat
ion
[
3,
6]
that,
t
her
e
is
a
data
re
quirem
ent
for
pro
per
super
vi
sion.
T
he
data
is
broa
dly
di
vid
e
d
int
o
tw
o
gro
up
s
1)
Data
s
ub
m
it
te
d
by
ba
nks
2)
D
at
a
gen
e
rated
or
c
om
piled
by
the
su
pe
rv
is
or.
Furtherm
or
e,
data
can
al
so
hav
e
oth
e
r
char
act
e
r
ist
ic
s
wh
ic
h
ne
ed
to
be
co
ns
ide
red.
Table
1
sh
ow
s
these
con
si
de
rati
on
s
s
ugge
ste
d
by
RB
I.
Con
si
der
i
ng
th
e
data
sta
nd
ar
dizat
ion
requirem
ent
of
RB
I
and
i
ncr
e
asi
ng
c
oncer
n
of
l
oan
fr
a
uds
com
m
it
te
d,
the
obj
ect
ive
of
t
he
stu
dy
em
ph
asi
zes
to d
e
fine
the
se
t of pa
ram
et
ers
w
hich help
to dete
ct
w
il
lful
de
fau
lt
beh
a
vior
and
bu
il
d a
da
ta
m
od
el
.
In
orde
r
to
unde
rstan
d
the
us
e
fu
l
ness of
t
he
r
equ
i
red
pa
ram
et
ers,
brai
nst
orm
ing
sessio
n was arra
nge
d
with
ba
nk
e
xperts,
com
pan
y
offici
al
s,
loan
su
pe
r
visors
an
d
fina
ncial
bro
ker
s
.
The
disc
us
sio
n
ha
ppen
ed
on
scenari
os
with
resp
ect
to
ba
nks,
com
pan
ie
s
who
are
ta
ki
ng
loans,
an
d
ot
he
r
fina
ncial
scenari
os
.
T
his
session
was
ext
rem
e
l
y
us
ef
ul
to
obt
ai
n
init
ia
l
br
oa
d
set
of
p
ar
a
m
et
ers
to
beg
i
n
the
pr
ocess,
These
are
s
how
n
in
Figure
3.
Af
te
r
ide
ntifyi
ng
init
ia
l
le
vel
of
pa
ram
et
ers
as
pe
r
br
ai
ns
t
or
m
ing
sessi
on,
f
ur
t
her
m
any
par
am
et
ers
are
ide
ntifie
d
by
le
arn
in
g
case
stu
dies,
t
he
li
te
rat
ur
e
s
urvey
and
discuss
i
on
with
the
dom
a
in
e
xp
e
rt.
Ba
se
d
on
al
l
these
inputs
an
d
stu
dies,
t
he
identifie
d
pa
r
a
m
et
ers
as
earl
y
ind
ic
at
ors
ar
e
gro
up
e
d
i
nto
six
gr
oups
as
s
how
n
in Figu
re
4.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Framew
or
k
to pre
dict N
PA/W
il
lf
ul d
ef
au
lt
s i
n
c
orpo
r
ate lo
an
s
: a
big d
ata a
pp
r
oac
h
(
Gir
ij
a
Att
igeri
)
3791
Figure
3. I
niti
al
Brainsto
rm
ing
par
am
et
ers
Figure
4. Br
oa
d
gro
up
s
of ea
r
ly
ind
ic
at
ors
Table
1.
G
uid
e
li
nes
sug
gested
b
y R
BI
for s
uper
visio
n
Data Type
Su
b
m
itted
b
y
Ban
k
Gen
erate
d
/Co
m
p
il
ed
by
Su
p
ervis
o
r
Structu
red
Nu
m
eric
al/
Fin
an
cial
1
)
DSB Retu
rns
(
X
BR
L)
7
)
Stan
d
ard Ann
ex
es as p
art
o
f
on
site in
sp
ectio
n
2
)
Fraud
Retu
rns
8
)
Ass
ess
m
en
t of
key
f
in
an
cial/capit
al
inclu
d
in
g
v
alid
atio
n
/re
-
ass
ess
m
en
t of
RBS risk
k
data f
u
rnis
h
ed
b
y
ban
k
3
)
FID
Retu
rns
9
)Sores f
o
r
ag
g
regatio
n
s o
f
vario
u
s ri
sk
s as p
art
o
f
IRISc
m
o
d
el
4
)
RB
S Risk
Data
(Data
Co
llecto
r
Ap
p
licatio
n
)
1
0
)
The
m
atic
/Sect
o
r/I
n
d
u
stry
/o
th
er
b
an
k
-
wid
e stu
d
ies
5
)
Fin
an
cial Co
n
g
lo
m
erate
Retu
rn (
E
x
cel)
6
)
Ad
ho
c data (
Da
ta Co
llecto
r
Ap
p
lic
atio
n
)
Textu
al
1
1
)
RB
S Co
n
trol g
ap
inf
o
r
