Font Size: a A A

Research On Customer Churn Early-warning Based On Neural Network

Posted on:2014-01-10Degree:MasterType:Thesis
Country:ChinaCandidate:Z J WangFull Text:PDF
GTID:2249330395494577Subject:Management Science and Engineering
Abstract/Summary:PDF Full Text Request
In nowadays, the market demands are changing continuously and the market isfull of competition, so minimizing the loss of customers is an essential method forenterprises to win the market shares and successes. Since we acceded in WTO,domestic industries have been facing on the increasing challenges from abroad forthe opening of the market, which aggravates the contest on customer resources.Meanwhile, the development of information technology brings electronic commerce.Owing to the network marketing’s advantages, the suppliers are accounted asopponents by enterprises, and this situation is especially severe for the retailers. Forabove reason, customer analysis is necessary and urgent in retailing.The point to decrease the loss of customers is early warning, which always takeadvantages of data mining technology. The work in this article is conducted on theframework of customer relations management, researching the early warning oncustomer losses by means of data mining technology. At first, relevant basic theoryon customer relation management is reviewed in this article, including the definition,reason and management of the loss of customers. Then discussions on customervalues and some kinds of algorithms are demonstrated. Secondly, the earlywarning model is constructed base on RFM customer values and IG-NN attributesselection. In this model RFM is used to calculate customer values, and the mainattributes are selected according to the information gains. The neural net analyses allmain attributes’ contributions on the loss rate of customers and then gives the keyattributes in accordance with the Rule of two eight. Another neural net is trained using customer values and key attributes as inputs and loss rate of the customers asoutput, so we say the constructed model is based on RFM customer values andIG-NN attributes selections. Comparisons between our model with single neural netas well as that only base on IG-NN attributes selections are also done in this article.The results show that our model is more satisfactory than those two on the aspects ofaccuracy, hit rate and speed improvement. The last part is the conclusion and somediscussion on future researches.
Keywords/Search Tags:Customer value, RFM model, Information gain, Retali businessNeural network, Attribute selection
PDF Full Text Request
Related items