Handling Of Missing Values

Question1 Fromthesamplegiven,thefollowingarethedatapreparationsthatneedtobeconducted;handlingof missingvalues,therearesomerowsforboththevariablesinsurancesoldin2019andpreviousclaim whichhasmissingvalues.Thenextdatapreparationisonthevariablepreviousclaiminwhichsome valuesareincorrectlyenteredasNandYinsteadofNoandYes.Thenextdatapreparationinvolve deletingthevariablecustomeridsinceitwillnotcontributetothepredictionoftheinsurancesoldin 2020.Thefinaldatapreparationwillinvolveconvertingthepreviousclaimintobinarysinceweare dealingwitharegressionproblem. Question2 HandlingOfMissingValuesforboththevariablesinsurancesoldin2019andthepreviousclaimFor thevariableinsurancesoldin2019wecanreplacethemissingvalueswitheitherthemeanoranyother measuresofcentraltendencies.However,forthevariablepreviousclaimwithreplaceeitherthemissing valueswiththemodeorjustaconstantvaluesuchasNoorYes. Forthevariablepreviousclaim,thosevaluesinwhichwerenotcorrectlyenteredforexamplethevalue NorYweresupposedtobeenteredasNoorYesrespectively.Inthis,weenterthecorrectvaluesto matchtherestofthevalues. Thevariablecustomeridisjusthighlightedandthendeletedfromtheentiredataset. Thevariablepreviousclaimwillbeconvertedfromnominalvariabletobinaryvariable.ThatistheNo valueswillberecodedaszerowhiletheYesvalueswillberecodedasones. Question3 Theanalystshouldapplythenumerousreductiontechnique.Inthistechniquewereducethevolumeof theoriginaldatasetintoamuchsmallerform.However,wewouldpreferthenon-parametricinwhich wecanapplytheclustering.Inapplyingtheclustering,weobtainthecentroidoftheentiredatasetand thecentroidclosertocertaindatapointsaregroupedtogether.Thisprocesscontinuesuntilallthedata pointsareclusteredintotheirrespectivegroupswithacommonclustercentroid. Question4 Themodelequationisasfollows; Chemicalprice=3516.4270.665packagedweight+0.008density0.946impurities+5.383reactive efficiency Question5 Thefollowingvariablesdonotplayastatisticalsignificantroletotheaboveregressionmodelat95% confidenceleveldensityandpackagedweight.Thisisbecausetheyhavep-valuesgreaterthanthe significancelevelof5%. Question6 Thefollowingvariableshaveahighmulticollinearitywiththeresponsevariablethatisreactiveefficiency andimpurities.Thisisbecausetheyhaveahighstandardcoefficient. Question7 TheR-squaredis0.729inwhichiscloserto1suggestingthatthisisagoodregressionmodelsinceitsfits thegivendatasetquitewell. Question8 Toremovetheattributesthatnotstatisticallysignificantorhavehighmulticollinearityintherapidminer tool,wewillhavetoapplythelinearregressionbutwithforwardselectioncriterion.Thisensuresthat non-significantvariablesaredeletedfromthefinalmodel. Question9 Theassociationrulebadtransport,badcustomerservicerepresentsacorrectandasignificant associationrulesincetherawfrequencyofoccurrenceis55. Question15 Inordertopreventthecustomersfromchurningwehavetoselectthefollowingvariableswithdecision ruleslasttransactiondate,discount,metroregionandweonlyconsidereastandsouth.Furthermore, wecanalsoconsiderthelasttransactiondataandthengreaterthanjuly19,2019toavoidthe customersfromleavingorchurning. Question16 Fromtheconfusionmatrix,wenoticethatthepredictionofnois87.41%whilethepredictionofyesis 68.26%.Thissuggestthatthemodelhasahighsensitivitythanspecificity.Theaccuracyis81.80plusor minus3.85indicatingthatthemodelisgoodquiteinclassifyingthecustomersaseitherloyalorchurn. The3.85valueindicatestheintervalinwhichtheaccuracycanobtainaminimumvalueandamaximum value. Question17 Basedontheaccuracyachievedtheanalystisveryconfidentthatthemodelisacceptedforthe objectiveofthetask.Thisisbecauseitindicatedahighaccuracyvalue. Question18 Theanalystshouldimprovethespecificityofthemodel.Thisisbecausethemodelachievedaless specificityof68.26%.