You are on page 1of 18

Data Scieuce Hasiqnment

1 Stody about Panda Ho dotafrarm e ets


impemente d B Pauch
Ans1-
Hhs1 Pandas is thon braTy
with clato
a
Py
Rels
0ed o worRi
o Ri

I Tt as fonchos to ceanin
analyzinq, cJedning
explovimg ,2
mannibu laing da la
Pandas alloun U to analy3e big data and

ma Re concoa icns bosed on 8ta HaHcal theo rie

Pandas Cau cJecn meg 8y dalo 2e & make


hem Teoadable resevat
Tt represen ta the data in abula wa
ie in roUR cofom&.
3icn
I t a o Rubporth n dexinq, aicng,
subne bling in a laxge da ta. geka.

These aHe 3 data 8tvucoes Pandan


i Serie (One- dimenaional
cii Data fsame (Two- dimenaiona
ii Pane (Mugh - di men &iono)

Data Frome
APonds Data Fra me i a 2 di mengional data
data

9toctuie, Gke a 2-D a3 1a, tabe


wih rowR & Coloum

Data Frome takes iga dic


ionavy, Beries

data frome a au drqumeub


ano thes

3 data stucturey, padas ii


Among a
Amon
most elfideut
#Suyotas
import pandas
as pd
t cseatng dicBonay

d name' : [Anand', Slikher, Kovtik,

centage [90, 25, 95] f


per
Po dat
pd DataFsame ld)
Oulput
name percenta
0 Anand gO
Sikhe 85
Kar iR 95
to dataeet
slo re bbiq
Aimple way
wa4

to ue CSV les Comm a eparaled Hes)


plai n text and i a

CSvN file Contal


Hhat Cau be ead by
t e Rnown oy mat
incudiug paudaas
Gnyoue
SV in to Data Frame
tho
load

import paudas as pd

pd. vend -cgv (dato csv')


d
to- sbing)) to.gtving
print (df. Aed fo print
tha eise data.
2 . Ditferen bate b/u inea ad non-i nea
eeon

Pn-2 Linean Regveazion


Re
ineax Reqsexa ion ia inear op pro o

tla elotiomuip blw a colen


do modelnq
res ponie One or mo re explo na tory

Yoyia bSe (deßendent ov in depeudenl vamia ble)


T hThe
e cCase o one exp la na tory vaTiab)e

called gimple n tor re9 ressi m, for


mOre tho
thcun one the þrore mu lhple

Rinean eqeiou.

Lineax Tegreto
i8 a inean mode, it
a8GUme a inear relation sup b/le inpuf
Varia ble (n) 2 ain q.o output vas iadle (y)

Cau be calcula fed fvom near

Co mbi na fion o input va riahle (n).


gor 8imple inea Teqreion problem
*
y= Bo t B,
B coefhcieu
mo des
Complexity oLinear reqenim
depeuds on numben of oefreseut
Parepaving dald fos linean Teqsewion:

iinea AsRUmhom
) emove Noig e
(in) Remove collinêai ty
riv) GaavARiom DiR kihu hon

(V Restale in pus.
Non- inean Regseon
T i a rm o1eqretion onalyinin hic
obRevvo lion al dato e modeled by aonchon
wich a non -in ean Com binq hon mo deO
porameleys depcu ds on one o move
indobekA
VOv iobes

Simple inea r e s e 8A1 OY elates two vaia Lles


(xv) oit a alrayghl ine (y = math),

while non -inneo veJafes e


two vaviables a non-inean (cuTved)
elahonsib
Non nean re qrexion U8es Jogarihmic onehonk,
ignemebnic fonchcns, Pxponen Hol pnchionus,
pou en tonchons, Loveu3 Cuvex a usAlo tuneHa

X othen tH9 e thods


Non-inear models axe move Complica fed tho
g near m odeh to developr berause to unehHon
inear
iCreated uxing eries e approximah'ns hat

nod
moy Rlem for trialå eyror.
Linean regressi on odel, wkuide he ypically
O1m aa
straight ine Can alka form cuxvCA,
debending dn e oTm nean veqr enion
Lhe wise, it's po2Ai ble to Use
algeha
to rou2 form non inean
Q3 ExIain Supfosl Vectov Mocwne aud ke Tne lg
in SVM.

