Data mining chapter for students of university

Data mining chapter

1. Data Mining

2. Prerequisite
Knowledgeof
DiscreteMathematics&ProbabilityTheory
DataStructuresand Algorithms
7

3. TeachingMaterials
TextBook
Datamining:ConceptsandTechniques,byJiaweiHanand
MichelineKamber,MorganKaufmann,ISBN1‐55860‐489‐8.
ReferenceBooks
DataMiningandAnalysis:FundamentalConceptsand
Algorithms,M.J.Zaki&WagnerM.Jr.,CambridgePress.
IntroductiontoDataMining,byPang‐NingTan,Michael
Steinbach,andVipinKumar,Pearson/AddisonWesley,ISBN0‐
321‐32136‐7.
MachineLearning,byTomM.Mitchell,McGraw‐Hill,ISBN0‐
07‐042807‐7
8

4. Topics
Introduction
DataPre‐processing
AssociationRuleMining
ClassificationTechniques(SupervisedLearning)
ClusteringTechniques(UnsupervisedLearning)
Semi‐SupervisedLearning
Applications
Socialnetworkanalysis
Opinion mining and sentimentanalysis
Recommendersystemsandcollaborativefiltering
9

5. Data
DataistheLatinpluralofdatum
Usedtorepresentunprocessedfactsandfigures
withoutanyaddedinterpretation oranalysis.
Generallyassociatedwithsomeentityandoftenviewed
asthelowestlevelofabstractionfromwhich
informationandknowledgearederived.
Datamaybeunstructured, semi‐structured,
and structured
Example:ThepriceofpetrolisRs.90perliter

6. Information
Informationisinterpreted(processed)datasothatithas
meaningfortheuser.
“ThepriceofpetrolhasrisenfromRs.70toRs.90per
liter”–isinformationforapersonwhotrackspetrol
prices.
Databecomesinformationwhenitisprocessedforsome
purposeandaddsvaluefortherecipient.
Asetofrawsalesfigures–Data
Salesreport(chartplotting,trendanalysis)–Information

7. Knowledge
Knowledgeisafluidmixofinformation,experienceand
insightthatmaybenefittheindividualorthe
organization.
“WhenpetrolpricesgoupbyRs.20perliter,itislikely
thatbusfarewillriseby25%"isknowledge.
Theboundariesbetweendata,information,and
knowledgeisfuzzy
Whatisdatatoonepersonisinformationtosomeone
else.

8. SummarizedView
Data–asindatabases,spreadsheets,textfiles…
Information–Processeddata
knowledge–Fluidmixofinformation,experience,and insight
OR,knowledgeisametainformationaboutthepatterns
hiddeninthedata
Thepatternsmustbediscoveredautomatically!!!

9. DataCategories&MiningTerminologies
DataarestoredinDocuments(Afile)
Unstructured Semi‐structured Structured
Afilestoredon
yourPC
(TextMining)
Awebpage
storedonWWW
(Web Mining)
Adatabase
(DataMining)

10. What is Data Mining?
AnimportantstepofKDD(Knowledge Discovery
from Databases)process
Dataminingistheautomaticextractionof
interesting knowledge(rules,regularities,
patterns,constraints) fromlargedatasources,
e.g.,databases,texts,web, images,etc.
Identifiedpatternsmustbe:
•Valid,novel(non‐trivial),potentiallyuseful, and
understandable

11. TheKDDProcess

12. DataMiningObjectives
Identificationofdataasasourceofusefulinformation
Automaticextractionofvalidandnovelpattersfromthe data
source
Useofdiscoveredpatternsforcompetitiveadvantages when
workinginbusinessenvironment

13. WhyData Mining?
DataExplosion(InformationOverload)problem
We aredrowningin data,but starvingforknowledge!
Datadataeverywherenoranydropofinsight!
(waterwatereverywherenor any drop todrink)
Explosivegrowthofdata:fromTerabytestoPetabytes
Automateddatacollectiontoolsandmature
database technologyleadtotremendousamounts
ofdatastored indatabases,datawarehouses,and
otherdata repositories

14. WhyDataMining?Cont…
Majorsourcesofabundantdata
Business:Web,e‐commerce,transactions,stocks,…
Science:Remotesensing,bioinformatics,scientificsimulation
Society and everyone:news,digitalcameras,
Thecomputingpowerisnotanissue.
Dataminingtoolsareavailable
Thecompetitivepressureisverystrong.
Almosteverycompanyisdoing(orhastodo)it

15. Why Data Mining Important?
Digitizationofbusinessesproducehugeamountofdata
Howtomakebestuseofdata?
Knowledgediscoveredfromdatacanbeusedforcompetitive
advantage.
E‐businessesaregeneratinghugeamountofdatasets
Online retailers(e.g.,amazon.com) arelargelydrivingby data
mining.
Websearch engines areinformationretrieval(textmining)
anddatamining companies

16. WhyisDataMiningNecessary?
Makeuseofyourdataassets
(knowledge‐based economy)
Biggapfromstoreddatatoknowledge
Transitionwon’toccur automatically.
Manyinterestingthingscan’tbefoundusing
database queries
Customerslikelytobuymyproducts?
Whysalewasdownafterdemonitization?
Whichitems should berecommendedtoaperson
purchasing computer?

