Escolar Documentos
Profissional Documentos
Cultura Documentos
PredictiveanalyticsusingSAPInfiniteInsight|agileDSS
PredictiveanalyticsusingSAPInfiniteInsight
agiledss.com/en/blog/predictiveanalyticsusingsapinfiniteinsight
Classification,RegressionandHowTheyCanSaveYouTime&Money
WhomshouldItargetwithmynextcampaign?Whatarethecharacteristicsofapersonthatwillchurn
withinthenextweeks?Whatisthebestlocationformyupcomingevent?Businessesusuallyask
questionsaboutthefuture,butthedatatheycanbasethemselvesonwillalwaysbehistory.
Now,predictingthefutureadmittedlysoundsabitlikemagic.However,manyquestionsaboutthe
futurecanbeansweredusingdatafromthepastandthehypothesisthatthesameunderlying
mechanicsthatformedpastdatawillbeatplayinthefuture.Regardlessofthetoolused,the
foundationisnotmagic,butstatistics.
Thisblogserieswillexplainsomeofthemostimportantstatisticalconcepts,theirfieldsofapplication
andanexampleofhowtheyhelpsavetimeandmoney.Readonforanappliedexampleofsegmenting
clientsbasedonsociodemographicdataforamarketingcampaignusingSAPInfiniteInsight.
ALittleBitofTheory
Classificationisamethodologyusedtoassignanobservationtooneofaknown,userdefinedsetof
classes.Astatisticalmodelisamathematicalformulathatexpressesatarget(output,dependent,
explained)variablebymeansofsomeexplanatory(input,independent,predictor)variables.Target
variableistheclassthatanobservationisassociatedwith,whileanyotherattributeoftheobservation
canserveasexplanatoryvariables.
Theclassesthatwereassociatingourdatawitharechosen,notderivedfromthedata.Thereliability
andusefulnessofsuchassignmentsheavilydependsontheactual,causalrelationshipsbetweenthe
variables,whetherknownorunknown.Ifthetargetvariableisdiscreet(orcategorical),i.e.onethatcan
takeonalimitedandusuallyfixednumberofvalues,wespeakofclassification.Ifthetargetvariableis
continuous(suchasatemperature)wespeakofregression.Targetclassesareoftendichotomous,
e.g.good/bad,dirty/clean,buyer/nonbuyer.Thetermclassificationissometimesreplacedbythe
wordprediction,especiallyinnonscientificliterature.
Heresanexample:Wewillberathersafetoclassifypeopleintorichandpoorbasedontheiraverage
incomeoverthepast5yearsthereisadirect,causalrelationshipbetweenthetwoandwecan
thereforepredictthecharacteristicrichnessusingtheexplanatoryvariableaverageincomewithsome
confidence.However,wecanhavelessconfidencewhenpredictingrichnessbasedontheageofa
person.Ofcourse,thereissomecorrelationbetweenthetwo,butnostrictcausalityoreven
dependency,letaloneotherinfluencingfactors.
FieldsofApplication
Whileitseemsthatourstatisticalweaponsarequiteweak,theirpowerliesintheirapplication.Aswe
arefreetodefinetheclassesthatwewantourobservationstoberelatedto,wecanapplythesame
http://www.agiledss.com/en/blog/predictiveanalyticsusingsapinfiniteinsight
1/7
9/23/2016
PredictiveanalyticsusingSAPInfiniteInsight|agileDSS
techniquetomanydifferentfieldsofapplication.Herearesomeexamples:
ThepropensityofaclienttobuyacertainproductbasedonCRMdata
Thelikelihoodofaclienttochurnbasedonhisusagestatistics
Theoptimalpricingforaproductbasedonlocationandsociodemographics
Theprobabilityofaclienttodefaultonhiscreditpaymentsbasedonhispaymenthistory
Thelifetimeofamachineundercertainenvironmentvariables
Thenumberofdefectsuppliesbasedonlogisticsdata
Thesuccessofaneventbasedonitslocation
Thefitofanemployeebasedonhispastemployments
Thefailureofadeliverysystembasedonmeteorologicaldata.
Keepinmindthatthebestmodelcannotpredictanythingusefulifthereareno(useful)correlationsin
thedataortoomuchnoise.