Awa3 Suppo«t Vectos Macline (SVM):


Svm U8e d
fo ClosLcaion as we os

it i
Howeve brma1 ly,
egenion psobleo
in HL
Usedo daniica Hon Pro ble
to Creae t
The
he Goal aSvM algo i hgt Cu
best ne 07 deciaion boonda
into claue
dimengio nd 3pare
egae he
te
8o hat we COu eaisly put
in tho
dala po int in t c6TTect Categovy
boundary call ed
fo tore. Th is deci 3ion
a helaue

SVM choose the ex tveme poinls/veckorg hat


These extreme
hel in creag thyperplane.
Case aye colle d upport vectov*
Posi bive Hplon

Maximu M

Morg'n
Su
ppot
p'one Vecto

Neothve
p e a ne

fo ide nify weatkor t i a


e 1we
waut

be
Cat 30 guck modek cau cveated
dy OY

gVH alg0
by iug Ne Oata

Mode fedich on Outpu


Pasl dabeled da
Trani
Lsnea SUM
dalo
dale
Linean Svmiv2ed fov ineovy xeporable
wich meau i a dala 8et be elomifed to

tuwo cdases by ugv Ringle yaiqht ihe, Hhou


suc dafa i te med a lHnea
lineary SepoTable
data &clasifien i ralled Iincar SUH clawihie
i Non -inean 9VM
Non-inea Sv) i U8ed for non-nea1y poraded
be
data, wlich meo n a dadosel Cau ot
hen cl dale
catyif1ed by u8Ing stai qht lre, w

de la & clai ther


teymed hon-nea

USed i called Non ncar alouifier.

Kerne in gv
TE a fun cion UAfd bo map a o wer-dimen Jiansl

data indo gho dimen3 i onal data.


Kernel ooc Hon gen exally trausfovm tha
rai
trawing
set of data 80 that a non- Lirea decis1m
80Tfa ce is able to tra so rmed to a n ea

eg va Hon tte n a ghey nu mben of dimeusion


3pareb

Btondard Kernel pnefion gu'


k(T) =1 f/al < =
k (7):0, oterwige

Major Kerney onchons


tor ioplementing Kewnel foncbioy, frt oall
bave to ina tall sci kit Jeovn".
gbra"y
-
we

CommOd
rompt
P'P in3 t a l 3cikt-Jearn
(0)au8siau Kerne
- Used for ausos mation, wheu here 1 no pYior

nousledae about dala


e-
knsy) = 2

Cna Ghavssla kes ned Radia ose toneio (RBF):


Thi
This onction is ged dor adding Tadia basc
me dhod to improve te trausoMoHe .

k( e (8rllx-y)
cii Siqmoid Kerne
Thu une ion i
eguivalewt to twoSayer, pe r cebtron
mo de
e0
neural
netwerk, w ich is U8ed a

achvatim tunch on oy orHteial neu


k)= tan ( T. n tr)
( Polynomiol Kerne
Tt repreen tx te imilari vecor in troui
ro
3e of data in a teature space ove polyuo tals

the origi nol vorIab las Used in kerne.

Kny)tan&(Yryt 7), 7>0


4 DscURs bauRing aie atudy teVecommunimhn
ion in ts llahu
Btudy a men
aohvidies
cou predio4rauduleu
How
Ho aualy Hea
in paniculan area)

Aws TeSe comm unlea bioww Case 8tudy


Type tedecom netwo1R'
bosrd
b o doo
netooskg ane diinguizh ed
eneyasN,
heis gegroplical pa
i PAN( Perzonol Area Nhwosk)-
Smallext N/»veyy perRonc to a uAe

deuice
Tuis incdvdes bloetoot euabled
ov
This
intro-ved enabld device
Ont"
PAN cOnnecivi ky Taui is uto
reles
I It
t incdude, wiyeles Ray boasd, mo u3e,
pvintos, TV e motey ete

i LAN (Locol Avea NetworR)


A compukes nekuoo 1k apanned ingide a bur
bui ldta
ldig

operoted under ing e adwi TaHive uate


twed a LAN
ixgnealy
LAN coverk orgowizaha, ajHee, gclhoola,
9cologe ele.
LAN U2es ei hen Ehevnef or
Tokeu-i
7oke -vivg

Tekn ologj U9e Atas +opolegy


MPN (Metvopolitan Prea Nefwosk)-
gi
MAN expands Hhro ygh out a i y ucL as oble

TV netwoYk.
be in te tom Ehes
ca
ne f
7oReu -Ting, PTM ov FD DT.
Back bon e d i kigh opority2kigh
speed ibe o lieg No1by in h/
LAN &wAN

Civ ide Avea Nelworh (tAN)-


i d e Pyea which may 3how ocro4 provin rCA
&eve auhole count
rounfY.

