| Summary: | 碩士 === 義守大學 === 工業工程與管理學系碩士班 === 93 === As the rise of “Customer Relationship Management”, the enterprises gradually have turned to the customer-oriented model instead of product-oriented before. Therefore, customer relationship management has already become an important activity in enterprises. Because the enterprises know their customers with each passing day, and realize all customers are not same. Some customers will make profit for them, but others won’t.
However, as for the enterprise, the data of demographics is easier to obtain in the market. It may predict the value of customer through the demographics of customers and then make marketing to high valued customers. Therefore, the enterprises could avoid too much and unnecessary marketing cost. Formerly, it is only used to one method to make the prediction mostly, in order to improve the correct rate of prediction, this study combine three methods to predict and compare to the single method.
The study has two models. Model one: use the RFM analysis pattern which is the main variable of customer segmentation to separate customers effectively through the SOM neural network model and K-means method. Then, use the parameter design to improve the quality of segmentation. Model two: formerly mostly only used one method to make the pretest, in order to improve the correct rate of prediction, this study combine three methods to forecast new customer’s value through the voting strategy. It will expect to have better correct rates than the single method.
As follows are the result from the study：
1.The study efficiently divided sample customers into six groups and named them are：high valued customers, grow-customers, low valued customers, new customers and non- valued customers and flow away customers.
2.The multi-classifiers’ exactitude of prediction is definitely higher than the single classifier’s.
3.In this case of the study, it is feasible to use the data of demographics for prediction.
|
|---|
