Defaults Assessment of Mortgage Loan with Rough Set and SVM

Credit risk is the primary source of risk to financial institutions. Support vector machine (SVM) is a good classifier to solve binary classification problem. The learning results of SVM possess stronger robustness. We adjust these penalty parameters to achieve better generalization performances with using grid-search method in our application. In this paper the attribute reduction of rough set has been applied as preprocessor so that we can delete redundant attributes, then default prediction model of the housing mortgage loan is established by using SVM. Classification performance is better than some other classification algorithms.

1. Introduction

Credit risk is the primary risk facing financial institutions. Arguably a cornerstone of credit risk modeling is the probability of default (PD) [1], [2]. The housing mortgage is an important component of bank loan. However, there are huge credit risks that banks take back the principals and interest of loan with default of the consumers. There is a credit scoring system for consumer mortgage loan application to produce an internal rating. It is a traditional approach based an empirical model, which takes into account various quantitative as well as subjective factors, such as the consumers' age, household income, interest rate, etc. Through this scoring system, analyzing all the pieces of information in your credit record and summarizing them in a number calculate a credit score of a consumer. A company named Fair, Isaac & Co. (FICO) developed a mathematical way to look at factors in your credit record that may affect your ability and willingness to repay a debt [3]. The problem with this approach is of course the subjective aspect of the prediction, which makes it difficult to make consistent estimates. Classification is used to predict unknown label by the classifier which is trained with experiential data. The credit scoring problems can transform to classification than predict the probabilities of defaults. Recent researches have shown that artificial intelligence methods [4], [5] achieved better performance than traditional statistical methods.