These nebovorks brovido (onnechiv i ty to p N

&LANs
They ane
eguipped ith vey gh abeed
Bock bone, WAs U2 ve ry ptive
netuos k egipm

WAN may be manag ed mu hhle Odui

2. Role e Analybes in Telecom Indestry :

The Tole ofdo ta analyhes in telecom ix 6


pTovide componieb it Ho ea3ieat way
Oncovrer insiGhtz Tom al e i r data.

Uaing
Using a tele com data ano yica 20l tha

wi help a Conpony ain beHer inaígk

wl0 gener oe re Profiba


Such a
e wGel
t TeSecom Data Analy ies Allou Betfer Uae
In veatuet»
The Tole f dato auoly h eg in tolecom i b give
eck ompany a uwieted vieo #eir date
acrORR depovtmen tal ineg Wle data sfreaud
data 800Tcey rougl
in te rom utiple
t Coupouy, te owpauy dau toe aduou tagt
Cow ou
(ome u wrh
of al0ik team

best Aol for every lalleng.


( nive Data Analyhiea To þriovitor TeJeco
leJecou

3UCes

euployees
Ne Necto Tedecom da a onalyios 3yateu,
e e dalo dis lay ed i hnnig fnjogopicy
In a
fagy undexsla vd format Tey
They
fo add is own no tes ko to data
be
be able
no hou
t disoover eveu moYe fuqh

Cowauy i pesorwig
0i Dak Analyti Top mboska ce forTeletom
% to
eslecom wi
8esvicea have nea
Loor lds totepex d/'ng ou &mayt da ta
oloud eckno lgy T your copavy g u't ou

boavd i t To wold- alas teleco da ka


de4t with
nalyHcA 2yate yoll be Roon

dino 8aUr death bloust0 te daBo -heedd

telero indus
(w)Connect He Daka Do wlt Ow Hed Borihes

a cen troliged data dwalyoR of, all te


depas tmeut data me vge in fo dve ceuhtyde

cac depavtnent oau add ib Oun aiguk iufo


te da a gleawed, gtvl anober perpeh'ut
Hhat wilo Dea to 8o
a
2ui cky
A 4t deba t we eoy ro
eac ohes, fnCreay 5
efrie bo pshug
pr oduch'y
) Mage aBeter uhre wl Teleco
Dak Aalyica
A Can accey e da ta aCcovdr
pr vious Aef by T dapk.
a ki g 1i A Qove Tu ed & AerUne Rol
Pu he colle chue kec ds dogeher to
nd
o we &evVice coAtx ive ovpa ulon,
a i m ediadly uhen KPI oag o
2wOoth profi
dept oil0 on oon a
geerat

Fvaud Deleeio
3 0plim12 aion-
The delecbion Fraudusod achvihes I oue of
ho biggeat challeqe to tte teleco imog y
Ttedecom induay, along wit having He
mogt no. uerg alta wit nexes a Jorge
num ber Cause o royd Accasdiuo

The most ommon raduleut ackvi he in


e bedecomwrld e un auBhoriged acoe,
faRo ro las, mauxe Csedit/debit caud
in foima i n , ete.

The tedeco Indoatrie ane atg VriouR


Un Bu pe vized ML alyer foy delecking Uh ojal
bsen aoHvihes &.preveHLe auds
TedecoM
eg enberpriRe Voda fpue i wokig
wih Argyle dala for deteohivg k prevewha

saud wit hel oaud oualy Hos.