m
atio
n
(D
a
ta Co
llecto
r
Ap
p
licatio
n
)
1
3
)
Co
m
m
en
ts/ad
d
itio
n
al inf
o
r
m
atio
n
on
Co
n
trol g
ap
an
d
Co
m
p
lian
ce b
y
SS
M
1
2
)RBS Co
m
p
lian
ce inf
o
r
m
atio
n
(
Da
ta Co
llecto
r
Ap
p
licatio
n
)
1
4
)
Co
m
m
en
ts b
y
Qu
ality
Ass
u
rance
Div
isio
n
Un
stru
ctu
red
1
5
)
An
n
u
al Rep
o
rts
1
9
)
W
o
rkin
g
do
cu
m
e
n
ts f
o
r
su
p
ervis
o
ry
asses
s
m
en
t
1
6
)
Po
licy
Docu
m
en
ts
2
0
)
Su
p
ervis
o
ry
R
ep
o
rts
1
7
)
Bo
ard Minu
tes
2
1
)
BFS R
ep
o
rts
1
8
)
Rep
o
rts of
E
x
ternal Aud
ito
rs
2
2
)
Co
m
m
u
n
icatio
n
s to
Ban
k
s
So
u
rce:
Rep
o
rt
to
t
h
e Co
m
m
itte
e on
Data a
n
d
I
n
f
o
r
m
ati
o
n
M
an
ag
e
m
en
t in
the Res
erve Ban
k
o
f
I
n
d
ia (
2
0
1
4
),
RBI
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
9
, N
o.
5
,
Oct
ober
20
19
:
3
7
8
6
-
3
7
9
7
3792
The
at
tri
bu
te
s
unde
r
eac
h of
t
he gr
oup are
de
fine
d
as:
1.
Fina
ncial
a.
Fina
ncial
leve
rag
e
r
at
ios
:
i.
Lever
a
ge
rati
os
ind
ic
at
e
fixe
d
ex
pe
ns
es
obl
igati
on
s
.
Since
the
fixe
d
ex
pe
ns
es
are
per
i
od
cost,
it
sh
ould
be
rec
overe
d
from
the
per
iod
i
n
w
hich
i
t
is
incur
re
d.
Worse
ning
le
ve
ra
ge
rati
o
in
dicat
es
that
th
e
com
pan
y i
s not
in
the
posit
io
n t
o reco
ver it
s fi
xed
obli
gatio
ns.
1)
Asset
C
ov
e
rage
Ra
ti
o:
It
in
di
cat
es
total
bac
kup
of
asset
s
f
or
each
r
up
e
e
of
loa
n
raise
d.
If
it
is
m
or
e
than 1 t
hen c
om
pan
y ca
n
m
a
nag
e
to re
pay it
s long
te
rm
lo
ans wit
h
e
xisti
ng assets.
2)
Deb
t
E
quit
y
Ra
ti
o:
This
ind
ic
at
es
ou
tsi
der
s
’
co
ntri
buti
on
to
ca
pital
com
par
ed
to
ow
ner
s
’
con
t
rib
ution.
I
deal rati
o i
s
1:1
bu
t
sta
ndar
d i
s f
ixe
d base
d on the
gestat
io
n peri
od and se
ct
or
.
3)
Deb
t
Se
rv
ic
e
Cov
e
rag
e
Ra
ti
o:
This
is
cal
culat
ed
base
d
on
interest
pay
m
ent
and
inter
i
m
per
iod
of
2
intervals.
If t
he
r
at
io is m
or
e,
loan te
rm
sh
ould
be
le
ss,
if
m
or
e
yea
rs
a
re
giv
en
the
n
it
is a
su
s
picion
4)
Deb
t/
EBIT
D
A
Ra
ti
o
:
Deb
t
is
co
m
par
ed
with
Earn
i
ngs
Be
fo
re
I
nter
est
Tax
and
Depreci
at
io
n
Asset(E
TBD
A
),
wh
ic
h
i
nd
ic
a
te
s the
burd
e
n of t
he de
bt on pr
of
it
5)
Fixed
As
set
s
to
Net
Wort
h:
This
i
nd
ic
a
te
s
to
wh
at
e
xt
ent
fi
xed
asse
t
is
fina
nced
by
ow
ner
s
’
con
t
rib
ution.
6)
In
te
re
st
Cov
e
r
age
Ra
ti
o
(I
C
R)
:
In
te
rest
is
com
par
ed
with
Ear
nings
Be
fo
re
I
nter
est
Tax
and
Depreci
at
ion,
wh
ic
h
ind
ic
at
es
the
bu
r
de
n
of
the
intere
st
on
pro
fit,
how
m
any
tim
es
pr
of
it
is
su
f
f
ic
ie
nt to
cover
the i
nterest
7)
Lo
ng
Te
rm
De
bt
to
Ca
pital
izati
on
Ra
ti
o:
It
ind
ic
at
es
that
bor
rowe
d
f
und
in
the
fina
ncial
structu
re
is
le
ss co
m
par
ed
to ow
ner
s
’ fun
d.