Practice:UsingSAPInfiniteInsight
AstheworkflowinInfiniteInsightusuallystartsbyselectingatechniqueorafamilyofalgorithmsrather
thanabusinessproblem,thetoolrequirestheusertohavesomeknowledgeaboutthesuitabilityofan
approachforagivenproblem.Likewise,itisadvantageoustounderstandthewaysapplicable
algorithmswork,thetypeoftargetvariablesonecanobtainandthewaystoinfluencethecreationof
themodel.However,withautomationinmind,thetoolusesaproprietaryalgorithmwhichremainsa
blackboxtoalargedegree.
Whatmightlooklikeadisadvantageatfirstsightisinpracticeamodestlossofflexibilityinexchangefor
ahugegaininproductivity.Stepssuchascuttingthetrainingdataintoestimation,validationandtesting
sets,selectionandbinning(regrouping)ofvariablesandparameterizationarelargelyautomated.The
usercan,butdoesnothaveto,interveneasaresultitbecomespossibletousealargenumberof
potentialexplanatoryvariableswithoutriskandletthetooldecidewhichonestouse,whichtranslates
intomoreaccuratemodelsatnoextracost.Itisalsopossibletocreatealargenumberofmodelsin
parallel,i.e.scoringindividualsforcampaignsforhundredsofproductsandselectingonlythosewiththe
highestlevelofconfidence,translatingintounprecedentedefficiency.
Classification/RegressionUsingSAPInfiniteInsight
Heresthescenario:Inordertomaximisetheimpactofourmarketingcampaign,wewanttomaximise
thenumberofpotentialbuyersamongthetargetedindividuals.Therefore,wewanttoassignascoreto
eachindividualthatdescribeshiswillingnesstobuyandthentakethetopindividualsthatfitintoour
campaignbudget.
Intheopeningscreen,weselectCreateaRegression/ClassificationModelandsubsequentlyselecta
datasource.
http://www.agiledss.com/en/blog/predictiveanalyticsusingsapinfiniteinsight
2/7
9/23/2016
PredictiveanalyticsusingSAPInfiniteInsight|agileDSS
ClickonAnalyzeinstep3togetapreviewofthecolumnscontained.Step4letsusselecttargetand
explanatoryvariables:Inourcaseweselecttwovariablesforpredictionandeliminateeverythingthat
wecantapplytootherdatasetsfromtheexplanatoryvariables.
http://www.agiledss.com/en/blog/predictiveanalyticsusingsapinfiniteinsight
3/7
9/23/2016
PredictiveanalyticsusingSAPInfiniteInsight|agileDSS
ClickNextandGenerateinthesubsequentscreenstocreatethemodel.Asummaryisbeing
displayed.ClickNexttoaccessthefollowingscreen,allowingyoutoinspect,applyanddeploythe
model.
http://www.agiledss.com/en/blog/predictiveanalyticsusingsapinfiniteinsight
4/7
9/23/2016
PredictiveanalyticsusingSAPInfiniteInsight|agileDSS
Usetheconfusionmatrixtoassessthepredictivepowerofthemodel.Wecanexpectroughly50%of
allrecipientstobuyourproductwhentargetingjust10%ofthepopulation.Whilethisisnotan
extremelyhighvalue,itrepresentsahugeincreaseinprecisionasthereareonlyabout19%ofbuyers
inthewholepopulation.ClickonContributionsbyVariableifyouwanttoinspectwhichvariablesare
themostpowerfulpredictors.
http://www.agiledss.com/en/blog/predictiveanalyticsusingsapinfiniteinsight
5/7
9/23/2016
PredictiveanalyticsusingSAPInfiniteInsight|agileDSS
Ifyouwanttodeployyourmodelanduseitinanexistingapplicationsuchasadashboardorreport,
simplyexportthemodeltoSQLusingGenerateSourceCodeandintegrateitintoyourDWH.
Clickhereifyouwanttoseethisandotherfieldsofapplicationinarecordedlivedemosession!
http://www.agiledss.com/en/blog/predictiveanalyticsusingsapinfiniteinsight
6/7
9/23/2016
PredictiveanalyticsusingSAPInfiniteInsight|agileDSS
http://www.agiledss.com/en/blog/predictiveanalyticsusingsapinfiniteinsight
7/7