4. Pxevenh u Customer Chusu'


vor
iou Servies ohened by Teleco ivdushy
TV, internel, plowd,tkc. Haking te
COtom er
believe thatyou ane

eis time X challehallegitesk,


Thop we
eed to ape proper & accoTate
Cn alkc fo undev a toudlung cu to meo 2
behavio Y
They
They ex vuct valuable
ingigkh obou Cutomera
eeling rom tte cUtomer tvanzocHon da e
analy3 the T elpr te tede co
Indugby buldiv 2qiasootory o to
R&O Tlis
belte
them in lp duT (
Vice
Gvo idivq cUoMer
chT

Ana yx is a
todiv IT 3enice provider
in
Furop Rnown tor applyi 03 aol
to mayk eng
Tlo Eusopean teJecomMUWcatioans opera tora
have teomed u wit tte Analyx
&el thew in ideu Hayivg9 fe botcuhel
chon e TuigLel e in ta ri4
ed fechve measuve o prvenio

CUg Fo m e churu .
Bankng Case 8bd
paheru i
bonka to deutidy
The ane
ty trawgachon data
am ount avai lab le
g avai

fintemeing elcaeutly i
tteir (UBto
wera
tho
Ho da la
da te
with DS in baniug,
Bauks uli0ige
igto y , treud,
om custo mex trausacfio us, previo U
CommUNca ti on oyaly
Variou
VarioU s method da ta analyi ko dala fugfon

&integ a hon, ML, Nabral Jaug uoqge oce


torelug,
9m al
9 pYoces lug,
prores cO be ued for tiz þuspose.

Fraud Detechon
The
The ban ks ae carcing o wa hat CO

detectraud as eavly a posible for Hinint13ig


t loe
Dala Scieuista Ca imbrove +hoevd ovkoMe
gecUi CO be do monitori
a ha3 ih9 difereut bo aoivi hes ote
CuA om er o hat fhoy Can detet
det ect d

80spiciouA 6s Halicioua ae Hvity.

The inckuded in raud detechon


najor steps
Proces ae-

i CollecHug a age numbtr ofdata 8ales


ode.
for trowng &Hestq
i Traiwiug te MOde o Makiv predictions
lla &
i) TesH to áccUTacy otto yeu

deß loyueu
thei bonds o
to have
d
Data Scieiat nee

iko O1oCa Hau,


Variov data mi
miui
g becligues

cdost evig,
evi cdasificakio, ele. ual fo worhig
Cost
difesu dalo keh entrociig bom e
wi au be obpl1ed
ha
meaningfu in3igh
to Yeafime baulsiy pro} es
hat
that holdx
o ldx
Ayatem
e Leb UR conider
a
ther trau2aotions 3uddeuly
uhu o a
sasaohas OccU so
o9rg no
no.

account Until ho ownor of


CUgtomer 's
account ielf verisies he
Pata
2. Managing Cugtome
aMe U2i ng vario u
Vasio0 bann9 orqaigabos
d echnigves om DS and ML just
tool an

to tranoTming bis dat into 8uch a fusmat


that it Can be Used o Rnowi
Rnowing eir
c9ienta beeter for devising ne9trateqies
fo bettey Te ve nue gener a tian

i726 aply Bevera


The Data Scien iat
me thods to 8epay te he data whi ch i

Use dtud rom al da t . Te anoly8i


4Hlis daba helps em to gan inzigh t about
abo ut

CuR tomer bekaviov, priovihe, etc. Tuig w hel


model.
in
building efftet
Applying differemt ML a
go Cau e banks
to derive new op porhuui ke for YevenUe
ne sa bion & taka 0me impon taut da ta-dsiveu
e
deci sio
3 Rigk Modelivg
This ideuh ficaion evaluaion risks ix
matter
matte Onceyn o r e inveatMen banks
banks
2
yes oRiak Mode liug
iCuedit Risk Modei
Thig alow Hte ban k to predi d whefhes a
Customes wi be able to Tepoy heii r
e loou

by
anysig peviov Lixtoy redt
Tepo of the CUs toe
The Cedit rigk Qualyai heSps calau lak a
Tigk aose do eaak indiviaua Case. Ten t

bank sanehon te Loa 07 uot debendi ug upou


te iak score valve.

d Inveatmeut ak Modelg
T n vest men banks vge TiAk mcdelha fo

debechng 712ky inveatuen Tu w helb the


to better nvetuens advi ce to f
ive
evgtal
4al kimg
ki nq ta g t decaiaso
Tkese ae easou
inCveasi ng ptoit
bank
Tigk modeliq ipoytou

