基于数据挖掘的电信行业中客户流失模型的分析与实现

《基于数据挖掘的电信行业中客户流失模型的分析与实现》由会员上传分享，免费在线阅读，更多相关内容在教育资源-天天文库。

1、浙江理工大学硕士学位论文基于数据挖掘的电信行业中客户流失模型的研究与实现AbstractInrecentyears，customerchumintelecombecomesserious．ForthethreeoperatorsincludingMobile，ChinaUnicornandTelcomwhoCallassuretheircustomersnottoabandonthedefaultgotoanotherone，atthesametimecangetthelostfromothers，wil

2、lbethefinalwinnerAbittercontesthaslaunchedamongoperatorswhodoalltheycouldtogainmorecustomers．However,operatorspaymuchmoreattentiontocustomerschurnedratherthanthosewhoarereducingtheirconsumption．Hardlyrealizethatthosepeoplearerunningoffgraduallyaspotential

3、lOSScustomers．Telecommunicationsindus仃)，hasamassofdataanddiversitywhichmeansthateverycustomerhasalargenumberofattributeswhichcalledvariablesinthedataminingmodel，suchasARPU，chargeway,downtimesetc．Inordertobettermodel，thispaperdesignswidesheetfromthreeaspec

4、tsofoperatorsendsmessagestoremindcustomer,customerperceptivevalueandvaluebehaviorofcustomerbasedontheassumptionthatthereasonsforcustomerschum．Thendividesthesevariablesintodifferentgroupstodecidewhichvariablestoparticipateinmodelmodeling．Combinedthecustome

5、r’Sowncharacteristicsandthegroupingvariablesthensubdivideallofthecustomers．Customer-chummodelbecomesmoreandmoreintelecommunicationsindustry，inordertoimprovethehitrateofthemodel，thispaperproposesacombinedmodelphilosophy．Thecombinedmodelisbasedontheconstrai

6、ntmodel，predictionmodel，markmodel．Constraintmodelselectsvariableshaslargedistinctiontobeconstraintconditions，predictionmodelscreensvisiblelossandrelativelyobviousvariables，markmodelselectsimplicitcustomerchumvariablestomakeupsmallamountofsamplestoidentify

7、customermorecomprehensively．Eachmodelhasitsownspecialvariables，asaresult，thecombinationmodelproposedinthispaperplaysasignificantroleincustomer-chummodel．ThismodelusesIBMSPSSStatistics，putsforwardtheconceptofpheromonedifferenceinantcolonyalgorithm，usingthe

8、improvedantcolonyalgorithmtosubdividethecustomertoimprovethecustomerclusteringeffect．Canonicaltransformationeliminatetheimpactofthedimensionlesscoefficientstomakethemodelmoreregular．Usinglogisticregressionalgorithmi