碩士 / 南台科技大學 / 行銷與流通管理系 / 97 / A customer is a valuable asset, but how to assess the value of a customer has become an important issue recently, especially in the area of customer relationship management. However, the calculation of the customer lifetime value (CLV) becomes complicated with the effort and extra cost to deliver a better model. In the other way, data mining has been used a lot to explore the consumer’s buying pattern and to generate useful models to predict buyer’s behaviors. Therefore, this study tried to demonstrate a procedure to predict a customer’s CLV by combining technologies of data mining and logistics regression.
This study adopted the AdventureWorks database provided by Microsoft SQL Server 2005 as the data source, which contains transaction records of 635 resellers to a fictitious bike company for three consecutive years. The data of the first two years was the training set to evaluate models, and the last year data was used to verify the model obtained in the training set.
Firstly, this study used the training dataset to estimate reseller’s CLV with two basic models: the model of the general value and customer retention model. Then K-means analysis was used to divide resellers into three groups: high, medium, and low CLV resellers. After comparing the group assignments of the training dataset and the verification dataset, this study found customer retention model was the better model to predict resellers’ CLV.
Next, the study conducted the logistic regression on 14 possible explanatory variables to explain the CLV group assignment. The results showed that the stepwise multinomial logit regression was better than the ordered logistic regression in explaining the dependent variables. Furthermore, the number of purchase items, the amount of purchase, and the promotion expense were found to be reliable indicators for the possible CLV groups. This study also found that the medium CLV group was the most potential target for marketing.
Finally, a data mining technology called sequence clustering was used to discover the association among purchase items toward the medium CLV group. Three clusters were indentified: promotion-preferred resellers, non promotion-preferred resellers A, and non promotion-preferred resellers B. Marketing strategies were suggested to these three clusters for better profit returns.
| Identifer | oai:union.ndltd.org:TW/097STUT0691015 |
| Date | January 2009 |
| Creators | We-jo-cheng, 魏若丞 |
| Contributors | 唐楚君 |
| Source Sets | National Digital Library of Theses and Dissertations in Taiwan |
| Language | zh-TW |
| Detected Language | English |
| Type | 學位論文 ; thesis |
Page generated in 0.0017 seconds