塑膠貨幣的興起，改變了人們消費與理財的觀念，根據銀行局統計，民國87年至93年的信用卡循環信用餘額平均每年的成長率為20.85%，現金卡放款餘額民國93年5月至94年2月平均每月的成長率為3.43%。這些數據究竟隱藏了什麼樣的資訊？它要告訴我們什麼樣的訊息？循環利息是信用卡發卡行主要收入來源之一，但相對的其風險也是最高，本研究利用資料探勘的技術，尋找產生這個族群的共同規則，提供發卡銀行卡務推廣參考，例如客戶償債能力、消費能力與對發卡行獲利能力之預測，以及信用額度准駁的依據。自資訊科技發展以來，不但資訊的產生速度以倍速增長，資訊的累積速度更隨著儲存科技的進步以人類未知的度量方式爆炸式的成長，面對如此龐大的資料，如何從其中萃取出可用的知識及還原資料本身所要表達的涵義，儼然成了人類近代科學中另一門熱門且渴望解決的話題。在眾多的分類（Classification）演算法中，規則歸納法(Rule Induction)是找出同質資料之產生規則最常用的方法，但隨著資料屬性的增多以及訓練資料量的龐大，使得整個尋找規則的學習(Learning)時間倍增，本研究將運用分層抽樣的技術，隨機抽取樣本再尋找規則，以進行信用卡族群共同特性之分析。關鍵字：知識探索，資料探勘，分層抽樣，規則歸納，循環利息。 In the advent of plastic cash, eventually we changed the spending and saving habits. According to the statistics of the Department of Finance announced in October 2003, the cyclic credit amount has increased 32.33 percent compared with one year ago. The cumulated interests and annual revenues increased 31.4 percent compared with same period of last year, as well. What is the implication concealed on this report? What should credit card issuing institutions respond to this information? The cyclic credit from the cardholders is the main income of issuing bank; consequently the relative risk it carried is very high. In our research, we are aimed to figure out the commonality of cardholders who apply cyclic credit, based on customer’s ability in paying back the debt, expense pattern, credit allowance and profitability of issuing bank. With the popularity of information technology, we are surrounded by data generated in an unpredictable speed. Facing this gigantic dataset, to extract useful knowledge and its implication becomes a hot research topic waiting us to resolve. Among various classification algorithms, rule induction is the most applied technique in searching the rules out from homogeneous dataset. But with more attributes of entities and large amount of training dataset studied, the learning time in inducing the rules also doubled. So that we tried to stratify the dataset first, followed by inducing the rules, hopefully it would save processing time significantly.
