Capstone Project chain stores

Capstone Project

OneofthebiggestissuesfacebychainstoresintheretailindustryistheSupplychainmanagement.ThecomponentofthesupplychaininvolvedindetermininghowbesttofulfiltheserequirementsarecreatedfromtheDemandPlan.Theobjectiveofdemandplanistobalancesupplyanddemandinamannerthatachievesthefinancialandserviceobjectivesoftheenterprise.

Ifwelookintothecaseofaretailchainstore,thebasicissueistoknowthedemandofproductsthataresoldinthestore.Ifthedecisionmakingauthorityisawareofthedemandofeachproductonaweeklyormonthlybasis,theywouldbeabletoplanthesupplychainaccordingly.Ifthecreationofademandpredictionmodelispossible,thiswouldsavethecompanyalotofmoneybypreventingover-stockingandbeingabletoplantheirlogisticsandstaffingaccordingly.

Objective

Tocreateamodeltopredicttheweeklysalesforachainofstoresbyanalysingthetrendsofsalesovertimeandtheeconomicfeatures.

Data

TostartthisprojectIamusingadatasetfromKaggle.com.IwouldbedoingmyanalysisandstudyonadatasetthatgivesmetheweeklysalesdataofWalmart.

Thereare3Datasets

WeeklySalesdataof98departmentsin45stores,primarykeyisacombinationof(Store,Dept,Date).TheFeaturesdatasetwitheconomicdetailslikeCPI,Unemploymentrate,Fuelpriceetc..ForeachweekintheregionoftheStoreprimarykeyhereisacombinationof(Store,Date)InthethirddatasetIhaveareaofeachstoreandTypeprimarykeyisStoreMethod

Iambreakingthisprobleminto2stages

1.Createapredictivemodelthatpredictstheweeklysalesof45Stores(InthisstageIamnotplanningtopredictthesalesonDepartmentLevel)2.ExtendtheprojecttoDepartmentLevel(FutureWork)

IdidsomebasiccleaningupofthedataIhadanddidsomebasicexploratoryanalysis.FromtheinsightsIgotfromtheanalysis,Ifurthermovedtocreatingmodelsandfinetuningofmodels.IuseddifferenttypeofRegressionModelsandTimesSeriesAnalysistoproceedandsolvethisstage.

ListofModels1.LassoRegression2.RandomForestRegression3.SupportVectorRegression4.TimeSeriesAnalysisConclusion

Ihavecomeupwithdifferentmodelsthathavehelpedmetopredicttheweeklysalesof45storeswitharound96.5%ofaccuracyfor3months(90Days).OutofthemodelsIran,IselectedtheLassoRegressionasmypredictiveModelsinceitgavemethehighaccuracyofpredictionwiththelowestresidual(error).

EventhoughIhaveselectedonemodelasofnowIamtryingtopredictsalesfor45stores.Iamsuretheimpactofeachfeaturemightbedifferentfordifferentstoresbecauseofmanyreasonsandthat聮swhyIcheckedhowdifferentmypredictionwitheachstoreis.

Herethebiggerthebubbleisthehigherthevariationinprediction.Sowecanseethatthereare3storeswhichhavemorethan5%variationinprediction.InthefuturestudyIcanlookintothesestoreandtrytoseewhatgoeswrongandusinganensemblemethodofgettingpredictionfromdifferentmodelstrytoimprovethepredictions

CheckthisinGitHub

clickheretocheckmycapstone