應用資料探勘技術建構顧客流失預測模型

Applying Data Mining Techniques to Construct Customer Churn Prediction Model

隨著 ICT 技術的進步，利用資料探勘技術從大量數據中挖掘所隱藏的知識對企業越趨重要，尤其在顧客關係管理中的流失顧客預測。雖然決策樹的易讀性以及羅吉斯迴歸捕捉變數間函數關係的優勢使得這兩種方法在流失預測的文獻中常被使用，然而結合此兩種方法優點的羅吉斯葉在流失預測中的適用性並沒有太多的討論。本研究鑒於決策樹的預測結果可視為是一種監督式的分群結果，再輔以羅吉斯迴歸來捕捉每一群中客戶流失的原因，非常契合精準行銷的先分群再建模的概念，因此主張以羅吉斯葉來建構流失預測模型。實證結果顯示羅吉斯葉確實有較好的檢定力與較低的錯誤歸類成本，而決策樹規則與羅吉斯迴歸的顯著變數也都可以提供管理者重要的管理意涵。

 

With the advancement of ICT technology, how to use Data mining techniques to discover the potential knowledge from big data is becoming more and more important for enterprises, especially for the churn prediction in the field of customer relationship management (CRM). Although the easy to read rules of Decision Trees and the advantages of Logistic regression in capturing the functional relationships among variables had made these two algorithms widely used in the literature on churn prediction, the applicability of Logit Leaf Model combining the benefit of these two algorithms had not yet been discussed too much. In view of the fact that the prediction result of the Decision Tree at the first stage can be regarded as a supervised clustering result, and it can further be supplemented by Logistic regression to find the causes of customer churn in each cluster at the second stage, which fits the concept of precision marketing, this study advocates using Logit Leaf Model to construct customer churn prediction models. The empirical results showed that Logit Leaf Model had the higher power and the lower misclassification costs among the algorithms, and the rules of the Decision Tree and the significant variables of Logistic regression can also provide decision makers important managerial implications.