It s
hould be
le
ss than 1.
8)
Current
asset
s
current
li
abili
ty
:
This
rati
o
i
nd
ic
at
es
sho
rt
te
rm
l
iqu
idit
y.
It
ind
ic
at
es
the
qu
al
it
y
of
work
i
ng
capit
a
l.
For
a
m
anuf
act
ur
in
g
sect
or
higher
w
orki
ng
ca
pital
is
essenti
al
com
par
ed
to
ser
vice
sect
or
s
9)
Total
Ex
pense
Ra
ti
o
(TER)
(
Total
ex
pen
se/
Turn
ov
e
r)
:
T
his
in
dicat
es
the
pro
portion
of
the
cost
in
rev
e
nue. L
o
we
r
rati
o i
s
bette
r i
nd
ic
at
or
of
pr
of
it
m
arg
in.
ii.
In
te
re
st
to
sal
es
rati
o:
This
in
dicat
es
the
pro
portio
n
of
de
bt
cost
to
the
revenu
e
ea
r
ned.
L
ow
e
r
rati
o
is
a
bette
r
in
dicat
io
n of p
rofit
abili
ty
a.
Credit
rati
ng
a
gen
cy
:
Cre
dit
rati
ng
s
are
give
n
wh
e
ne
ver
t
her
e
is
ne
w
iss
ue
of
se
cu
riti
es
apa
rt
f
ro
m
the co
m
pan
y a
s a wh
ole. E
xa
m
ple
CR
IS
IL
s
cor
e
etc.
b.
Wr
it
e
offs: I
t i
nd
ic
at
es
poor c
ollec
ti
on
poli
cy
an
d ham
per
t
he pr
of
it
abili
ty
.
c.
Current
li
abili
ty
to
fi
xed
asse
ts
:
Highe
r
rati
o
in
dicat
es
higher
ris
k
[S
ho
rt
te
rm
fu
nd
for
long
te
rm
pro
j
ect
s]
2.
Op
e
rati
onal
: T
hese att
rib
utes
ind
ic
at
e
op
e
rati
on
al
as
pects
of a c
om
pan
y.
a.
Delay
in pay
m
ents to
sup
plier
s
b.
Delay
in pay
m
ents fr
om
the cu
stom
ers
c.
Losing c
us
t
ome
rs
d.
Sudd
e
n
c
ha
ng
e
s in
t
he
s
uppliers,
buye
rs
e.
Fr
e
qu
e
nt c
hanges i
n
the
busi
ness
m
od
el
3.
Ad
m
inist
rati
ve
a.
Dive
rsion
of
f
unds
:
Th
e
loa
n
am
ou
nt
is
be
ing
us
ed
for
pur
po
se
oth
e
r
than
f
or
w
hich
loa
n
was
sancti
oned
or t
he
am
ou
nt is
d
i
ver
te
d
f
or
per
s
on
al
gain
.
b.
Lack
of
co
ope
rati
on
f
r
om
the
key
per
s
onne
l:
If
the
ke
y
per
s
onnel
is
avo
i
ding
the
di
scussion
with
fina
ncial
instit
ution o
r has
ne
gative
ou
tl
ook.
c.
Chan
ges
in
ad
m
inist
rator
s
f
r
equ
e
ntly
:
This
ind
ic
at
es
the
pro
blem
with
the
com
pan
y
if
there
is
fr
e
qu
e
nt c
hang
e in the
ad
m
inist
rati
ve
posit
io
ns
.
4.
Ind
us
try
a.
Decli
ne
in
th
e
bu
si
ness g
rowt
h
b.
Chan
ge
in
i
ndus
try
re
gula
ti
on
s
af
fect
the
prof
it
abili
ty
of
the
com
pan
y.
It
can
be
m
on
it
or
e
d
by
analy
sis o
f ne
ws
a
rtic
le
s
c.
In
c
rease i
n
c
ost
o
f
the
raw m
a
te
rial
s
d.
Em
erg
ing
m
ark
et
s,
c
om
petit
i
ve
c
om
pan
y pe
rfor
m
ance also
aff
ect
t
he pr
of
it
abili
ty
o
f
the
com
pan
y.
e.
Chan
ge
in
c
ust
om
er
beh
avio
r
with
res
pect
to
segm
ent:
It
can
obse
rv
e
d
that
if
a
custo
m
er
segm
ent
changes
the
n
it
has
ad
ver
se
e
f
fect
on
t
he
co
m
pan
y.
It
can
be
stu
died
f
r
om
br
oker
a
naly
sis
of
A
nnual
repor
ts
.
5.
So
ci
al
a.
So
ci
al
b
e
ha
vior a
nd li
fe
sty
le
o
f
the c
om
pany
o
f
fici
al
s
b.
