lOMoARcPSD|34962087

Data mining unit -3

Computer Science and Engineering (Jawaharlal Nehru Technological University,

Hyderabad)

Scan to open on Studocu

Studocu is not sponsored or endorsed by any college or university

Downloaded by Priyanka Gaikwad ([email protected])
 lOMoARcPSD|34962087

13
ayes c
lass ificodion Bayesvan_elemicaln
-Boye sion lassievs ave statslel classhievs,
perforin pro bablistic pved alan
A statbtcalassifier-

(ie) predicbs cla6s me mbership probabiltizs

probab,litit
probsblitits
dit c lass
class me mbership
mbexship
to aa patler
partulav
BaySian classifievs Can pre
ple be longs
belngs
that a given
Sweh as the prob.bilily

class 8uys
theove
theore n
n

is bascd on
olaskation
an
oyesi
orarje
dak.ba ses
is
geod
n d e p e n d en t e

C o n d t i o n ad geet e
c las tht the
aume
clhssnfiers independenty the
the
Naive Bayesian indepondent oy
c laSs is
a given
Vae
om
an attibule
other atribules
vals the
the

Bags Thm taple (

dala cvidence
be a
d.
-het x
Considere

3
X altibutes
J n Measremenb
Bayehian
terms Set o n the dl
ha de
ha that
Snch
Sus
the &is
Some hypo
Aet H be clhss c .
a Specificd
deErmine
delermine
PlH/x)
P[H/x
t plexbelongs
Problen ue
e ANt to
oAnto
problen H holdsiven
clhssificaton the hypothesi's
-For tht
dli -plex
the probnhi observed
the "evidence cond: oncs on X
H
i
the poslaror probili
-
PlHIX)

Downloaded by Priyanka Gaikwad ([email protected])

 lOMoARcPSD|34962087

PlH) is pY ioy pYoba bili y H

obsevved
that sa
mpk d l is on H.
Plx) is probab:liy Condiond
Conddioned
on
H
probabilik Xx
is the
the pus&sior
36 PIx/H)
=
raln). Ply
P(H/x)
SayoThn Plx)

Bauo clh ssiha

Najves
classfler

is a Simpe Bayoran Posteyiori

Naive Bayes
to derive the maximum Posteyiovi
classification
is
theis
theia assoiated
assxiated chss
clss
and
and

Let D be o trom7 set y tuples trib.t vela

n-D
cach tple s prtseded by an

bes and

x4,X h clsses ,4--- m a iTnd

the maxi oad
de) the
lEs so u
Suppose these are

leaion)
desive
the hu
m pos
c l a s s i b i c a t i o n
is to

6 a y e o 1kemem
Tkeen

Pcix) eriVed
ved m Bayio
cn
be l C alx)- P { k l , j Ac;)
can
This

Px)
all clesses onl,
constant for edst he maXimized
he maximijed
haeds t
Sinie
Plx)s
Plxle;) Plc;) the he ohich c;
Plclx)= 0bel is
delzd class
the pre
wos ive Bagsian
Jn ottor
maxima

P(xlc;) PCa)
is the classlabel
sing
Pre dictng
classification naive Bayesian
a 4uple
tple ASing
clam lbel o
Prr e t
the

classificntion

Downloaded by Priyanka Gaikwad ([email protected])

 lOMoARcPSD|34962087

Customer
all eleedvonit
clam-hebeled Training Tiples rom

9 Olbase

s t. dent
cvedt class b5aGompuli

Incomc
RIO

faiv
o n t h
high exellent

g o n t h
high fair

middle.agcd nO
foir
3 medium
tair
S e n i o v

Low
e x e l e n t

S e n i o d

e x t e l l e n t

Lo
Seniod

iddle-age Lo
lo fai
medium a i r

Sonbh
oh

c Xc e l e n t E
mediun

Seniod

eXe l e n t
mediu
Onth

ediu aiv

2 high E x d h e n e

13
hiddle-0ge medium

Seniot
4 t the clhs5.

ire tupls bconging

Thee
tthe css
buys-compley
A V e p l %
belon
belons

T h eb
r e- t
ar e
disti valsiysna
ompl no

has
- tompley
The class lbe
Lef =) bp-compuwlar
byp-tompmls a no
2
Downloaded by Priyanka Gaikwad ([email protected])
 lOMoARcPSD|34962087

