Unit No: 3

Subject: Class Notes Lecture No:

Faculty: Link to Session
Topic: Planner (SP):S.No...of SP
Book Reference:
Date Conducted:
Page No: |
Unt-3

bTh claulicaton algoithy a a supenvisd lerning

echique that is wed to iolurtty te a

hew cbemadtons on the bois

-frniring data
bh ellicaten, a prgam leams fro the qien
dataset on) obserhons and then clasitis
obsenators oto ember
a dases
Sueh No, oQ) 1, cat or and ete.
Yu
n clasitiakn algonthm, a discrete output tunchin
is mappey to inpt variable Co).

short, clasttcaion is atom Pattern

Reungriton. Here claviiatten
dhta the Same patern in ctuture data.
leames n clasiicalion problemd.
2) Enge kone%s.
 Subjec
Facul

) Lay leanned: 4 iost stres the Topi

forr the teat dat set

dataset blore clastlicten
omte. uhen using a
a
lay learmer, the
h tránng dataset mect
pppiat dati: me
but more
Spet on predichons.
b kn: Case-base rasonia KNa algoith.
|a) kager leamer! Before obtairnig a text dataset,
eager eagnens buslt
datoet.
-hyspent more dine shudying and les time for
prediction.
bays algoith, Decision tree
mL clasili ation tiqoatlhns
) LinaY Mode: T isa Staftical pproach uyc &to
dycribe the relatiuychip ao a depndet Cnd oe or
independnt vasiale .
he
YPeulane thd ini nàeth
t lww the predited & ckual valy a
dependent valuep
 Unit No: 3
Subject: Class Notes -Lecture No:
Faculty:
Topic: Link to Session
Planner (SP):SNo...ofsP
Book Reference:
model. Date Conducted:
Page No:
) Non- linear model! Ex: non-int

docgt not Linca

bo te depondent vanite dindepundunt vain
: Logistie Rgreotin, Neural wo.

Muidass labls are

rteredas mubiclas claihication.
maltelas cassification, tasku are tquently modeted,
a. mode hat ore casts a mut
probability etoibuton tor cach emample.
casitcatio) mearg Cadegori data auttg
groupr bayed the Qinilaitiy. na datuct
the indeponolnt OY

lhe data.
 Subjec

When e tally ot muLtilas assiat Facul

Top

Qe nave more than Our dependet

vostaje, a ont seen in telo
A A
A A A

Ex: mutelaus otsutoaton

, wve Bayu, Deiciy tsu, sym, Randon rt
 Unit No: 3
Class Notes Lecture No:
Link to Session
Planner (SP)s Mo.
Book Reference:
Date Conducted:
Page No: 2

* Nleural Networks
simila
Neural netwoks proceses intomati on in a
the does a rd -these netork
human brain does
actually lears from enmles, which ie
is

Aontt need tob provide al the intmien qaing

ny pec task. So that was the monreas0n
neua netoork foto emisterce.
hy
3
neural network i o three main layens ohich
are

DCoput layer? Accepts all the Aoput that are

provided by the
9 rdden layerr Lo beteen the the 'nput nd oupat
there tsis a set t hidden ayer 7) which
computatios are pertorne that further qulu
in the output
i) Dutput e the input lhyer
Seies ob transtomations, ohil
the htden laye, it result in olriAelived by plhy.
 Subjec

Facul

# Multt Layer Neural Netorka

Top

cnnected mult toyer nYa netaonk

ully
muilaer perceptn(mLp).
ke a Singeager preptn madel, a mulklayer
pceptn model also ha the sae model structure
but has
arater ounber f hiden lager.
hidden layr with
4 unitt 4jas it
3 input unts plus (d=4+)
bias wnt (m=3+)
output layer ith s
oubp nts (t-3).

