Font Size: a A A

Study On The Market Segmentation Of Credit Card Based On Consumer Data

Posted on:2015-04-14Degree:DoctorType:Dissertation
Country:ChinaCandidate:Y S DuFull Text:PDF
GTID:1109330422493421Subject:Management Science and Engineering
Abstract/Summary:PDF Full Text Request
It is highly necessary for commercial banks to improve management level andoperation capacity of credit card by market segmentation, yet current research and practicehave some serious problems and can’t meet the rapid business development of credit card.Literature survey found the following issues. Seen from the number of researches, theamount of segmentation researches on cardholders are more, while those on merchants arequite fewer. Seen from the research perspective, most segmentation of cardholders paymore attention on customer value neglecting their credit risks in the same models, while thesegmentation of merchants mainly depend on industries and formulate different merchantdiscount rates, however such policy has no practical value for supervision management.Seen from the marketing need of banks, nearly all segmentation researches didn’t considerthe association between cardholders and merchants, thus the marketing strategies often losstarget and can’t achieve foremost purpose. Facing these problems, the study proposed threemarket segmentation models for credit card management which is based on transaction dataand implemented under data mining technology. We hope these models can provideeffective methods for commercial banks to improve their segmentation management oncredit card markets.To solve the problem of current segmentation which use customer value and takesfewer credit risk into account simultaneously, we proposed RFMDA model which is basedon RFM model for market segmentation of credit cardholders. In addition to the definedthree features in RFM model, i.e. Recency, Frequency and Monetary Value, the RFMDAmodel adds two new features, i.e. Default Risk and Arbitrage Risk and integrates credit riskfactors into customer evaluation. With the five features, we use K-means++algorithm tocluster all cardholders into9groups, and then we label each group according to value (high,medium, low) and risk (high, medium, low) by comparing their feature values of centroids.Comparative experimental results show RFMDA model outperforms RFM model inpractical value and effectiveness.To deal with the contradiction between rapid growing merchants and short of staff aswell as lack of supervision, we proposed the MSBRF model based on random forestalgorithm for merchant segmentation. The model presents10decision variables out oftransaction data for identifying risk characteristic and segmenting merchants. With the10 features, we build a classifier by learning partial samples of merchants, and then use it toclassify other merchants into two separated groups, one of which represents possible riskmerchants. By means of selecting risk merchants, banks can spend limited resources intoeffective supervision. Experimental results show the effectiveness of MSBRF model insegmenting merchants. Comparative experiments demonstrate the random forest algorithmperform better than the decicion tree algorithm. Compared with other algorithms, therandom forest algorithm used in MSBRF model can evaluate the importance of featuresafter classification, which can provide valuable reference for practical merchantmanagement.To cope with association between cardholders and merchants in segmentation, so thatbanks can formulate targeted marketing policies, we built3types of network and proposedsegmentation mothods of cardholders and merchants based on technique of social networkanalysis. The first network, namely transaction association network (TAN), is built basedon the transaction behavior between cardholders and merchants. Then the island analysistechnique is used to discover closely related cardholder and merchant groups in TAN forthe management of credit card supervision. The second network, namely cardholderassociation network (CAN), is built based on the relation of same merchants betweencardholders, while the third network, namely merchant association network (MAN), is builton the relation of same cardholders between merchants. We use clique discovery andcommunity detection technique to discover closely related cardholder groups in CAN ormerchant groups in MAN. These close connected groups provide important basis forprecision marketing of credit card. Experimental results show the effectiveness of proposedmethods.
Keywords/Search Tags:credit card, market segmentation, cardholder, merchant, social networkanalysis, data mining, customer value evaluation, credit risk
PDF Full Text Request
Related items