4. Read ime Pediekve Analyalg


I n bn
ban nq,
nq we a apply vaviou anayica

metho ds aud desive Some se infoymahio


p1tdictutose evente

These e boslcalyto ye analyhca used


in ban Kia
bonks o Conside
aReal -

bne an
aly hea enabls

Comet2e nario& Ga ka ackon accosdivly


h
t predicf
bPaedickve analyhoy hel bans 01
to ftat
about t We
ometi vg utore.
to phe dicd prbl ta
&elp Le
1&els Louk a

fek
ight aphean in e nea futo1e&ek
8utable achow Juat to iui e tupoet
bUA iu eM

Coabomer Lifekme Vae edchon

CLV waoe vefesg t predicting value


Value
net
neb projt 6 Tt t vaJve that a boxine
teis
Tueiy
wigain rom A to mer dustng
en ive reSa Howsu
moat Co mmon' ug ed Dak Sctenoc tool
Ciloni heaho 2Reqve io
pur pORe
l(CART), 2lepuise Reqreion Grenesaliged
Lineax Nodels CCLH).
Using Data Sci eu ce mode opredeh va
TLe CLV acuatomer oi hel te
8uitqble de a o u
anigations to take 2 oe

do their 9TowX proit

6Coatome uþont
Vosi ovs CUg omer 8u pbost i e 3Ervice pyovi ded

bythe bonk incode CU2do'men'


9ues tHow å complai l* n a eart a POsbe for
un
dergtandi rg a bettey wa
U
Data Science in ban king helping te Bonkig
Indus ty to automale hi Sesvice that w

psovi de bettey ae vesponaes


coso mev & wi0 alao hel te Couepawes

reduee i i n ves tueut o He 3 wOuey


ebloyee
Cus tomer Seq men taton
uslo me Sey weutobi o-
Banks pom e acivity
dividtng customesg Into apoic0Toup
behaviou 1e
9sMoups Cau be On bi of
behavioToleqwentadiom 0 o te boaiA of
specio ch araodevig tieg o vdomer

demog vaplie deqeutatio


here
There aMe dilevent Dale Soieu ce teobives

&uch as osoakving , deciz iion tveck, oqix he


Yeq ession, eBe. thad ca e p banks o it
the, t e Cau prediot the CLV dor diyfe aut
s Men Cugto mer accoydingyost fov
idenHinq ighlw -valve coatower eqwe ytz
Cvgtom er SeqHenta hon UAed or pro id u
better costomer Bevvi cea & inpsovitq e
oyaly c t Cogtouera

8. Recom mendation Exgines


data Scien tixh taRe a e uge dota s0
their thiey previoug Bearch iatoy, sousack ou
daa b oualyze then 2 the predicf
hisoy,. pofile
ha+ migt intes et
t mo:t accoale itema

Tho Ye commen dation


e
aOu be bult
T kb o Cola boTaive
alcos
be ei her Usfower
terit
teris method
OY
tha
iteM- cen bri e
Cau

Tt evalua tes te
centic
bAergbelaviov provl do
Contenb-Baied
he Re Cond one i
tesg algo, t eCo m m en ds

axe p y ed by Ho
tomsto du ey 0t
inleracded duv tg
rodu eh
i ch hey
aoti vi He.
hels previou)
elho ds Cou be Ae tor
An o above m

building co mmendatton engino accosdta


a 7e

toyou1 gools iscu n 2-lemce

Ho YES BANk vsed Data Seience 7


They are
docig on
predichi ng futove eve
even

have 2ome tpact on t e eal


hat
ime b Uain ex rna king Dafe
Screu ce ban Ki
They have p a nded y 2e n g up a eparae

team of Data Seren H z for makiug 8oe


eam
im po tat dala- hr iven decinio dor providig

be t poARiHe
gevice Heir elte.

Te YES bo,k Pestosmed aReceh cy, yevey

tat a RFM analy 21A


Monetary
daba CUgtomer' debit ard UAage 20

that they Co rovid perso nali3ed


Uaomey experieune by p ovidi g Horqeted

The bamk also developed a jovedv'chve


mode by Jage aoU

CUB Fomev dala Tig helbed teu

tnce a)e &ales opera Houal gol

You might also like