No ot layers Le3.
that
LConeet 4st nn bías bas neon

i (hidden
(ou~put layeo]
has 3 ayes tndutng
has More than 1 hiden ayer. t is caled as
dup ANN. An mp u a
teedtod arifiaa
ificial neura) netooTk .
 Unit No:3
Subject: Class Notes Lecture No:
Faculty: Link to Session
Topic: Planner (SP):s No....dSP
Book Reference:
Date Conducted:
Page No:
4
h number o layes and the nunber ewrs
leret as hyper parneten t a neural
etot any hese ecd leaning
-must be sed to ioa
Coss- validaion techniques
ideal valwe tor these.
the is don via
wocigt adjustboert triring
Deepr neiral netorks
back prpagaion. De«pr ar

better at prceing dota.

toweneY, deepen layes Can lead to
qradient pblens Speiad algonithas are

to solve this issue

a,Tin
ain a cin]

anCin Xtin)
a,cin)
a,ch bias uit nd iu 1.
 Subject

Facult

vies o the muthayer prurty Top

shous Connected three

here, his Tmag etwork
a

oith 3 înput neuen

neura
and 3 output neurons.

output

stntre o Neural Netoort.

'mde
W
lahuork)ybutput)

piologica! nlau netsnk prentt tpt

fo ANN, Cely hueclear ereentt naes, Synpse
epreents ocight represen ouput.
 Unit No: 3
Subject: Class Notes Lecture No:
Faculty: Link to Session
Topie: Planner (SP)s Nn.. ofs2
Book Reference:
Date Conducted:
Page No: S

Percaptn
an ANN unit that helps t detct
Peyceptn i
Centain inpu data computations înn busines frliqurce.
Percertn u also tratd as
Simple tys ot ANN. owever, t s a supevisy
algnthn of inag clautlier
laing oside ft as a
Herce, we
neural netoork Ûth eur main paramets, ie
foput vaes, weigbs & Bias, net sum, &activatin turct.
Basic cmponen Percepten
There age thrree ain conpenLnt
Loput
Net inp nchivatio)
tuncion fnhen

(X
 Subye

F a

Iut Node (on) Ipu

which 4he fnitia! dta int
acoph 4he
fr Junthey proreswing cach input ne tertoi
eat numyical nlu0

Weight pamnmeter nopretenhwni4ethe hicchrergth ot

Hhe Cmnec ti cn of buteen is anothr

important panametes od perceplrn eomfprnts

|eight Aectiy
rptiona) to the
output Funther
he input neurn in
Bias can be decitng
considered al the li4 o irtercpt
a
lincan quaien
ndivati on fuetion :
the tin &impntat tenfernk
that helps to detegnine whettey the
t e on not Actvatio furetin
skp tu ton
-Artnaton fuehons

kp funetbon
 Unit No. 3
Subject: Class Notes Lecture No.
Faculty: Link to Sesion
Topic: Planner (SPAa d19
Book Reference:
Date Conducted:
Page No.
F Back Brpaatin 6

7he back an

the

hiddenThe layer.neumon pertrms a weighted Gum upon the

the actvation
Toput lyr, uhich ü then ued ay
funeton as fnput, eipeialliy
funchon
supeviad learing t
Heach the netirk. t wpat1 the
constanty upt
weights o the netonk uat) thi desived output
s met by the etworh.
t ineude the folouáng tactx that Telp:nsit
-ble, t the pformance d the eiwetk
oang
Random value o weight
* Numbe d hiden ne u r S

the taaining set

* Teachtng pinmetin
poameter aluer Ceuniy nte)
 Subjeu

Fad

Need o Bnck prgin

impenent

other pannmolx to perrfurn trraintng

haue kind d prierr knuldq.
So, it tende ut do be mre
tlzible
a
standad meho) hat ruull! well
woking d Back progmtion
input
layers

olrlayer

1. 7he prye connectep mhy nnn he input x

Then the weight w Tandomy selectec
whtch arse wted to mod +he input
 Unit No: 2
Subject: Class Notes Lecture No:
Faculty: Link to Session
Topic: Planner (SP)s o ntsP
Book Reference:
Dale Conducted:
Page No: 1
3- n{ten, then, the output i caluulattd tr eny
individu neurnn that pauee fron the irp
to the kiten layer and hen to the
hy
ouput layo.
Laty, the are evaluated in the
oubut
Erre Ackuad output - Dutred output
Sent back to the hitden
S. 1he esmos, are
the output layr
the to lesser the