The tple weishto classity 5

sldend= yes, cndk vlny hid

X =
(age yath,
= omemedium,

eneed to moximig PxlC:) P(c:), for i Can

be
Cach ClnsS,
Pc) is the pioy probabilhly of
bsed on the traimi tpleS:
COmpulcd
o-64
P (buys-com pulay ye5) 14
P(bgs-compuly no) S - o35
theollowig
the o l o 7
2 e tompfe
0compulE
Plxle;) or i
Conditional probabilities

o22
Plage=ynathI bus-com puJEs yes)
bugs_compul
no) 0'6

lage youth/
1 o 44/
=
o41/

medium l bugs-com pul

=
yes)
P(intom ece
no) o
bwys-compuy=
Plineome med um
bugs-compuli ) 6
P(sldent= yes
ugs-tomp wk n
Plsladent= y»l
bvapcompa =ya)
Plcveht.raingfai
bugs-compufy no)
z

Plevedit-vaing faiv| =

obtain
these prebabik bies, ve

VSin
yonth/ bmgp-tomplä =yeo)
Plye=
P[XI bgs-cohpulir ys) b=y) x P[st= ylb=y)
x Pli m

x Ple f l =y)

o-6 xo 6 : o D44
o 222 X oh X

Downloaded by Priyanka Gaikwad ([email protected])

 lOMoARcPSD|34962087

Stmilarly, oVX02x0
o6x
no)
X bugs-compuläy
P(X/c)plci),ae
ue tompule
P(XIci)lCi),
that cs, ma my.
P(CGx)mbe 643
TO
ind
the class , o.04yxo-63
o04Yxo

=yes)
=ye5)
Compnly
ly
Plbuyscomp
Plbuys 0 026

P(xlbys-comp.a
es)
O 0 19X o 3 5

z
no) =

0007
bugs-Lompulay

no) P(
buys-compnfe
y
s

PXI Sscompula

prediclo
elasifier
ian

Threys
the naive
bay
X
for
the iven tuple

Rule Based classitication

ules.
uses
then
classifiers
Rude bascd aSed o r c l a s s i f i c a t i o n .

nles ac
How Such
classificatiom
Rules oy nfosmatiom o b6
USing -Then epesenting

Rules art a
ood
clasificalian
ules clasificatin.

ules od
Tf-Then
Kmaale dy clas5ifres
asesSet
nses
set of
of -Then
A le based
of
oT
the
the form
is an
cpYsSi on
cxpvasion
ule
An I-Then
Conclusion
Condikion
THEN
IF
R, THEN
THEN Compus yes
ssCompbs
is ale
An c9 AVO slident =
yes
ayoh prc Co nd:l
anbcedento4
RI:IF antzcedent 4 pre
s ca lled ke
called
l e

p the e cosequen
Then" par(o vight side)
morc
the Condtion is
cono onc or

In the nle antcedent,

thd
tha avt
av
tesb legi- age= yatk
and
sla dent1»)
altvib.le
Logcally ANDedDownloaded by Priyanka Gaikwad ([email protected])
 lOMoARcPSD|34962087

NatNe Ba
sheu. ld
then the player
15 Sum,
Test the weathes

0D not ?
pl
dalaset
ont look
play

Ralmy

Sumy

oveacast

3
ove KCasst

No

Sv
O v e r C a s t

NO
Raimy NO
S my

S NO

12 Rai
OVer Cs P(G)- P ) P(G)
13
oVercst
Plx)

plis Sume
ll tuplos

weather i,
Ci . Not p

0 0H 029
No
play iy 029
P(Cl pley
o7

Downloaded by Priyanka Gaikwad ([email protected])

 lOMoARcPSD|34962087

MMM
ueAROu JnOT9 SAXVMLIOSIAA

3 = o-3

P/outlooK = Sumy
10y
P(I,)
P(Yes) o t =
o35
(T
P(Sm .5

Plotlook -5wnny( playmo)