Investm
ent p
at
te
rn
of the c
om
pan
y
c.
Ex
pen
se
s r
el
at
ed
to
tra
vel and
oth
e
r
re
quire
m
ents
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Framew
or
k
to pre
dict N
PA/W
il
lf
ul d
ef
au
lt
s i
n
c
orpo
r
ate lo
an
s
: a
big d
ata a
pp
r
oac
h
(
Gir
ij
a
Att
igeri
)
3793
d.
Key
pe
rs
on
al
ou
tl
oo
k,
direct
or
of
the
c
ompany
is
res
ponsi
ble
f
or
l
oan
process
w
hich
is
ta
ken
f
or
bu
si
ness/c
orporate re
qu
i
rem
e
nts
e.
So
ci
al
b
e
ha
v
io
r of
t
he direct
ors
of
m
anag
em
ent p
e
ople
ca
n be
ob
ta
ine
d
t
hroug
h
s
ocial
m
e
dia posts
6.
Ba
nk
: T
he
re a
r
e seve
ral p
a
ra
m
et
ers
wh
ic
h B
ank
s
m
ai
ntain
for
eac
h
loa
n, so
m
e o
f
them
are
li
ste
d bel
ow.
a.
Pu
r
pose
of the
loan
b.
Past l
oa
n
sta
tu
s
c.
Ann
ual Inc
ome
d.
Gr
a
de of
the l
oa
n
e.
Credit sc
or
e
f.
delinqe
ncies
g.
delay
in pay
m
e
nts
Ap
a
rt
f
ro
m
these
pa
ram
et
ers,
Com
pan
ie
s
A
udit
or’s
Re
port
Order
(C
ARO
)
can
be
c
onsid
ered
as
the
m
ast
er
do
cum
ent
to
analy
se
the
par
am
et
ers
m
entioned
in
.
Fo
r
in
sta
nce
if
the
CARO
re
port
say
s
the
asset
s
are
no
t
validat
ed,
t
hen
it
is
a
nega
ti
ve
ind
ic
at
or.
The
n
asset
rati
os
nee
d
no
t
be
consi
der
e
d
e
ve
n
if
they
lo
ok
good.
Com
pan
ie
s
Act,
2003
require
s
that
the
aud
it
or’s
re
port
of
s
pe
ci
fied
c
om
pan
ie
s
shou
l
d
in
cl
ud
e
a
sta
teme
nt
on
the
pr
esc
ribe
d
m
at
te
rs.
These
repor
ti
ng
re
quirem
ents
hav
e
been
pr
esc
ribe
d
unde
r
the
Com
pan
ie
s
(Au
ditor’s
Re
port)
O
r
der
,
2016.
CAR
O
repor
t
has
inf
orm
ation
on
Fixed
Asset
,
I
nvent
or
y
,
L
oan
giv
e
n
by
Co
m
pan
y
,
L
oa
n
to
direct
or
an
d
in
vestm
ent
by
the
com
pan
y
,
De
posit
s
,
Cost
Re
cor
ds
,
Stat
uto
ry
Du
es
,
Re
paym
ent
of
Loa
n
,
Util
iz
at
ion
of
IPO
and
f
ur
t
her
public
offe
r
,
Re
portin
g
of
Fr
a
ud
,
A
pp
roval
of
m
anag
erial
rem
un
erati
on
,
Nidh
i
Com
pan
y
,
Re
la
te
d
Party
Tran
sact
io
n
,
Pr
ivate
Pla
ce
m
ent
of
Pr
e
f
eren
ti
al
Issu
e
s
,
No
n
Ca
sh
T
ra
ns
act
ion
,
Re
gister
unde
r
RB
I
Act
1934
.
This
in
f
or
m
at
ion
al
so
need
to
be
c
onsidere
d
as
po
t
entia
l
par
am
et
ers
f
or
identific
at
io
n
of
N
PA
s
ba
se
d
on
pre
-
loa
n
and
post
-
l
oa
n
pe
r
form
ance
analy
sis
of
the
sa
m
e.
The
pre
-
l
oan
and
post
-
loa
n
perform
ance
analy
sis
of
the
par
am
et
ers
m
e
ntion
e
d
a
bove
need
to
be
done
to
unde
rstan
d
th
e
patte
rn
of
perform
ance.
If
po
st
-
loa
n
perform
ance
decli
ned
a
s
c
om
par
ed
to
pre
-
loa
n
perform
ance
then
it
is
a
neg
at
ive
ind
ic
at
or.
Howe
ver,
co
ntinuous
m
on
it
ori
ng
of
the
l
oan
is
req
ui
red
for
early
detect
ion o
f
th
e w
il
lful
defaul
t beh
a
vior.
Con
si
der
i
ng
al
l
the
a
bove
br
oad
groups
,
th
e
pa
ram
et
eriza
ti
on
process
i
s
car
ried
out
t
o
i
den
ti
fy
eff
ect
ive p
a
ra
m
et
ers
for
data
m
od
el
.
