M T W TF SS

Page No.:

VECISION TREE CLASSIFIER Date:

4||23
YOUVA

> Gupervised learnfnqitechiqw that be used fos

both ctacsific atio but i t mostly
prefemed for Solvin claccification pro bems.
problem Gtuement
yay tIOoK Tempertre aind Humid playTenois
HOt ueat HO

HOt
Higb Ho
Hgh
Overaut HOt tléak Ye
Kain Weak Hormay yes
5 Kafn Coo1 Homa yes
6 Kafn Coo1 Stong Hona Ho
Overco0t 1COo| 9Hiqh d yes
1faeak Horma1o
eak NOrma Yes
Kaln MId (4eak HOrmay
12 QveScaut High tr ye
13 Oyercaut stoong HormayS
14 Kaio HigS
>7iagram 1

9-yes 5-N0
:Outlook'

es]2-NO

Overraut

tyes ONO
príty sptit check
Wbether optit pure in pure

Entsopy ) meaje of
trnpurit
Gini Imprity
 MT T F S S
there Page No.:
YOUVA
D-yg 3-NO Date:

9yes|s-NO1
Jmpure spfit Outlook

4-yesCO-NO

Overrau) Kalo

fure spitt onecategory

is: there tha 13
Jeaf NOde
4-ye9.
2.) klhfeature nee selet Stat tes
Iofoimatfon
) PURITY eHECK
> Conoiderbirary classifica ior

H CS) -p + loqoPtP- o9,

probabi 0tgf post atgory
probabíty of

1
 M T WT F S S

Page No.:
YOUVA
Date:

G-yeg g-NO
()
leaf node

Pt: 3-yes
3-yes t g-NO probabitity p Of
post
B-NO 3
3-es t3-NO 6 2

H CC) = -3
6 6

entrops of HCc) = 1 :(Conmptetey (mpre

need further
dinstoo get
Dode

HCa) lo9, Pt -P - J092 P

there 3-yes
cateqory
pt = 313 p- =
-NO

13 3
pure Spi
po!
(e if probabftityof PtjP
qet entsop4 of 1
 M T W

Page No
YOUVA
Date:

9-) Gin? Tmputity

I-PCp)

eompiete mpure Sptit

pure ptt
eotoop

o-S

pt,p
Consider elassifroio probie
3

entropy
HeS) =-Pej loqe Pe1 - Pca lo9 Pe2- pcs to9g Pe3
’ Gint entropy
GI + Cpo+ Cpg)
 Page No
YOUVA
Gate

feature Gelet Start

Gais (s, fi) HCs) - |9y| H (GY )

Ve vay Gf chtegoriy.
S’ Sampe Vataue
entoopy
fifeature f3
Sv’ Subsapie 9-es|S-NO - 14
H ’ entropy (e) HCe) = 0-94
s G yes |D-o
O-81
HCS ) = -Pt I09-Pt-P-1092 P

14

O-94

o-81

HCc2) 1 = tmpute Sptit

Hod, Gain CS, f) - HCs) -

HCS) = 0-94 6-948O-81 6

(Sv foc category

C2
8

|s| 4 Ga'o Cs, J" O049.

 M W

Page No.
YOUVA
Date:

Gao, f2) - o.0sI Gai (s,fy - 0.o49

nced to Start Splittfoq featre

entooPy
the datuet Sma Shod
entropy. Lioq formwa 4akes +fne
) beneve the datauet1 19 Buge Shoy
Gfo? entoopy simple math’ take (ess

wha if my feature {sContiooW!

Sort the fet ure
23 yes thoeshold Value

3.6 yes k2.3

NO

yes -yes 3-ND

Ho mny
yeo
than e qual
t0 29

1-yes,
Ftsadvantaq : ) we need Create decision tree fos
each and every thshold
TIme -0hen datue
buge.
 MT W
|Page No
YOUVA
Date

PosT PRUNHING AND PRE PgunMING’Keduco q

Conciderraîninq daa
overfiting,
f3

This
cpittng CootinQusly
going the Jeaf

node oCCerSy
Nheoever Spfit +be
decisio toeto1tS depth ti| lef dode
|eads overfiHing

AccCuray ubigh T i i 1
TeSt data accu o0.

Afm to get generauzed MOdel

reducn9 ov erfitoq ustng post prunnin

preprunnin
poST PRUMNING:
’ Cuttiog brancbe blod9
prunning means
Const entire decisiodtoee to Compree teaf
node.
decision
prunisg dataue: becauye t take
Suitabio forSrmouer
+fme COnSto et decisto9 foee.
2PRE PRUNMI G
HyperpQranter tor seet tbe best
pramete
 W F S S

Page No.
YOUVA
Date:

7ECISON TReEREGRESSORt
regrp Ssion problemGtaregent
be ConHDow.
> Congider he datset
Exp Career qap 9alary COutput ) - Cootinouy
yes
yes
52k
4
45 yes
y=SoK- Avg Slary (m ean
Start Splitinq Exp feature "
]threshol d \raueA

Lok42k ,52 k, 60k5 6IkJ

Exp <2

4K,52k,60k, 5GK
2) tseshold vaue 25
1

yes

40,42K

fioal am Calauate vartance edetio.

\ariane
error
C-yJ’(nean 9uare
error

’Varfance (ROot )
4o-so)°+ C43-s64Ce2-ro
+(60-so +(s6-SO)
 Page No
YOUVA
Dete

varlrool G08
/ariane Cref) Cko-s0

) Veurince C42 -s0)2 FCr2-so J 4 C60-s0

(sG-soj
L 64t 4 + + 36

reduton

\/arian eduuHo var (aoot) Z o? v r Cchfid

Go-8- * 00 4-4/ * s
/OriCnce reduCHOf iS ero

VOur ooo4 ) = 60-8

F 2

448c + too
Varince
redwO 60-8 X82 4666

VariOnce recu
Which Spu+ should

wh ose
vartance eduwton t bigb.