MULTIDIMENSIONAL

ANALYSTS
Knowleda
kDD (kaoaledge
Discouty
Biscovery in Database )
js aprocess Hat involveshe extractioh
of useful unkaoos ànd potenially
previously
valuable intormations om lazge data sets

Data is Ah pssetal sdep of Kaswledge

Mining
Discavery
List of steps in Knsualedge biscovesy
Process

Data Extaction Data is exta cted om

various dataàses

• Data Cleaning: Data cleaning is ceined as

removal of hoisy and irkelevant
data from collechon

Daa Tatqaion: Data inegrahioa defned as

heteogeaeot data fom

muhiple SOyrces (onbined jn SoUhce
 done by
Cbataarehouse) Data indeqration is

LLLLL tools, data miarahon tol

data intequation
using tools
data' sunchhonizhon

dehined as
•• Data Selecion: Dat selecion is

where dati e levahtto

the Drocess
tom the data
Ahd retieve
analysis is decided
colle cion

ahsfor bmaton is deßned

• Data Transfornàhon: Data
te process of hanstorming

data ibto iae torn veGuired by mininc

apprspri
procedure. lay petozming summay
operations:

Daa Mining Data Mining is de fined as

intelli aent mettods techhigyes thad

applied to exht pattens which

potatially useful

Palern evaluahon is defned

• Patten Evaluation:
ideschifyina, stity
pattens represehn knoale e
increasing
based meas UreR

T tnds interestingness scDre eachpattehs

uses Summahi Zaion Aad wis ualizahon to

ma ke data un derstandable by
 Difference betueen knowledge distovery
and Data Mining

Kns aledge presentaton Tkis ihvolves Dresehhing

he eults in that
be used to make decisions

Inte pretahon/
Evalyathon Knowle dge

Data
Mining

APatterns

|Transfomation

Trànstor med

Preprocessing Data

Preprocessed
!
Data
Selechon

Ta-geti
Data

Data

Diagamiaàhe Representaton of
Knaaledge
in Databases
Disovery
 Applicathon

Archite cture

Data
Mining

BIG Data Mining Extra ct valuable

DATA
insights /
infomation
Intelligent Algorithms/
MechaniShn

Custome Data
Valuable Insights
Purchase
Freqyent buyer
Purchased item Data Mining Occassional buyer
Buying Fequenty Bargain buyer
No of items

Decision Making
Strategie Plannina
Sales Promotion

Pricing Stategy

Need of Data Mining

Handling Lage Data. Volumes

•Tmproved Decision Making

• Compethhe Advàntage

Tdenhying Hidde, Pattekns

• Cost Reducion
 User Intefa ce

Pattetn Ev aluatio
Descriphve

PredicivekData MininA EngineKnoaledge Base

Task
Data w3e House Serve

Da ta cleaning Data
Transformation, Data
Thtegration, Data
Selection

Database Dataw dehoyse 0+he Repositoies

Process f Data Mining

Data Mining process iny olves seueral

process hupically inuolues key
steps of4referredo s the CRISP -DM,
which stands for Cross Thdusty stàndard
Process for Data
Mining

Follocinq is the breakdowh

stage, of
Crisp "data mhing

ndestandigg RusineSS

)
() Obiechve Defhiniion 1
 Tderhfy the bu siness problem or
oppor-tunity
oProject Plan:

•Outline
tke data mihi,q aoals and approà ch
2. Data
Understanding
) bata Collection: Gaatker data
data fom
from various SOurces.

ü) Data Explorathon: Analyze te data to understand

it huctre
content quality àhd

(iji) Datà Desciption: Summabixe te data chakateisic

including aay peelimi hary insighi

3. Data Prepakaion

i) Data Cleaninq Hasdle missing values, re mave

duplicates an coect ertorS
6) Data Thansfohthion: Nomalize àggregate or
Convet dataY as nee ded

i) Featue Selechon: Choose relev

relev at athibute that
will help ih
modeling

4 Modeling
 )Model Selecton Choose
Ch 00se Dpkopriate algorithm
(Classithicaion, Clusteing

) Model Trainina Use the prepaved data to

train the model

i) Model Evaluation: Assess the tmodel

perfo nàh ce uSing matrices
(Accura Precisioh et)
5. Evalyation.

i)) Revie u Result : Co Dare te model outones

stt business obiecthues
-àgains

i) Desision Making Determin e if the model is

tegdy fo
deployment
equites adjustment

6 Deplaynent
i) Tmsle,ehtatioh: Tategbate the model inta tte
busiess proceSs

G) Monitaring (ontinMbusly tack the models

penfok ma make necessa
ypdates.
 Maintenànce

()
w Peiodic evaluate the model
Revieo i
Keguaa
dta to ensre
continued
relevance ahd accuray

Conclu sionh This stucted approach helps

ensureS he data mining projects
effecive relevant ah d alighith bus
goals.

