在現今資訊科技發達的社會中，因同樣類型的商品替代品相當氾濫，企業開始採取不同的行銷手法如：以顧客導向為原則之客製化商品，不只希望吸引新的顧客也希望能保留住舊有的顧客，也就是留住忠實顧客並避免忠實顧客之流失，以維持或增加企業之利益，由此可知辨認忠實顧客及預測顧客行為的改變是非常重要的，但要如何判別忠實顧客以及流失顧客，又該如何訂定行銷策略便成了當今企業所煩惱的課題之一，因此本研究之目的在於找出忠實的顧客及預測忠實顧客流失，以避免企業發生顧客流失及利益損失的情形。
本研究將利用Foodmart所提供之1997年及1998年的交易資料來作為分析的資料來源，首先將資料分別依照RFM的特質做分類，接著輔以Weka軟體中的K-means分群法來作分群，判別出忠實顧客的群組為何，並比較不同年度分群結果找出流失之顧客群。最後基於顧客行為資料用以建構出一預測忠實顧客流失的決策樹模型。
實驗結果顯示，決策樹模型的準確率及回應率比貝氏分類法和類神經分類法的要好，本研究所提出之方法提供了一個有效的機制以預測忠實顧客之流失。
In the well-developed of technology society, there are many kinds of replaced products for customers to select, the enterprises start to use different marketing tactics, for example, customer-oriented of customized products is not only to attract new customers but also hope to remain regular customers, that is to retain the loyal customers and keep them, in order to maintain the enterprises benefit. Therefore, it is important to identify loyal customers and anticipate their behavior changes. But how to differentiate the loyal customers and the lost customers and how to make marketing strategies is the most difficult topic for all enterprises to concern. So, the purpose of this study is to identify the loyal customers and build a model which can predict the customer churn and can avoid losing customers and profits for enterprises.
In this study, the transaction data provided by Foodmart between 1997 and 1998 will be used as the sources of analysis. First, we categorize the data by the characters of RFM. Next, we use the K-means clustering algorithm from Weka to determine which group is loyalty customer and compare the result of different year, in order to find the losing customers group. Finally, a decision tree model is built based on the customer behavior data to predict the churn of loyal customers.
Experimental results show that the decision tree model outperforms the Bayes approach and the neural network method in terms of accuracy and recall. The proposed approach provides an effective mechanism for customer churn prediction.