6. Untl he desired sult i achieved,

koap Herating the tucess

 Subject

Facuit

# Support Vactor Machine (svM) # opi

of tha mot
pSuppot Vect mahin (Sum)
algrithm, which is wsod
popular pervitod laortriog "grsion prchlarns
tor classilfcation as wo

Mowever, paimry ctaslicen

mmchine
chocces -the entene poin k/veca that help
the hyrplane These Lntrrec
are calle? Support vects, and hence
fs termed Suppont Vecton machine:
the qoa o tho svm ts to create he best line
(o) decislon bouga that Can

n- dimensio space into clayes, o that

ently put the new data poiot in the coTeet
fo the tuture
H This decision bounta is caled the

Manim >povhve hypenplane

Do
Suppost veckox
 Unit No: iL
Subject: Class Notes Lecture No:
Faculty: Link to Session
Topic: Planner (SP):S No of SP
Book Reference:
Date Conducted:
Page No: 2
kmanjk!
Cat ha also
havo cce
Suprote
has Some
a
atuyaiely uri
ohege el is a a cat trr So, such
createf
be
oill
by
siot
uig
in
So that Con loasn about aheysnt
eatures cats and dogs,
this
So, al Sipport vecty
vectr Coeates decísirn
borda
choosee n e
betwecn these doo data ( cat d do) an
cto). et l See 4he th eattmma
Cases ( Support e
CAse c cat ad doqs, On the bais of Suprert vechos,
Cat.
new dak it
Cat a Cat

(pait labales modet Prlichion

data). Arininy
 Subje

Wwed lo "face dolechen F a

alqunilhn
chuilicadion, tat catrgisndim "ete
er plane & Suyport Voctox o sVm :

There Can b
Hhe casses îo -dimenciona space, but
Hir out he but decicion boundaries hat hop
dhta póos. his
T5ie best bouray
Knon as the (Vm.

Dimension of the hypesplane deperd on the

eatue ruent fo datnset, kch mears f thee
are 2teturey, then hiper plane wil be
line an thee are 3 featunes, thern
be Aoo- Aimensiona! plane
a

We alnyi cmeate a hyprplane that ha a

maimun margin, wkch mean the Manimum distanc
betucen Ahe dalapoiote
* Support Vectr:
The aapoint ve chos that ane the clat te
hyeaplane ang wtich lect he posifion
the hypopkane Sppont veckoy
 Unit No: 3
Subject: Class Notes Lecture No:
Faculty: Link to Session
Topic: Planner (SP)s Mo ofsP
Book Reference:
Date Conducted:
Page No:

) LioeaY <Vm

limy ccpevn deHe data, which

chassc
means, a dtasel corn bo clessiliod frto to
Rne, thn
toe, such dhata
by ung Cingle stnight
Hegme Ineanly seporade ctositien is caad
clasctier
|lineanly sepoade
A
.yperplane
A
A

a) Non-linea SVm

Groóng bejony linsmnity

No,línear
|-cd, data, obich meanS a dataset cannct
clasibied staight ine, then Such data is
Non línean dta
d lassilies
terned
caled Non lioean CVm ctasiig.6
 S u b j e c

Facu

lorking Tanged, then sapat

To

sraght Kne, but tor noninear

data, we cannot dro ingle straigbt lioe,
Cosider the blo image.

So, to seperate thesn data poink, we need to acd ou

dmensions. for (oear dota, oe bave wed
dimesions non- linear data, we
Hhivd lan be caleulatod
as,

So, now SVm wir divide the dataaets classe

athe follouing A AA
eonsider the
A

Cast hyplane.
 Unit No: 3
Subject: Class Notes Lecture No:
Faculty: Link to Session
Topic: Planner (SP):s.No..ofSP
Book Reference:
Date Conducted:
Page No:

hence ft is looking
-’ Since
Sin we in 3d Space, locking
to the
1Rke plane paallel become s
t io d spaca usth 2-l,then

A A
Bet hyperplane.
A A