For
ea
ch
par
am
et
er
su
it
able
data
ty
pe
a
nd
ra
nge or
va
lue
i
nd
ic
at
ing
go
od
loan
are
i
den
ti
f
ie
d.
The
pa
ram
et
ers,
data
ty
pe
and
values
a
re
sh
ow
n
in
Ta
ble
2.
T
he
pa
ram
et
erizat
ion
pro
cess
fo
ll
owe
d
to
ide
ntify
the
par
am
et
ers
is
un
i
qu
e
and
ef
fecti
ve
a
s
al
l
aspects
a
nd
sce
nar
i
os
relat
ed
to
l
oan
pr
ocess
hav
e
been
ta
ke
n
into
co
ns
i
derat
ion
w
hile
def
inin
g
the
final
li
st
of
par
am
et
ers.
He
nce,
th
e
pr
oc
ess
is
hig
hl
y
feasible
to
im
ple
m
ent
with
th
e
help
of
Inf
orm
at
ion
an
d
C
om
m
un
ic
at
ion
Tech
no
l
og
ie
s
(
ICT).
As
the
num
ber
of
ba
nk
s
an
d
c
om
pan
ie
s
are
i
ncr
easi
ng
an
d
al
so
num
ber
of
loans
i
ncr
ea
sing,
the
data
ca
pturin
g
an
d
a
na
ly
sis
process
for
al
l
these
pa
ram
et
e
rs
is
not
able
t
o
be
im
ple
m
ented
us
i
ng
t
rad
it
ion
al
ICT
.
F
ur
t
her,
pa
per
desc
ribes
the b
i
g data
ba
sed n
ov
el
fr
am
ewor
k desig
ne
d for loa
n p
ro
c
ess and
d
at
a a
na
ly
sis.
4.
FRAMEW
O
RK FO
R NP
A/WILL
FUL
DEFA
ULT I
D
ENTIFIC
ATI
ON
The
pa
ram
et
erizat
ion
proces
s
fo
ll
owe
d
to
identify
the
pa
ram
et
ers
is
un
iq
ue
an
d
e
ff
e
ct
ive
as
al
l
aspects
a
nd
sc
enar
i
os
relat
ed
to
l
oan
proc
es
s
ha
ve
bee
n
ta
ken
into
c
on
si
der
at
io
n
w
hile
def
i
nin
g
the
fi
nal
li
st
of
par
am
et
ers.
Hen
ce
,
the
process
is
highly
feasible
to
im
ple
m
ent
with
the
help
of
Inform
at
i
on
a
nd
Com
m
un
ic
at
io
n
Tec
hnologie
s
(I
CT
).
As
the
nu
m
ber
of
ba
nk
s
a
nd
com
pa
nies
are
inc
rea
sing
a
nd
al
s
o
num
ber
of
lo
ans
i
ncr
e
asi
ng,
the
da
ta
captu
rin
g
a
nd
a
naly
sis
process
f
or
al
l
these
pa
ram
et
e
rs
is
not
able
to
be
i
m
ple
m
ented
usi
ng
t
rad
it
io
nal
ICT.
Furthe
r,
pap
e
r
desc
ribe
s
the
big
data
ba
sed
novel
fr
a
m
ewo
r
k
desig
ned
f
or
loan p
ro
ces
s a
nd d
at
a a
naly
sis.
A
no
vel
fr
am
ewor
k
f
or
NPA/
W
il
lful
default
identif
ic
at
ion
is
de
si
gne
d
an
d
is
re
presented
i
n
Figure
5.
T
his
fr
am
ework
m
ai
nly
pro
vid
es
te
ch
nical
so
l
ut
ion
t
o
handle
the
c
om
plete
loan
proce
ss
s
ta
ring
from
sancti
on
ing
to
ea
rly
ide
ntific
at
ion
of
the
will
fu
l
de
fa
ult.
F
or
t
his
pr
ocess
al
l
the
pa
ram
et
ers
req
ui
red
for
early
detect
ion
of
N
PA
/
will
ful
def
a
ult
are
i
de
ntifie
d
t
hroug
h
data
pa
ram
eter
iz
at
ion
proce
ss.
T
hese
par
a
m
et
ers
need
t
o
be
c
ollec
te
d
at
the
lo
an
ap
pro
val
le
vel
an
d
the
n
c
on
ti
nu
ous
m
on
it
or
in
g
has
t
o
be
do
ne
unti
l
loan
is
com
plete
d.
Durin
g
m
on
it
or
in
g
the
patte
r
n
of
loan
paym
ent,
transacti
ons
carried
out,
be
hav
i
or
al
a
nd
s
ocial
trai
ts
are
analy
zed
and
if
th
e
patte
rn
is
not
no
rm
al
it
is
identifie
d
as
outl
ie
r
beh
a
vior
and
he
nce
po
ssible
def
a
ult case
.
