Abstract

      The credit industry is concerned with many problems of interest to the computation community. This study presents a work involving two interesting credit analysis problems and resolves them by applying two techniques, neural networks (NNs) and genetic algorithms (GAs), within the field of evolutionary computation. The first problem is constructing NN-based credit scoring model, which classifies applicants as accepted (good) or rejected (bad) credits. The second one is better understanding the rejected credits, and trying to reassign them to the preferable accepted class by using the GA-based inverse classification technique. Each of these problems influences on the decisions relating to the credit admission evaluation, which significantly affects risk and profitability of creditors. From the computational results, NNs have emerged as a computational tool that is well-matched to the problem of credit classification. Using the GA-based inverse classification, creditors can suggest the conditional acceptance, and further explain the conditions to rejected applicants. In addition, applicants can evaluate the option of minimum modifications to their attributes.

      With the rapid growth in credit industry, credit scoring models have been extensively used for the credit admission evaluation. The credit scoring models are developed to categorize applicants as either accepted (good) or rejected (bad) credits with respect to their characteristics such as age, income and marital condition. Creditors accept the appli¬cation provided that it is expected to repay the financial obligation, and vice versa. Creditors can construct the classification rules based on the data of the previous accepted and rejected applicants. With sizable loan portfolios, even a slight improvement in credit scoring accuracy can reduce the creditors’ risk and translate considerably into future savings. From the Brill (1998) study, the benefits of credit scoring include cost reduction in credit analysis, faster credit evaluation, closer monitoring of existing accounts and improvement in cash flow and collections. Linear discriminant model (Reichert, Cho, & Wagner, 1983) is one of the first credit scoring models, and it is commonly used today. Linear discriminant analysis (LDA) for credit scoring has been challenged due to the categorical nature of the credit data and the truth that the covariance matrices of the accepted and rejected classes are likely to be unequal (West, 2000). Practitioners and researchers have also applied statistical techniques to develop more sophis¬ticated models for credit scoring, which involve logistic regression analysis (LRA) (Henley, 1995), k nearest neighbor (KNN) (Henley & Hand, 1996) and decision tree (Davis, Edelman, & Gammerman, 1992).

      Classification is a commonly encountered decision making tasks in business. Categorizing an object into a predefined group or class based on a number of observed attributes related to that object is a typical classification problem (Zhang, 2000). In addition to credit scoring and corporate distress prediction, neural networks (NNs) have been successfully applied to a variety of real world classification tasks in industry, business and science. A number of performance comparisons between neural and conventional classifiers have been made by many studies (Curram & Mingers, 1994; Markham & Ragsdale, 1995). Conventional statistical classification procedures such as LDA and LRA are constructed on the Bayesian decision theory. In these classification techniques, an underlying probability model must be assumed in order to calculatethe posterior probability upon which the classification decision is made.

      In credit industry, NN has recently been claimed to be an accurate tool for credit analysis (Desai, Crook, & Over-street, 1996; Malhotra & Malhotra, 2002; West, 2000). Desai et al. (1996) have explored the ability of NN and traditional statistical techniques such as LDA and LRA, in constructing credit scoring models. Their results indicated that NN shows promise if the performance measure is percentage of bad loans accurately classified. However, if the performance measure is percentage of good and bad loans accurately classified, LRA is as good as NN. The percentage of bad loans correctly classified is an important performance measure for credit scoring models since the cost of granting a loan to a defaulter is much larger than that of rejecting a good applicant (Desai et al., 1996).

      In credit industry, NN has recently been claimed to be an accurate tool for credit analysis (Desai, Crook, & Over-street, 1996; Malhotra & Malhotra, 2002; West, 2000). Desai et al. (1996) have explored the ability of NN and traditional statistical techniques such as LDA and LRA, in constructing credit scoring models. Their results indicated that NN shows promise if the performance measure is percentage of bad loans accurately classified. However, if the performance measure is percentage of good and bad loans accurately classified, LRA is as good as NN. The percentage of bad loans correctly classified is an important performance measure for credit scoring models since the cost of granting a loan to a defaulter is much larger than that of rejecting a good applicant (Desai et al., 1996).

      In the field of corporate failure analysis, which is also an important classification problem in business, NNs were also reported to be successful. Coats and Fant (1993) have utilized both LDA and NN to classify firms obtained from COMPUSTAT as either viable or distress. Coats and Fant concluded that NN is more accurate than LDA, remarkably for predicting the distressed companies. Salchenberger, Cinar, and Lash (1992) reported that NN performs as well as or better than the LRA in the prediction of the financial health of savings and loans. From the computational results made by Tam and Kiang (1992), NN is most accurate in bank failure prediction, followed by LDA, LRA, KNN and decision trees.

      In the field of corporate failure analysis, which is also an important classification problem in business, NNs were also reported to be successful. Coats and Fant (1993) have utilized both LDA and NN to classify firms obtained from COMPUSTAT as either viable or distress. Coats and Fant concluded that NN is more accurate than LDA, remarkably for predicting the distressed companies. Salchenberger, Cinar, and Lash (1992) reported that NN performs as well as or better than the LRA in the prediction of the financial health of savings and loans. From the computational results made by Tam and Kiang (1992), NN is most accurate in bank failure prediction, followed by LDA, LRA, KNN and decision trees.

      In order to investigate the possibility of translating a rejected decision into the accepted class for applicants, creditors can suggest modifications to the adjustable attributes with minimum modification cost. This approach lessens to a degree the deficiency of applying NN for credit scoring in explaining the rationale for the decision to rejected applications. Creditors can suggest the conditional acceptance, and further explain the conditions to rejected applications. On the other hand, applicants can evaluate the option of minimum modifications to their attributes. Some of the factors are adjustable, and they may change currently or in the near future.

      This study presents a work to the credit industry that demonstrates the advantages of NNs and GAs to credit analysis. NNs have been emerged as an important and widely accepred technique for classification. Recently, a hyge amount of research works in neural classification has established that NNs are a promising alternative to varioustraditional statistical methods. In this study, the NN-based credit scoring model is used to properly classify the applications as either accepted or rejected, and thereby to minimize the creditors'risk and translate considerably into future savings.

      The GA-based inverse classification technique reassing the rejected instances to the preferable accepted class, which balances between adjustment cost and customer preference. From the computational results of credit dataset, the proposed evolutionary computation based appoach hes shown enough attractive features for the computer-sided credit analysis sysrem.