PNO) 0 29

P(Xlc) = 06 o
o 3x o 71
Yes 5uny)
35

o 21X05 041
PNoSun 0:35

Pl Yes Can

Highof
the a boVC
that player
we Sce vo
0 aS

the anc
Sunn dag
play
35, 25
Dalsel stolen Co, 13,32,15,1,
0igin
No colo TYP Yes
Downloaded by Priyanka Gaikwad ([email protected])
I854, 3 , 5 ,
Doms & ns4
 lOMoARcPSD|34962087

The ules conseqnent contains a cle5s prcdiction (in this

Case, we are predictin whether custo me llby Lompule)

RI can albo be itlen a

(stadent yes) {buy scompulir s )
RI:lage = yeuth) holds tre J
a

a nle anle cedent

the con d. tion is stshed
the ule anle ecdent
that
Say
Accr
e

gVen lcple, an d

be a[Sessedy
cove
set, D le1
ovns
e R Can
dala
A taseled
lnss
Civen pleX yrom
R.
coveved by
be the ne of tplks clessibied
cl-ss:hie d by
y **

be the no o tples coweetly

coweet
covYcel
ples in .

-IDI be the no

and aCCurag
R aS

the coverage
wL can detine lu
hoveYS oveS stst)t
CoveYage [R)
=
a l e anleueden

DI fhok

tuples
ouurocylK) = coYYect omt
0 6oh antle d n t

r l e Cosennt-
nLoNeS th ar
the

is the 7 tuplks
the
l s coveYage hold tre
true o Te
Th is
ttyibwe vakes
vales
hold
theio attribulë
b, the nle (
co Voved

w anle cedent). that it covers

look t the t»pls

we
e

For a
ules accuracy

conly claSsify
Coweely cl»ssif
them
them the nle cen Can

what 7
and s4 Catomer d
dkbae,
kbae
ekeron
Cstomer

ddasase uill by a

All Cutomer
for the
a
whelr

OuY l a k s to pvedik
the
ytplo:
Compuer
Rwheh cove
Cosider h 19 20
2 3 S6913 5
3
36 31
Downloaded by Priyanka Gaikwad ([email protected])
33 3
 lOMoARcPSD|34962087

It Con
Covctly classihy both tuplo
Coveyage (RI)= tM 287.

ACcevacy (Ri) 00.

de-bnsed ahs:keatan to
hets See how wC Can

pre dic the class bel dor agiven pleX

is Said
to be tiggoed
ane
3 Saloficd by x, the ne
haVc
For 2SPpose
we

Cedtvaling
=
fas)
fair)
sladent ys,=

imcomen medium,
=
lage =y uth,
X accovding to buys-Com puley
wold LKe to classity
we
the
-

thene
XSaloges R which tvigzers
then the efires by
the ony e sa ked
R s
diclon j» K
clss pre
the
rekrning tonfket veSolution
need
mle ave tri9gered,
than one
more

trig2oing nals that

tha
Sige ondexivg to
o the iqyering
the

the
the hiyhest prio the most
attri bJe
atti bule tesb)
lob)
AsSign
AS5iq (witk
ment
has the toughst

class based ox deig nisc lassiication cost

prcYaene ov
decrensin? ovdey veer

pex cl5s

>Rale based ovdeing

ino onelong pri
prioai5 kst, aceovdin
0iist, accon

rles are oanged

l aliy o xpe
pes
to Some
mcaure
ay

Downloaded by Priyanka Gaikwad ([email protected])

 lOMoARcPSD|34962087

Decis.on Tsee duetion

Decsion liee algn IO3
s
desgned KroUn as

CART (c la sei fi cation and Regye ssion Twe s) deseribed the

generation of binay dec Si0 n vres

C45 abo decision Gee lgn

CART, I03 C5 adc
top doun approeches
The algm
contain
3pasameles s
)Athibwelist
3) Atyib.te-selection_melhod

Ois a dalz parlLon

lst aftyibutes deseyibing the taples
attriblekst is oya
tos
hcuviste procedre
Atlibeseselection-mehod Speahs a

thad bet" disnmin ats the

the iven fp les
given tles
Selecting the atyil
aclovding to class.
o m a mgain oo Gini rdx
whtS

the
as the
tst
test condla
Condlrn
as
t the fcatre
best featse
ut nodes
each stogc(node), pick linleynal
(inleynd
Dt
poss,be
Be outcana
oama
ben
the node int the
into Condluons
C ondilhnk
have
huve
)Na Spl tall
the lst