T
his
process is
c
arr
ie
d o
ut lo
ng
it
ud
inall
y u
ntil
the loa
n
is
f
ully
p
ai
d o
r decl
ared as
NPA.
Accor
ding
to
E
\
&Y
s
urvey
[
1
8
]
early
wa
rni
ng
sig
ns
to
i
de
ntify
def
a
ults
m
us
t
le
ver
age
te
chnolo
gy
and
data
analy
ti
cal
capab
il
it
ies.
On
ly
te
ch
nolog
y
can
bri
ng
rev
ol
utio
nar
y
sh
ift
in
NPA
m
anag
em
ent
i
n
India.
Assistance
of
Au
t
om
at
ed
so
luti
ons
in
data
a
naly
sis
can
ena
ble
early
i
nd
ic
at
or
s
t
hat
will
gen
e
rate
al
erts
before
the sit
uatio
n be
com
es w
orse.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
9
, N
o.
5
,
Oct
ober
20
19
:
3
7
8
6
-
3
7
9
7
3794
Table
2
.
Param
et
ers
ide
ntifie
d f
or
t
he data
m
od
el
Sl no
.
Featu
re
N
a
m
e
Valu
es
Ideal Values
1
Ass
et Co
v
erage Ra
tio
ratio
>1
2
Deb
t
Equ
ity
Ra
tio
ratio
1
:0
1
3
Deb
t Ser
v
ice
Co
v
erage Ratio
ratio
I.5
4
Deb
t/EBIT
DA Rat
io
ratio
<1
5
Fix
ed
Assets
to Ne
t W
o
rth
ratio
<1
6
Interest Co
v
erage
Ratio
(
ICR)
ratio
<1
7
Lon
g
T
er
m
Deb
t t
o
T
o
tal Asset Rati
o
ratio
<1
8
Cu
rr
en
t assets
cur
r
en
t
liab
ility
ratio
>1
9
Total Exp
en
se Rati
o
(
TE
R
)
ratio
<1
10
Interest to
sales
r
at
io
ratio
<1
11
Cred
it r
atin
g
agen
c
y
Ran
k
in
g
Po
sitiv
e gro
wth
12
W
rit
e of
f
s
Frequ
n
cy
Frequ
n
cy
13
Cu
rr
en
t liability
to f
ix
ed
assets
ratio
<1
14
Cred
ito
rs velo
city
ratio
1
15
Sto
ck
velo
city
ratio
1
16
Deb
to
rs velo
city
ratio
1
17
Los
s o
f
sales
catego
rical
lo
w
18
Su
p
p
lier’
s lo
y
alt
y
catego
rical
h
ig
h
19
Cu
sto
m
ers
loyalt
y
catego
rical
h
ig
h
20
b
u
sin
ess
Mod
el
catego
rical
No
21
Div
ersio
n
of
f
u
n
d
s
Bo
o
lean
valu
e(Ye
s/No
)
No
22
Ou
tlo
o
k
o
f
KM
P
Bo
o
lean
valu
e(Goo
d
/Bab
)
Go
o
d
23
Ad
m
in
istrato
r
Tur
n
o
v
er
Bo
o
lean
valu
e(M
o
re/less)
Less
24
AR:In
v
en
to
ry
Val
u
atio
n
Bo
o
lean
(
Do
n
e/Do
u
b
tful)
Do
n
e
25
AR:Lo
an
sactio
n
ed
Bo
o
lean
(Goo
d
/Ba
d
)
Go
o
d
26
AR:Statu
to
ry
Du
es
Bo
o
lean
(Yes/No)
No
27
AR:R
ep
ay
m
en
t of
L
o
an
Bo
o
lean
(Yes/No)
Yes
28
AR:Man
ag
erial
r
e
m
u
n
e
ration
Bo
o
lean
(Yes/No)
No
29
AR:Priv
ate Pl
ace
m
e
n
t of
Pr
ef
erenti
al I
ss
u
es
Bo
o
lean
(Yes/No)
No
30
AR:No
n
Cash
T
ra
n
sactio
n
Bo
o
lean
(Yes/No)
No
31
SL:Purch
ase h
isto
r
y
Bo
o
lean
(High
,
m
o
d
erate
,
lo
w
)
Mod
erate
/lo
w
32
SL:I
n
v
est
m
en
t:
Bo
o
lean
(High
,
m
o
d
erate
,
lo
w
)
Mod
erate
/lo
w
33
SL:Social
Lif
e
Bo
o
lean
(High
,
m
o
d
erate
,
lo
w
)
Mod
erate
/lo
w
34
SL:T
r
av
el
Bo
o
lean
(High
,
m
o
d
erate
,
lo
w
)
Mod
erate
/lo
w
35
SL:App
arels
Bo
o
lean
(High
,
m
o
d
erate
,
lo
w
)
Mod
erate
/lo
w
36
SL:Social/r
ef
erenc
e gro
u
p
s
Bo
o
lean
(High
,
m
o
d
erate
,
lo
w
)
Mod
erate
/lo
w
Figure
5.
