基于消费行为的客户细分模型应用分析

《基于消费行为的客户细分模型应用分析》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、对所识别出来的消费群体进行有针对性的营销。最后，我们尝试将此模型应用于信用卡欺诈行为的探测。实验表明对客户细分后再使用BP模型进行检测的正确检出率略有提高，错误检出率降低，这也从一个侧面说明我们的细分方法是有效的。关键词：客户细分模型，聚类分析，市场营销，欺诈探测ⅡResearchonApplicationofCustomerSegmentationModelBasedonConsumptionBehaviorMajor：ComputerSoftwareandTheoryGraduateStudent：ZhuYan-liTutor：YuanDingLiuFangThetrem

2、endouscompetitionourfinancialindustryis伍cingisforcingbankstoadoptaclient-centeredmanagementmodelafteroureountry*sjoiningtoWTO．Inaclient-centeredmanagementstyle,themostimportantandurgentistoclassifythecustomersproperly．GroupingcustomerintocategoriesgivesUSbetterunderstandingofcustomersandpro

3、videspersonalizedservice．Therehasbeensomeresearchabroadonconsumerbehaviorappliedinthefieldofconsumersegmentation,likeRFMmodel,CustomerValueMatrix(CⅥ哪model，etc．Mostofthesegmentationcriterionsincurrentpracticeareproducedbycustomergeographicdistributionandtheircharacteristics，butseldomCallwefi

4、ndthosecriterionsbasedoncustomerbehavior．Thesesegmentationmodelareclassifiedonlybyonesideofthecustomer'sbehavior．Inthispaper,wepresentanewclusteringapproachtogroupcustomersanddefineconsumptionexpenditureandconsumptionfluctuationrateassegmentationvariabletodeterminethesubdividedvariablestoov

5、ercometheshortcomingsoftheexistingsubdivision．Thispaperaimsatestablishingacustomersegmentationmodelbasedonconsumptionbehavior．Themodelcanguidethebankstodetecttargetedmarketingandfraudrisk．Firstly,weuseimprovedK-MEANSalgorithmtoconstructtheⅢmodel，fromwhichwecannotonlygetcharacteristicsofcons

6、umerbehaviorpatterns，butalsofindinterestedcustomergroupsthroughidentifyingcharacteristicpauems．Secondly,weuseDBSCANalgorithmtoconstructthemodel．Throughcomparingthetworesults，wefindthatDBSCANexcludesthefactthatanisolatedpointhasaninfluenceofclusteringintheprocessofclustering,givingthecustome

7、rtypesclosertoreality．Finally,wetrytoapplythismodeltodetectcreditcardfraud．Experimentsshow,thecorrectdetectionrateusedbyBPalgorithmaftercustomersegmentationhasincreasedslightly，whileerrordetectionratehasdecreased．Thisprovesthatthesegmentationmethodwhichw