3) Repeat the above steps

nasted mt led
dts
stawing let Conhen
the
How to pick
1tCan b Enfsopy, yon.nJa-n
Can be oblamed
n-decson lsus
Downloaded by Priyanka Gaikwad ([email protected])
 lOMoARcPSD|34962087

Besic Ag
S
baSit Alm jo3
decision tee ton udhen
The decision tves in a lepdown
that Cowstc
e d agn manne

ecusive
divide& Conques
adl tobl
dala, ue)
set D of dassifica un
a) Given a
tyaining
clas atlyible
distinguished
ith o
ioncd into Smdler

Set 15
vecwys
vely paxh built
This tdaining
3) the l5te isbeimg
(daltables)
Swbsebb all
node lroot) depes ening
STARTS as a Single
Toce
dset D (smplea)
taI D
attyible from
5 w e choose o
oot
tbibJe
SPLIT
G)ts
called a vale defimed
eath
a r

branch is cxcate d fo islabeled by 5

A
attyibe and
the node olcoxdr
D abc pastliened
Samp les
Val-es and the
t otn
whus the Same roLun e CuYsively
8ihe Ag each paen
deosion Lte
node, it necd
node, net be
necd not
Yed t a
occus
atvibde hs
an
Oncee descendmts
th hod
othey o
Conbidetd

SToPS on
on when
he any one
Thc 3cluySive pashlning

ollawing Condlenbs yne

ayt
a
he Same cloas
n pasther
l l the Sample
then th rode be comes the ler seld t ltk.t clam
wwh dlz
ats ib
may bevty pstl Downloaded by Priyanka Gaikwad ([email protected])
 lOMoARcPSD|34962087

Month
Tokal

Class-Lobeled Tvaiming
,
diming Tuples
leples em
Am twth Alelecironil
RID Sn tome stdent md:t. alng cle sss
cle

Bus-emp-

no
yoth high fais
inelea
yoth high
no
Seniod mediuh
Lo
oiD
Senio
Excellent
Senio s Lo y txellent
nidle-ae mO
dai
mediu
Low

medit
Senib tKte lent
medium

eXte
medium
12
midsten
13 iddle-a ged high ex celles
medium
Seno
4

atn bwe sekahen

as
a
ws
intovmahm g
ID3

meas n e is choser
wik te highs
iomhon 9
The t r i b N.
rode N
ode
tyila fo
Tlyalë fo im Dis gmn
Splitimg
Tededto derig tnpe
a
ab the
T h ee x p e d i o m p h k n

JofolD) 2 l,P)
an
ar mbia
asbiyar

the rore pobabi k th

where Pis
umalid by
and is
D belonyp t
tein eno ded m bib
log iS widbeCaur the i/omlon
InfolD) is abo
Downloaded by Priyanka knaon
Gaikwad as trioP
([email protected])
 lOMoARcPSD|34962087

TOtal

(o) jo(o,)

nfolo) og,m S

0 9 40

is
dened the drene bton th orgn.!
Jnyomln gan
cquivement
D s t a n o

and ne
Jomalien eqivement D t
Iyolb) Jyo, (o)
-

Grain(A)

Vao,b) =

a ) +(
x
o694 6
lo)- poal o) -(o-6412) (-01)
Gaimlage)=
Jyo
024 bib
o o 94-o69 = - 0 35(-
0
Compe Gain (Income)
2- 015
we Can

Info
(D) x
In com e

0.9

09-0.9
0-0296.b
hain(Inone)
0:15
aim (sdet) =

0048
hain( Cye kknt) =

Downloaded by Priyanka Gaikwad ([email protected])

 lOMoARcPSD|34962087

st
spliry on Ag
Ag Senio
3 0

iddle

3
2-

on
slde
based
dvde
So, mow

C v e e h t - v l i

st.ak EXeelle
nidle-odfai

L30

310 Credt

slident
Xel

Downloaded by Priyanka Gaikwad ([email protected])

 lOMoARcPSD|34962087

Downloaded by Priyanka Gaikwad ([email protected])

 lOMoARcPSD|34962087

Downloaded by Priyanka Gaikwad ([email protected])

 lOMoARcPSD|34962087

Downloaded by Priyanka Gaikwad ([email protected])