19. Data Mining: On what kind of
data?
RelationalDatabases
DataWarehouses
TransactionalDatabases
AdvancedDBandDataRepositories
Object‐oriented and object‐relationaldatabases
Spatialdatabases
Time‐seriesdataand temporaldata
Textdatabasesand multimediadatabases
Heterogeneousand legacy databases
Webdatabase

20. Data Mining Functionalities:
Characterization(1)
Adataminingprocessaimstofindrulesthatdescribe the
propertiesofaconcept.
Standardform:
Ifconceptthencharacteristics
C=1A=1&B=3(Support:25%,i.e.,thereare25%
recordsforwhichtheruleistrue)
C=1A=1&B=4(Support:17%)
C=1A=0&B=2(Support:16%)

21. Data Mining Functionalities:
Discrimination(2)
Adataminingprocesswhichaimsistofindrulesthat
allowustodiscriminatetheobjects(records)
belonging toagivenconcept(oneclass)fromthe
restofrecords (classes)
Standardform:
If characteristicsthenconcept
A=0&B=1C=1(Support:33%,Confidence:83%)
Confidence:The conditional probabilityofthe concept given
the characteristics
A=2&B=0C=1(27%,80%)
A=1&B=1C=1(12%,76%)

22. Data Mining Functionalities:
Classificationand Prediction (3)
Findingmodels(rules)thatdescribe(characterize)
and/ordistinguish(discriminate)classesor
conceptsfor futureprediction.
Classifycountriesbasedonclimate(characteristics)
Classifycarsbasedongasmileageanduseittopredict
classificationofa newcar
Presentation:
DecisionTree
ClassificationRules
NeuralNetwork
Bayes Network

23. Data Mining Functionalities:
Prediction (statistical)(4)
ADataMiningprocesstopredictsomeunknownor missing
numericalvalues.
Outputspace:continuous

24. Data Mining Functionalities:
Association Analysis(5)
ADataMiningprocesswhichaimstoidentify
patterns (akafrequentitemsets)indata
Forexample:
Buy(X, Printer)Buy (X,Cartridge)
Buy (X,Bread)Buy (X, Butter)Buy (X,Milk)

25. Data Mining Functionalities:
Cluster Analysis(6)
Unsupervisedlearning
Aimstogroupdatatoformnewclasses
Clusterhouses tofind distributionpatterns
Basicprinciple:Maximizingtheintra‐class
similarityand minimizingtheinter‐class
similarity

26. Data Mining Functionalities:
Outlier Analysis(7)
Outlier:Adataobjectthatdoesnotcomplywiththe
generalbehaviorofthedata
Itcanbeconsideredasnoiseorexception,butisquite
usefulinfrauddetection,rareeventsanalysis,etc.

27. Major issues in Data Mining
Miningdifferentkindsofknowledgeindatabases
Interactiveminingofknowledgeatmultiplelevelsof
abstraction
Incorporationofbackgroundknowledge
Dataminingquerylanguages
Expressionandvisualizationofdataminingresults
Handlingnoiseandincompletedata
Patternevaluation:theinterestingnessproblem
Efficiencyandscalabilityofdataminingalgorithms
Parallel,distributed,andincrementalminingmethods

28. Major issues in Data Mining
(cont…)
Handlingrelationalandcomplextypesofdata
Mininginformationfromheterogeneous
databasesand globalinformationsystems
(WWW)
Applicationofdiscoveredknowledge
Domain‐specificdataminingtools
Intelligentquery answering
Processcontrolanddecisionmaking
Integrationofthediscoveredknowledgewith
existing knowledge:Aknowledgefusionproblem
Protectionofdatasecurity,integrity,andprivacy

29. Data Mining Applications(1)
Targetmarketing,customerrelationmanagement,
marketbasketanalysis,crossselling,
Forecasting,customerretention,qualitycontrol,
competitiveanalysis
Textmining(newsgroup,email,documents)andWeb
analysis.
Intelligentqueryanswering
Buyingpatterns
Decisionsupport
Frauddetection

30. Data Mining Applications(2)
ScientificApplications
Networks failuredetection
Controllersdesign
GeographicInformationSystems
Genome‐Bioinformatics
Intelligentrobots