Fram
e
work for
NP
A/W
il
lful
d
efa
u
lt
i
dent
ifica
ti
on
Bi
g
data
te
c
hnology
is
s
uitab
le
to
deal
with
data
that
is
not
only
struct
ur
e
d
bu
t
i
n
a
ny
f
orm
at
,
in
real
tim
e.
The
earl
y
detect
ion
of
will
fu
l
def
a
ul
t
need
s
analy
s
is
of
unstr
uctu
red
data
a
nd
ge
ner
at
e
al
ert.
Hen
ce
util
iz
at
ion
of
Bi
g
data
te
c
hnology
is
esse
nt
ia
l.
Su
c
h
ea
rly
wa
rn
i
ng
syst
em
with
Bi
g
D
at
a
capa
bili
ty
help
i
n
identify
in
g
str
ess
in
banks
a
nd
im
pr
ov
e
l
oa
n
m
anag
em
ent
li
fe
cy
cl
e.
Further
Bi
g
Data
can
be
le
ve
ra
ged
in
loan u
nd
e
r
wr
it
ing decisi
on m
a
king a
nd NPA
m
anag
em
ent.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Framew
or
k
to pre
dict N
PA/W
il
lf
ul d
ef
au
lt
s i
n
c
orpo
r
ate lo
an
s
: a
big d
ata a
pp
r
oac
h
(
Gir
ij
a
Att
igeri
)
3795
Cl
assifi
cat
ion
al
gorithm
s
are
re
quire
d
t
o
buil
d
pr
e
dicti
on
m
od
el
for
N
PA
/wil
lf
ul
de
f
ault.
He
nce
pr
e
dicti
on
al
go
rithm
s
are
i
m
ple
m
ented
us
i
ng
m
achine
le
ar
ning
util
iz
ing
var
i
ou
s
str
uctu
red
a
nd
un
st
ruct
ur
e
d
par
am
et
ers.
T
he
se
m
achine
le
arn
i
ng
predict
i
on
m
od
el
s
are
desig
ne
d
us
in
g
m
ap
reduce
l
o
gic
on
ha
do
op
big
data
platf
or
m
[19
-
21]
.
T
he
cl
assifi
cat
ion
al
gorit
hm
s
co
ns
ide
red
are
Nai
ve
B
ay
es
[22],
L
og
ist
ic
Re
gr
essi
on
[
23]
,
S
upport
V
ect
or
Ma
c
hin
e
[
24
]
,
Ne
ur
al
Netw
ork
[25]
and
Ra
nd
om
fo
rest
[
26
]
.
Th
ese
ar
e
i
m
ple
m
ented
us
ing
Ma
p
-
Re
duce
te
chn
i
que
of
Bi
g
Data
on
Hadoop
Cl
us
t
er.
The
m
od
el
s
are
com
par
ed
base
d
on accu
racy
obta
ined
a
nd the
al
gorithm
w
it
h
best acc
ur
a
cy
is co
ns
ide
red f
or predict
io
n.
5.
EVAL
UA
TI
O
N
O
F P
REDICTIO
N
MO
D
EL
S
The
eval
uation
of
the
m
od
el
s
is
do
ne
co
ns
i
der
i
ng
str
uctu
r
ed
an
d
unstr
uc
ture
d
data.
Str
uctre
d
data
fiel
ds
i
nclu
de
Loa
n
ID
,
Cu
stom
er
ID
,
Cu
rrent
L
oan
Am
ount,
Te
rm
Credit
Sc
or
e,
A
nn
ual
I
nc
om
e,
Y
ears
i
n
current
job,
Hom
e O
w
ner
s
hip,
Purpose
, Mont
hly De
bt, Year
s of Cre
dit Hist
or
y, M
onths
since la
st
delinquent
,
Nu
m
ber
of
O
pe
n
Accou
nts,
N
um
ber
of
Cr
edit
Pro
blem
s,
Current
C
red
i
t
Ba
la
nce,
Ma
xim
u
m
Op
en
Credit,
Ba
nkruptci
es,
Tax
Lie
ns
et
c.
The
dataset
co
m
pr
ise
s
of
a
round
t
wo
la
cs
of
r
ows.
U
ns
tr
uc
ture
d
data
co
ns
ide
r
e
d
include
s
synth
esi
zed
so
ci
al
m
edia
data.
Senti
m
ent
analy
sis
us
ing
A
pac
he
Hive
i
s
do
ne
on
this
data
t
o
get
so
ci
al
outl
ook
value
[
27,
28]
.
If
this
va
lue
is
po
sit
ive
it
ind
ic
at
es
posit
ive
li
fe
sty
le
.