Data Tasks / Functions

Mihing

Data Mining cdeals with

ohat kiad of patterns
Cah be ined

basis of kihd of data t be mined

HRere are two kinds functions. ihvalue d in
te process A data mining

Pediciye Funcion /Task Descigtive Funchioa /Task

Objective: Predictive Ohjective: Desciphye tuncton

tasks work with the

are
executed withhe obechive to suh mahi'ze
Lobjective of axd ànalyze Aistoical

fösecastig future
fut data
events tends

Output: Predichye mo del, Outp uti Genekates the

 Difference between
bet $upervised and Unsupevise d

Learning

probabilitie output like pattern S

clusterS. rules and
visualiz ahions

Example: Classiticathon, Example, Clysteihaasociaton!

keqrtssion, tine axalysis, namaly
series aaly sis deducon Sunmarizato

Ue: Predicive is use Use: Desiptive is used

for sales forecast fo market analysis.

Descaiphue Data Mining Techniques

Assnciaioh Analysis t dis Covers a pa ttern

relationship betaeea
vàriables data set
large

Example: People who formula milk also buu

liapers

2 Cluste AnalysisiT Ainds group of closely

helated sbsevations

•It is based Uns upeuised learnihg

Clustering
based Smila chaacteistice
is

Example stedent based

Custering
 Aqoithm includes:

WK-Means Clustering
Pastition data ito
k- distinct clustes
G) Hierakchica) Clusterina: Create a tee of clusters
based On simi lariies

3_Anomaly Detection (Outlie,): Idexities

observaions whose
characeristics are different om rest of he
data

used detection:

4. Sum marix aton: Short coh c lusion of large

data

Example: Total amout Total ites

5. Text Mining Tavolues extrachag insights fran

hraa
dextal data troah technique
lilke:

Sentiment Analysis
Topic Mode ling
Extty Recogmition

Predichve Techniq4eJ

. Classifeation : Tt inv o lueg

aigaing
idems
 dataset cto taraet classes.
categoe.

discreate inite tàrget value

Example.Peson buy a paricula book

bosk o not

Classiticatione do mathe maical grouping

Techniques include :
(Decision Tee: Tree like model that solit data
based feature Valyes

Neural NetworKs: Dsed fo complex patern

zecagnition specially is deep
leaning

Regessioni T+ orks corinyoys taaet.

vable s

Example: Price of he book

Technique inclvde

0 Linea Regessioh : betwee

Models relatios hip
dep endent vaiable
more indlependet va bia blea.

Reqsessioni D is sed fox made ing

he probabi lity of a
certin class event

3.ime Series Analysis 4 inv olves methods

time

ordered data asonal

to identfy treAds
heads.
patterns ànd
ànd cyc lic beAaviour

Examplei ECG matches to hormal values

 AsSIG NMENT-IV

Difference betoeen knauledge discovey

Data Mining

Knole dgeDiscovey
0 Encompasses tte enthre process of exta cting
insights fam data
meaningful

i) Tacludes steps data ceahn

ike cleaning, intearation,
selechon trànsformatiDn minina evauaton.
ovahation Sd
Cin) Aired at tanshrmn data into usel
Kaauledge
(iv) Lnvolves interdiscipinar methocds such as

staistics AT and taàchine earning

( Ends with kaouledqe presentation where

ormat
are Sharedin undestandable
insights

Data Mining

)
G) A fo Cysed step
ocess.
within the kna wledqe
Kno
discovery pr

G) DHlixed algprithns to find patten sS, têads

relationships inda ta

G) Ofleh emplo4s machine learning. statistical

models clusteeinq techaques
(iv) Data Minin a ou tput pyovide acionable
patterns preolicticns

() Helps decision -making hy uncavering

hidde
patterns. and aSSociaion.

Advantaaes ànd Disadvahtaqes issues of

Data ining

of Datà
Data
Advantages Minig
i) Pattern Dis Covery. Reveals hidden patterns and

kelationships in Large data set

(i) Tnp Dv ed Decisibn - Making:Aids bysintss ec

)
data-drive,
de cisions naking

Gii) Predichve Capabilhes: Anticipa te tr ends

futre outtomes:

EPhiençy Automatea data analysis Savihg

ime resour ces

) Personaliz athion: Ehhares cystome expejences

hrotak tage ted màketing

Disadvantaqec /Tssuee Data Mining

cu Conce bns: PesOnal data usaqe

Pivay ai
 ethicalL issues
privay
be unre liable if
ü) Data Quality: Results
the data indccurate
in comp lete

(i) Camplexity: Requires specialized kaswledge

ànd ools hich Can be cos+lu

(iw ecuity Risks Lage ataset are vulnerable

to rpaches and mis uSe.

Ouenhtkinq: Models become too tailored

to specifi daba limiting
entralizabilih
* Diference between Sunervised and Uhsupe wised

Leanig
Supervised Leaning

()Uses labeled data cwith input - output paiks

(i)rais the model t predit specifhc

outomes

usedl for classihcaton nd

eg besiag
tasks

(jw Examples Spam image

dekction imàqe recagnition
 Dasupervised learning

Uala beled da tàwith predehne d

) Taetifies pattens clu steS structure in

olata

used to dnd dimensionality

used clusein
reductio

iv) Exa mple Customer seamentation, anomà ly

dedection