混沌理論在資料探勘上的應用-以國內某藥品公司之客戶信用分析為例

Abstract

企業要能夠穩健的經營，除了追求利益以創造其存在的價值之外，任何有可能影響到企業營運的因素都為企業所不容忽視。本研究是利用一般企業內資訊系統所能夠取得的交易歷史資料進行資料探勘(Data Mining)，以混沌理論為理論基礎研究客戶交易行為與客戶發生跳票問題間的關係，並找出客戶發生信用問題的特徵，以避免增加企業不必要的營運成本與降低經營風險，並藉由健全交易客戶的信用體質，達到加強企業與客戶間合作關係的目的。

經由研究結果顯示，客戶交易行為存在著混沌現象的特性，利用李雅普諾夫指數分析與R/S分析所得到的相關指數變化，歸納出正常客戶交易行為的樣板，做為檢驗客戶異常交易行為的檢驗指標，並初步建立一套客戶信用交易之預警機制，在具備足夠的客戶交易記錄的條件下，可有效的篩選出發生退票行為的客戶。利用預警機制搭配一般企業所採用的信用管理政策，可為客戶信用管理帶來一套新的思考方向。

This study applies the data mining technology and Chaos theory to examine the relation between the behavior of customer transactions and the customer credit. Methodologically, this study focuses on the analyses of the sales data of a pharmaceuticals sales company with the use of Lyapunov exponent and Hurst exponent as indicators to the measurement of chaos. Furthermore, by the observations of the change of these two exponents, this study generalizes the pattern of the behavior of normal customer transaction and use this pattern to separate abnormal clients from normals. Under the analysis of Lyapunov exponent, the result shows that the curve of normal clients rises steadily, while abnormal one falls. Similarly, under the analysis of R/S, the result shows that the curve of normal clients converges toward a stable value, while abnormal one shocks. From these results, this study develops forecasting rules that provide a new way to help making decisions in customer credit management.