Paym
ent
data
is
al
so
consi
der
e
d
t
o
get
sp
e
ndin
g
pa
tt
ern
s
values
.
These
value
s
a
re
ad
de
d
to
the
data.
Sp
e
ndin
g
patte
rn
a
nd
So
ci
al
ou
tl
oo
k
are
t
he
par
am
et
ers
f
ro
m
Table
2
a
nd
a
re
synt
hes
iz
ed
f
or
t
he
pu
rpose
of
valida
ti
on
.
T
he
ai
m
of
t
he
m
od
el
is
Loa
n
de
fau
lt
pr
e
dic
ti
on
.
For
t
his
pur
po
se
pre
dicti
on
m
od
el
s
a
r
e
buil
t
us
i
ng
m
achine
le
ar
ni
ng
on
Hado
op
a
nd
spa
rk.
Mult
iple
m
achine
le
arn
i
ng
m
od
el
s
ar
e i
m
ple
m
ented.
These
m
od
el
s
are
eval
uated
ba
sed
on
accuracy
obta
ined
.
Ma
chine
le
arn
in
g
al
gorithm
s
con
side
red
f
or
bu
il
din
g
pr
e
dicti
on
m
od
el
s
are
Log
ist
ic
regressio
n,
Ne
ur
al
Netw
ork,
Ra
ndom
Fo
res
t,
an
d
Naive
B
ay
es.
The
res
ul
ts
ob
ta
i
ned
f
or
the
m
od
el
are
sh
ow
n
in
the
Fi
gure
6.
As
dep
ic
te
d
in
the
fig
ur
e
Neural
net
wor
k
has
the
high
est
accuracy,
hen
ce
is
us
e
d
in
the
process
of
pr
e
di
ct
ion
of
NPA
and the
re
by wi
ll
fu
l defa
ult.
Figure
6.
Eval
uation o
f
cl
assi
ficat
ion
al
gorithm
s f
or predict
ion
6.
CONCL
US
I
O
N
Ba
nk
i
n
g
is
th
e
m
ajo
r
se
r
vice
sect
or
to
bala
nce
the
ec
onom
y
of
the
co
untry
.
T
he
loa
ns
goin
g
ba
d
intenti
onal
ly
are
no
t
only
affe
ct
ing
t
he
bank’
s
pr
of
it
abili
ty
bu
t
al
s
o
ca
usi
ng
set
bac
k
f
or
the
ec
onom
y
of
the
country
as
a
w
ho
le
.
Th
e
te
ch
no
l
og
ic
al
asse
s
sm
ent
and
s
upp
ort
f
or
early
identific
at
io
n
of
s
uch
will
fu
l
def
a
ult
is
the
need
of
the
hour.
It
is
i
m
per
at
ive
that
custom
ers’
en
ti
re
pr
ofi
le
including
be
ha
vioral
,
fina
ncial
,
so
ci
al
par
am
et
ers
hav
e
to
be
c
ons
idere
d
an
d
m
on
it
ore
d.
I
n
this
pa
per
a
proces
s
for
ide
ntific
at
ion
of
crit
ic
al
par
am
et
ers
is
desig
ne
d
for
e
arly
identific
at
ion
of
will
fu
l
def
a
ult.
This
pa
ram
et
erization
proces
s
nee
ds
to
be
integrate
d
into
the
process
of
loan.
He
nce
a
novel
f
ram
ewo
r
k
w
hich
ta
ke
s
in
to
acco
unt
sta
rting
f
r
om
loan
sancti
onin
g
ti
ll
com
pletio
n
is
desig
ne
d.
The
fr
am
ewo
r
k
is
bu
il
t
us
in
g
big
data
te
chn
ol
ogy
as
it
need
s
to
deal
with
both
struc
ture
d
and
unstructu
re
d
para
m
et
ers.
In
or
der
to
ch
oose
the
best
pr
ed
ic
ti
on
m
od
el
in
the
fr
am
ewo
r
k
an
exp
e
rim
ent
is
cond
ucted.
It
is
carried
out
on
th
e
loan
da
t
a
set
wh
ic
h
is
structu
re
d
an
d
the
gen
e
rated
sy
nt
hetic
un
str
uct
ur
e
d
data.
Va
rio
us
cl
assifi
cat
ion
m
od
el
s
are
buil
t
us
in
g
m
ap
red
uce
and
com
par
ed
base
d
on
the
accu
ra
cy
.
The
res
ults show
that neur
al
netw
ork
has
the
be
st
pe
rfo
r
m
ance,
a
nd
he
nce
it
is
i
m
ple
m
ented
in
the
f
ram
ewor
k.
The
r
esults
al
so
in
dicat
e
that
in
order
to
ide
ntify
will
fu
l
def
a
ult
un
st
ru
ct
ur
e
d
c
om
po
ne
nts
play
a m
ajo
r
r
ole.
Evaluation Warning : The document was created with Spire.PDF for Python.