銀行在金融體系上，扮演著資金供給的角色，對企業貸款的利息所得是銀行主要的獲 利來源之一。若借款企業倒閉或無法償還所借的債務，則會造成銀行的不良放款，導致盈 利降低，因此對企業信用風險的評估，也就是對企業的償債能力的分析，是銀行放款時的 重要工作及依據，好的企業信用評等品質，可以減少銀行的逾期放款金額，降低逾放比率， 保障銀行的獲利。 為了避免金融機構因同業競爭而放寬信用評等，造成各金融機構的呆帳普遍增加，國 際清算銀行對於信用風險規範了一套協定「巴賽爾資本適足條約」，採用更嚴謹的評審標 準，各金融機構也以憑此為依據，發展出自己的一套企業信用評等評估標準。 傳統上，銀行審核授信，對於借款企業的信用，大多依靠主觀的經驗，或是5P 原則， 或是6C 原則。近年來，資訊科技的發達，資料挖掘的技術也越來越廣泛應用在信用風險 上，發展出許多信用評等模型，如支撐向量機模型、Logistic 模型、類神經網路模型、模 糊徑向基網路模型等，不同的模型皆有不同的優缺點。 本研究運用基因演算方法，以DNA 編碼與繁殖的原理，遴選適合環境的「下一代」， 且不斷的演進，並且與企業評等機構現有的個案數據及違約機率做比較，計算出信用評等 模型的各項財務指數的權重，跟四個信用評等模型做比較、分析。借款企業的各項財務指 數透過此系統，能夠計算出一個值，用來預測此借款企業的客觀違約機率，以提供銀行做 貸款決策。

Commercial banks play an important role in the supply of capital in the financial market. Interest income of enterprise loans is one of the major sources of their profits. An enterprise or a debtor cannot repay debts due to the business failures will result in the increasing of banks’ non-performing loans and lower their profits. Therefore, the assessment of enterprise credit risks, by analyzing a solvency of enterprises, becomes an important task. Better quality of enterprise credit rating can protect bank profits by reducing non-performing loan ratio and the total amount of overdue loans of banks. As business competitions among various financial institutions, they loosen the loan credit rating criteria in spite of the increasing of non-performing loans. In order to prevent the situation, Bank for International Settlements adopts more stringent assessment criteria to standardize a set of agreements “Basel Capital Accord” for the credit risk. Every financial institution therefore develops its own enterprise credit rating assessment criteria based on “Basel Capital Accord.” Traditionally, a bank assesses a debtor’s loan credit rating according to subjective experience such as 5P principles or 6C principles. In recent years, as the advance in information technology, data mining technology is also more widely used in credit risk assessment. Many credit rating models are developed such as support vector machine model, Logistic model, neural network model, fuzzy radial basis function network model, etc. Different models have their own pros and cons. Our model applies the genetic algorithm to enhance the evaluation of loan credit rating of enterprises. Through the simulation of DNA encoding and propagation principles, better financial index assessment parameter weights are generated. These weights are then used to compare with existing enterprise cases and analyze the ability of their debt payments. Values of these weights are improved by running the process over and over until no better values are generated. These weights are employed to produce a value which is able to predict probabilities of bad debts of enterprises. Banks can then make better decisions depending on the output value of our model.